Nature Neuroscience March 2000.pdf

© 2000 Nature America Inc. • http://neurosci.nature.com

contents volume 3 no 3

march 2000


http://neurosci.nature.com

Mice lacking NMDA receptor function in the CA1 region of the hippocampus have impaired nonspatial memory. Exposure to an enriched environment, however, ameliorates these learning deficits and increases spine density in CA1. See pages 205 and 238.

editorial Mysterianism lite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

news and views Predicting perception from population codes . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 Jennifer M. Groh ➤ SEE ARTICLE, PAGE 270

ChIPping away at potassium channel regulation . . . . . . . . . . . . . . . . . . . . . . . . . 202 Min Li and John P. Adelman Toying with memory in the hippocampus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Howard Eichenbaum and Kristen Harris ➤ SEE ARTICLE, PAGE 238 Place cell firing during space flight. Page 209.

Attention - brains at work! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 Roger B.H. Tootell and Nouchine Hadjikhani ➤ SEE ARTICLES, PAGES 284 AND 292

Signaling dendritic growth in vivo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 Sandra Aamodt ➤ SEE ARTICLE, PAGE 217

brief communications Three-dimensional spatial selectivity of hippocampal neurons during space flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 JJ Knierim, BL McNaughton and GR Poe

Predicting perception from population coding. Pages 201 and 270.

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature America Inc., headquartered at 345 Park Avenue South, New York, NY 10010-1707. Editorial Office: 345 Park Avenue South, New York, NY 10010. Telephone 212 726 9200, Fax (212) 696 9635. North American Advertising: Nature Neuroscience, 345 Park Avenue South, New York, NY 100101707. Telephone (212) 726-9200. Fax (212) 696-9006. European Advertising: Nature Neuroscience, Porters South, Crinan Street, London N1 9SQ. Telephone (0171) 833 4000. Fax (0171) 843 4596. New subscriptions, renewals, changes of address, back issues, and all customer service questions in North America should be addressed to Nature Neuroscience Subscription Department, PO Box 5054, Brentwood, TN 37024-5054. Telephone (800) 524-0384, Direct Dial (615) 377 3322, Fax (615) 377 0525. Outside North America: Nature Neuroscience, Macmillan Magazines Ltd., Houndsmill, Brunel Road, Basingstoke, RG21 6XS, U.K.. Tel: +44-(0)1256-329242. Fax: +44-(0)1256 812358. Email: [email protected]. Annual subscription rates: U.S./Canada: U.S. $595, Canada add 7% for GST (institutional/corporate), U.S. $195, Canada add 7% for GST (individual making personal payment BN: 14091 1595 RT); U.K./Europe:£435 (institutional/corporate), £185 (individual making personal payment), £99 (student); Rest of world (excluding Japan): £480 (institutional/corporate), £195 (individual making personal payment), £110 (student); Japan: Contact Japan Publications Trading Co. Ltd., 2-1 Sarugaku-cho 1 chome, Chiyoda-ku, Tokyo 101, Japan, phone (03) 292-3755. Back issues: U.S./Canada, $45, Canada add 7% for GST; Rest of world: surface U.S. $43, air mail U.S. $45. Reprints: Nature Neuroscience Reprints Department, 345 Park Avenue South, New York, NY 10010-1707. Subscription information is available at the Nature Neuroscience homepage at http://neurosci.nature.com. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Application to mail periodicals postage rate is paid at New York, NY. Executive Officers of Nature America Inc: Nicholas Byam Shaw, Chairman of the Board; Jan Velterop, President; Edward Valis, Secretary-Treasurer. Printed by Publishers Press, Shepherdsville, KY, USA. Copyright ©2000 Nature America Inc.

nature neuroscience • volume 3 no 3 • march 2000

i


contents

articles NMDA receptor-mediated control of protein synthesis at developing synapses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 AJ Scheetz, AC Nairn and M Constantine-Paton Rho GTPases regulate distinct aspects of dendritic arbor growth in Xenopus central neurons in vivo. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 Z Li, LV Aelst and HT Cline ➤ SEE NEWS AND VIEWS, PAGE 208


Mapping shifts in visual spatial attention. Pages 206, 284 and 292.

Anatomical and physiological evidence for D1 and D2 dopamine receptor colocalization in neostriatal neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 O Aizman, H Brismar, P Uhlén, E Zettergren, AI Levey, H Forssberg, P Greengard and A Aperia Growth cone and dendrite dynamics in zebrafish embryos: early events in synaptogenesis imaged in vivo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 JD Jontes, J Buchanan and SJ Smith Enrichment induces structural changes and recovery from nonspatial memory deficits in CA1 NMDAR1-knockout mice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 C Rampon, YP Tang, J Goodhouse, E Shimizu, M Kyin and JZ Tsien ➤ SEE NEWS AND VIEWS, PAGE 205

Muscles express motor patterns of non-innervating neural networks by filtering broad-band input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 LG Morris, JB Thuma and SL Hooper D1 and D2 receptors in the striatum. Page 226.

Microsaccadic eye movements and firing of single cells in the striate cortex of macaque monkeys. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 SM Conde, SL Macknik and DH Hubel Lack of cortical contrast gain control in human photosensitive epilepsy . . . . . . . 259 V Porciatti, P Bonanni, A Fiorentini and R Guerrini Learning to find a shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264 M Sigman and CD Gilbert Seeing multiple directions of motion—physiology and psychophysics . . . . . . . . 270 S Treue, K Hol and HJ Rauber ➤ SEE NEWS AND VIEWS, PAGE 201

Imaging synapse formation in vivo. Page 231.

A multimodal cortical network for the detection of changes in the sensory environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 J Downar, AP Crawley, DJ Mikulis and KD Davis The neural mechanisms of top-down attentional control . . . . . . . . . . . . . . . . . . 284 JB Hopfinger, MH Buonocore and GR Mangun ➤ SEE NEWS AND VIEWS, PAGE 206

Voluntary orienting is dissociated from target detection in human posterior parietal cortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292 M Corbetta, JM Kincade, JM Ollinger, MP McAvoy and GL Shulman ➤ SEE NEWS AND VIEWS, PAGE 206

classified advertising see back pages


ii


editorial


Mysterianism lite A philosophical view known as ‘mysterianism’ holds that even though there is nothing supernatural about how consciousness arises from neural activity, the human brain is simply not equipped to understand it. The reason we find the mind–brain problem so baffling, the argument goes, is that humans did not evolve sufficient cognitive abilities to solve it, just as armadillos did not evolve the ability to understand arithmetic. This argument has been advocated by philosophers such as Colin McGinn and cognitive scientists such as Steven Pinker. Now it has been taken up by a prominent science journalist, John Horgan, whose new book The Undiscovered Mind offers a view of brain science that might best be described as ‘mysterianism lite’. It is not just consciousness that is beyond our grasp, he says; neuroscience as a whole is failing, because the brain is too complicated for human understanding. Horgan attracted attention, even notoriety, for his 1996 book The End of Science, in which he argued that the age of great scientific discoveries is coming to an end because most of the big questions have been answered. The brain is an obvious exception, but Horgan now argues that neuroscience too is reaching its limits, not because it has succeeded in its aims but because those aims are unachievable. The subtitle of his new book is “How the human brain defies replication, medication, and explanation”; its thesis is that the achievements of neuroscience (along with psychology, psychiatry and other related areas) are being oversold, that the supposed practical benefits have been exaggerated, and that the field is now confronting an ‘explanatory gap’ that may never be bridged. Unlike particle physicists or molecular biologists, says Horgan, neuroscientists “…have yet to achieve their reductionist epiphany. Instead of finding a great unifying insight, they just keep uncovering more and more complexity. Neuroscience’s progress is really a kind of anti-progress. As researchers learn more about the brain, it becomes increasingly difficult to imagine how all the disparate data can be organized into a cohesive, coherent whole.” It is tempting to dismiss this as another example of what Richard Dawkins once called “the argument from personal incredulity”, but Horgan is surely not alone in finding neuroscience difficult to approach. The brain is immensely complicated, and in the absence of a grand unifying theory for how it works, researchers tend to study very diverse problems that often seem unconnected to each other. It is therefore understandable that their achievements do not always seem intellectually satisfying to nonspecialists. Part of Horgan’s critique may reflect how neuroscience is reported in the media. Among the stories that attract the most attention are the identification of genes or brain areas that are associated with particular behaviors (think of fosB, ‘the gene for maternal behavior’, or the orbitofrontal cortex, ‘the brain’s moral compass’), but typically such findings are only the first steps on a long road toward mechanistic understanding. In contrast, many of the most important mechanistic insights into how the brain works (at all levels, from nature neuroscience • volume 3 no 3 • march 2000

biochemistry to computation) tend to go unreported, because they are very difficult to explain to lay people. The problem goes deeper than this, though. Even where mechanistic explanations of brain function have been possible, they do not ‘feel’ like explanations of mental processes. Consider the paper by Treue et al. on page 270 of this issue, which presents a striking example of how far our understanding of perception has progressed. Based on knowledge of how motion is represented by populations of neurons in the visual cortex, the authors were able to predict an entirely unexpected visual illusion; two different patterns of moving dots are perceptually indistinguishable, apparently because they both evoke the same pattern of activity in a cortical area called MT. Of course a graph showing the distribution of neuronal firing rates in MT doesn’t ‘feel’ like an explanation of perception. But why should it? The criterion for a good theory is not that it feels right, but that it can successfully predict unexpected results. If a physical theory of neural processing can predict an unexpected mental phenomenon, that is surely a substantial achievement. It goes without saying that Treue’s study raises many further issues—how is the population activity decoded, what other areas are involved in representing the stimuli, and so forth—but there is no reason why questions of this type should not eventually be answered. Certainly, it will be a challenge to understand how (say) a moving red bar is perceived as a unitary stimulus if its orientation, motion and color are each represented in different cortical areas. Horgan may well be right that existing hypotheses to solve this so-called binding problem (such as synchronous oscillations) will prove incorrect. But to deny the possibility of further progress seems perverse. A deeper understanding of the mechanisms underlying mental processes should follow from greater knowledge of anatomical and functional connectivity, better methods of recording and manipulating neural activity, and more realistic computational models, all of which should be achievable with enough time and effort. For ethical reasons, we may never know as much as we would like about human brain activity, but it seems reasonable to expect that insights from other organisms will some day provide good models for much of what happens inside our own heads. The mysterians might turn out to be correct in claiming that we will never fully understand how brain activity leads to subjective experience: why the firing of MT neurons feels like visual motion, why dopamine release in the nucleus accumbens feels pleasurable, or why electrical stimulation of the anterior supplementary motor area feels amusing. But that is very different from claiming that these phenomena can never be explained in physical terms, or that neuroscience is, as Horgan puts it, “bumping up against fundamental limits of science.” Most neuroscientists, fortunately, take a more optimistic view than Horgan; the explanatory gap may not be closed at a single stroke, but it is getting narrower by the day. 199


news and views

Predicting perception from population codes Jennifer M. Groh


Treue and colleagues use electrophysiological recordings in monkeys and psychophysical experiments in humans to suggest that the shape of a population response in a motion sensitive region of the brain (area MT), rather than the peak of the response, determines motion perception. When neuroscientists can consistently predict what people perceive by studying their neural activity, we will have achieved a remarkable level of understanding of brain function. A notable advance toward this goal is presented by Treue, Hol and Rauber 1 in this issue of Nature Neuroscience. These authors have used the response profiles of neurons in a motionsensitive area of the monkey brain, area MT, to predict how humans will perceive moving stimuli. The brain encodes many kinds of sensory stimuli using maps of neurons that are tuned to the properties of those stimuli. How does the neural activity in these maps subserve perception and sensoryguided action? Because neurons are broadly tuned, a single stimulus typically activates a large population of neurons— the so-called population response. Several different theories have been proposed for how population responses in turn mediate perception and action. The most obvious possibilities are that perceptual outcome is determined either by the peak of the population response or by its center of gravity (also known as the vector average of the response). When only one stimulus is present, the peak and the center of gravity of the population response are the same. But what happens when two stimuli with different features occur at the same place and time? Both stimuli influence the population response, but are they perceived as independent? Do they both contribute to behavioral responses? How do the two stimuli interact? If the peak of the population response is the most important feature, then both stimuli would be perceived so long as the two stimuli are sufficiently different from one another that the population response contains a separate peak Jennifer Groh is at the Department of Psychological and Brain Sciences, Center for Cognitive Neuroscience, Dartmouth College, Hanover, New Hampshire 03755, USA. e-mail: [email protected]

for each. In contrast, if the center of gravity is important, then the location and number of peaks should not matter. Under the latter mechanism, subjects would perceive a single stimulus intermediate between the two actual stimuli, regardless of whether the population response has two separate peaks of activity. Visual motion processing is one domain where these issues have been explored fairly extensively. Motion perception is thought to rely on the population responses in visual area MT, which is specialized for processing moving stimuli and contains a columnar organization for motion direction (for review, see ref. 2). Because of this topographical organization, microstimulation can be used to activate a population of neurons with similar motion preferences, thereby simulating the response to real motion. Microstimulation in concert with an actual moving visual stimulus is presumed to cause a population response in MT that corresponds to two different directions of visual motion— the actual direction of motion of the visual stimulus and the preferred direction of the cells being stimulated electrically.

What do monkeys see when this happens? Salzman and colleagues trained monkeys to indicate the perceived direction of motion, and found that they alternated between reporting the real motion direction and the stimulation-induced motion direction, as if perhaps they could see both and were simply picking one of the two on each individual trial3. However, we trained monkeys to track the motion using eye movements, and found that the animals responded as if they saw only the vector average of the two directions4. Both of these experiments likely involved a population response composed of two peaks of activity: the neurons whose preferred direction of motion matched the visual stimulus and the neurons at the tip of the microstimulating electrode. Perceptual judgments were correlated with the locations of these peaks, whereas eye movements were correlated with the vector average of activity in MT. Microstimulation is artificial, of course. What happens when real stimuli moving in two directions are presented? When two patches of moving dots are superimposed on each other (a situation

Bob Crimi

Fig. 1. Electrophysiological recordings in visual area MT of rhesus monkeys by Treue and colleagues1 suggest that the population response to a transparent motion stimulus with two components separated by ±40 degrees is probably the same as the population response to a transparent motion stimulus with three components (+50, 0, –50 degrees). Treue and colleagues predicted that human observers would therefore perceive the two stimuli as containing identical motion. This prediction was confirmed: human observers judged that both stimuli contained the same upward and rightward component, even though in one case this component had an angle of 40 degrees and in the other case it had an angle of 50 degrees.


201



news and views

known as transparent motion), humans can perceive the two stimuli as distinct provided the directions of motion are separated by at least 10 degrees5. Clearly, the center of gravity could not subserve this percept—or else we would always perceive transparent motion as containing only one component of motion at a direction intermediate between the actual directions—but what about the peak(s) of the population response? Does the population response in MT contain separate peaks for each component of a transparent motion stimulus? Do these peaks merge together into one broad peak at the point where the two directions are too close to be resolved? In an elegant series of experiments, Treue and colleagues1 tested this hypothesis. Although the responses of MT neurons to both single and multiple stimuli have been well characterized (for review, see ref. 6), it is less clear how the population response varies as a function of the relative directions of the components of multiple stimuli. Treue and colleagues first studied the responses of monkey MT neurons to transparent motion stimuli. Their results show that because these neurons are broadly tuned for direction, the populations of neurons responding to each component of motion overlap quite extensively. For directions separated by less than about 90 degrees, only a single broad peak exists (although when the directions are farther apart, two separate peaks do appear). Importantly, this single peak occurs in monkey MT even when the directions are sufficiently different to be readily distinguishable to human observers (and presumably to the monkeys). Thus, the relationship between neural activity and perception of the components of transparent motion does not seem to be based on the presence or absence of segregated peaks of activity, as would have been predicted by algorithms that identify peaks of activity (for example, winner-take-all). Rather, the transition from perception of two directions of transparent motion to perception of a single direction of motion must depend on some as-yet unidentified aspect of the shape of the population response in MT. If the overall shape of the population response is critical to motion perception, then Treue and colleagues reasoned that stimuli that produce population responses having the same shape should produce the same percepts. Based on their recordings using two-component stimuli, Treue and colleagues designed three-component stimuli that should produce the same 202

population responses as certain two-component stimuli. For example, the population response to a transparent motion stimulus consisting of two components 80 degrees apart should be the same as the response to a motion stimulus with 3 components each 50 degrees apart (see Fig. 3 of ref. 1). If so, and if motion perception relies on this population of neurons, then the direction of motion of these two stimuli should be indistinguishable. They tested this hypothesis in human observers, and found it was indeed the case: these two very different motion stimuli appear perceptually to have the same components (Fig. 1). A number of issues remain to be resolved. For example, do MT cells actually respond identically to the two- and three-component stimuli? Do the demands of the psychophysical task affect how MT represents motion information? Monkeys can certainly be trained to perform motion tasks like the one used by Treue and colleagues in humans, but there is reason to think that the task itself might influence population responses in MT. In particular, previous work by Treue and others has demonstrated that when an animal is attending to only one of two directions of motion, neurons in MT represent the attended direction much more strongly7–9. Thus, if MT neurons were studied while monkeys performed the psychophysical task used here in human observers, the presence and/or location of peaks in the population response might be different.

Perhaps the most intriguing aspect of this work is the notion that the shape of the population response in MT can be important for motion perception. There are welldefined algorithms for identifying peaks of activity (winner-take-all), or computing the center of gravity (for example, via vector averaging) to arrive at a perceptual judgment or behavioral response, and it is comparatively easy to imagine how neural circuits might perform these calculations (for example, see J.M. Groh, Soc. Neurosci. Abstr. 23, 1560, 1997). Yet the findings of Treue and collegues suggest that perception can be affected by details of the shape of the active population, details that are lost through either of these calculations. Therefore, we need to explore new algorithms for reading population codes. 1. Treue, S., Hol, K. & Rauber, H.-J. Nat. Neurosci. 3, 270–276 (2000). 2. Albright, T. D. in Visual Motion and its Role in the Stabilization of Gaze (eds. Miles, F. A. & Wallman, J.) 177–201 (Elsevier, New York, 1993). 3. Salzman, C. D. & Newsome, W. T. Science 264, 231–237 (1994). 4. Groh, J.M., Born, R.T. & Newsome, W.T. J. Neurosci. 17, 4312–4330 (1997). 5. Mather, G. & Moulden, B. Q. J. Exp. Psychol. 32, 325–333 (1980). 6. Britten, K. H. & Heuer, H. W. J. Neurosci. 19, 5074–5084 (1999). 7. Treue, S. & Maunsell, J. H. Nature 382, 539–541 (1996). 8. Groh, J. M., Seidemann, E. & Newsome, W. T. Curr. Biol. 6, 1406–1409 (1996). 9. Seidemann, E. & Newsome, W. J. Neurophysiol. 81, 1783–1794 (1999).

T.

ChIPping away at potassium channel regulation Min Li and John P. Adelman Kv4 subunits form A-type potassium channels. To replicate native currents, these subunits require additional factors, now shown to be a family of calcium-binding proteins. In a recent issue of Nature, Kenneth Rhodes and colleagues1 present results that resolve long-standing questions John Adelman is in the Vollum Institute, Oregon Health Sciences University, 3181 S.W. Sam Jackson Park Road, Portland, Oregon 972013098, USA. Min Li is in the Department of Physiology, Johns Hopkins University School of Medicine, Baltimore, Maryland 21205, USA. e-mail: [email protected]

concerning the molecular identity of Atype potassium channels. They describe the isolation and characterization of a family of calcium-binding proteins, the KChIPs (K + channel interacting proteins; Fig. 1), that bind to the intracellular amino (N)-terminal domain of cloned Kv4 channels and endow them with many of the properties ascribed to native A-type potassium channels. Coexpression of the KChIPs and cloned




news and views

Kv4 subunits increased current densities, shifted the voltage dependence of activation and speeded their recovery from inactivation. Mutagenesis experiments that eliminated the ability of the KChIPs to bind calcium showed that their channel-modulating effects require calcium binding, although the physical association between the interacting proteins was calcium independent. A-type channels are voltage-dependent potassium channels that activate in the subthreshold range of membrane potentials and completely inactivate during depolarizing pulses, while other voltage-dependent potassium channels are just beginning to activate. As a result, A-type channels influence the time required for membrane depolarization to reach threshold for action potential generation as well as the time between action potential spikes. Thus they are important determinants of the firing frequency in excitable cells such as neurons and cardiac myocytes. Sequence homologies among cloned subunits distinguish several subfamilies of voltage-dependent potassium channels, the Kv subfamilies. The Kv channels are tetramers, and heteromeric channels are assembled only from subunits in the same subfamily. Heterologous expression studies show that members of the Kv4 subfamily form channels similar (but not identical) to native A-type potassium channels 2. In particular, they demonstrate inactivation mediated by a specialized intracellular domain at the extreme N terminus, the ‘ball’, which physically occludes the pore when the channel is open. The expression profiles of Kv4 subunits in the CNS and heart are consistent with this role, and Kv4 subunits have been identified as the components of A-type potassium channels in rat neostriatal cholinergic interneurons 3 and cardiac ventricular myocytes4. The functional profiles of Kv4 channels expressed in different heterologous cells are variable5, suggesting a contribution by factors present in the host cell. This is further supported by differences between Kv4 channels expressed in Xenopus oocytes with or without rat brain mRNA. Co-expression of the lowmolecular-weight mRNA fraction increases current density, most likely reflecting an increased number of Atype channels. This co-expression also shifts the activation voltage of the channels to more negative potentials and allows faster recovery from inactiva-

tion 6 , increasing their similarity to native channels. In addition, a role for calcium in modulating A-type channels has been suggested from recordings of cholinergic neostriatal neurons, where blocking voltage-dependent calcium channels with cadmium shifts the voltage dependence of A-type current activation and inactivation to more depolarized potentials3. Until now, however, the relationship between these observations has remained obscure. The new report from Rhodes and colleagues 1 reconciles these differences between native A-type potassium channels and cloned Kv4 channels at the molecular level. Using the intracellular N-terminal region of Kv4.3 as bait in a yeast two-hybrid hunt through a rat midbrain cDNA library, they captured and characterized three members of a family of Kv4 channel interacting proteins (KChIPs). The KChIPs interact selectively with Kv4 subunits, and they are expressed in tissues that also express Kv4 subunits. When co-expressed in heterologous cells, KChIPs were localized with Kv4 subunits. KChIPs were selectively

immunoprecipitated with Kv4 subunits from transfected cells and rat brain membrane preparations. Immunocytochemistry confirmed that the two proteins are colocalized at cellular and subcellular levels. Remarkably, co-expression of KChIPs with Kv4 subunits reconstituted the functional characteristics of native A-type channels, as well as replicating the effects of co-expressing rat brain mRNA on cloned Kv4 subunits. Current densities were increased, the voltage dependence of activation was shifted to more hyperpolarized potentials, and the channels recovered from inactivation much more rapidly. The KChIPs range in size from 216 to 256 amino acids. The N-terminal domains, ∼50 amino acids, vary considerably, but throughout their carboxyl (C)-terminal domains, they share ∼70% sequence identity. The conserved regions of the molecules contain four E-F hand calcium-binding motifs. Calcium binding to the KChIPs was confirmed by calcium-dependent mobility shifts. Interestingly, a mutant KChIP-1 that could not bind calcium still inter-

K+

α

K+

α

α KChIP

KChIP

α

KChIP

Ca2+

KChIP

α

KChIP Golgi

Ca2+

α KChIP Ca2+

Fig. 1. KChIPs are integral components of A-type potassium channels. The KChIPs bind to the N-terminal domain of Kv4 subunits, close to the membrane, and close to the ‘ball’ domain that mediates channel inactivation. The physical association between KChIPs and Kv4 subunits does not require calcium binding, but the effects on channel gating are calcium dependent. The KChIPs are also present in the Golgi, where the association with Kv4 channels during their biosynthetic development may regulate the levels of functional A-type potassium channels present in the surface membrane.


203



news and views

acted with Kv4 subunits, suggesting that the association between channels and KChIPs is calcium independent and may be constitutive. However, the mutant KChIP-1 no longer modulated Kv4 channel function. Sequence alignments show that KChIPs belong to the recoverin family of calcium-binding proteins that includes the Drosophila protein frequenin, which regulates transmitter release, and mammalian homologs NCS-1 (neuronal calcium sensor-1) and hippocalcin7. Although KChIPs 1 and 2 have not been previously described, KChIP-3 is identical to DREAM (downstream regulatory element antagonist modulator), which is reported to regulate transcription in a calcium-dependent manner8. KChIP-3 is also identical to calsenilin, a protein that resides in the ER and Golgi and interacts with the C-terminal domains of the presenilins, transmembrane proteins found in the same subcellular compartments. Mutations in the presenilin genes account for 40 percent of early-onset familial Alzheimer disease and sensitize neuronal cells to apoptosis, possibly by disrupting intracellular calcium levels9. In this regard, it is tempting to speculate that the increased current density (thought to reflect an increased number of channels in the membrane) induced by co-expression of KChIPs with Kv4 subunits may reflect effects on membrane trafficking, which could be related to the interactions between KChIP-3 and presenilins in the trans Golgi. Taken together, the results suggest that KChIPs are integral parts of Kv4 channels, both in the plasma membrane and during their biosynthetic journey through the trans Golgi. On elevation of intracellular calcium levels, such as during an action potential, calcium binding to KChIPs induces a conformational alteration that is rapidly transduced to the channel, resulting in altered gating. The previous localization of KChIP-3 to the trans Golgi suggests that the increased current densities may result from regulated cell-surface expression of the channel complexes. Several questions await further experiments. For example, just how does calcium binding to the KChIP result in altered channel activity? Which amino acids in the N-terminal domain of Kv4 subunits interact with KChIPs? Does the interacting domain of Kv4 overlap with binding sites for other proteins? How selective are the different 204

KChIPs for the various Kv4 subunits? What other proteins might KChIPs bring into the A-type channel complex? This is the latest in a series of findings suggesting that intracellular calcium signaling can modulate membrane potential. Fluctuations in intracellular calcium levels have long been appreciated as an important modulator of ion channel activity. A generally accepted model posits that second messenger systems, such as calcium-sensitive protein kinases or phosphatases, alter the phosphorylation status of the channel, affecting channel activity and cellular excitability10. These are relatively slow processes that require calcium interaction with the signaling molecule and subsequent interaction with the ion channel. Recently, several reports have suggested a faster calcium signaling process by demonstrating that calmodulin is constitutively bound to the ion-conducting α subunits of voltage-dependent calcium channels (VDCCs) and small conductance calcium-activated potassium channels (SK channels). In both cases, compelling evidence supports a model in which calmodulin is an integral part of the channel complex, and calcium binding to calmodulin induces structural alterations in calmodulin, which are transduced into conformational changes in the channel proteins that alter their function 11,12 . Local signaling induced by calcium entry may be much more rapid than second-messenger-mediated processes, and can respond rapidly to discrete, localized alterations in intracellular calcium. For example, VDCCs are the likely source of the calcium ions that would bind to channel-associated calmodulin, and the precise distance between SK channels and VDCCs or intracellular calcium release sites strongly affects the dynamics of burst frequency. The work by Rhodes and colleagues1 suggests that KChIPs are an integral component of Atype potassium channels, acting as direct calcium sensors that affect channel gating properties, analogous to the role of calmodulin for SK channels and VDCCs. The KChIPs join an expanding list of proteins that bind to potassium channels and influence their activity, but do not contribute to ion conduction. Among the voltage-dependent potassium channels, the intracellular N-terminal domain has been shown to mediate heteromeric subunit assembly, restrict-

ing associations to members of the same subfamily. The N-terminal region also interacts with the Kvβ proteins, some of which endow inactivation that is modulated by protein kinases13. More recently, two alternatively spliced proteins, ZIP1 and ZIP2, have been identified that act as molecular bridges, linking the β subunits to protein kinase C ζ (ref. 14). At their C termini, many voltage-dependent potassium channels bind to PDZ-containing proteins, such as PSD-95/SAP90 family members, to modulate the distribution and surface expression of the potassium channels, which is affected by the presence of a β subunit at the N-terminal domain in some cases 15 . These PDZ-containing proteins, in turn, interact with other putative regulatory molecules. The emerging picture suggests that the α subunits of potassium channels are embedded in large, multimeric protein complexes with components that sense a wide range of metabolic signals. Indeed, the ZIP proteins, which do not themselves interact with the α subunits, begin to define a larger microdomain, a molecular neighborhood, in which the channels reside. The two-hybrid screen can be used with each newly identified resident to determine the next nearest neighbor. 1. An, W. F. et al. Nature 403, 553–556 (2000). 2. Serodio, P., Vega-Saenz de Miera, E. & Rudy, B. J. Neurophysiol. 75, 2174–2179 (1996). 3. Song, W.-J. et al. J. Neurosci. 18, 3124–3137 (1998). 4. Yeola, S. W. & Snyders, D. J. Cardiovasc. Res. 33, 540–547 (1997). 5. Petersen, K. R. & Nerbonne, J. M. Pflugers Arch. 437, 381–392 (1999). 6. Serodio, P., Kentros, C. & Rudy, B. J. Neurophysiol. 72, 1516–1529 (1994). 7. Pawlowski, K., Bierzynski, A. & Godzik, A. J. Mol. Biol. 258, 349–366 (1996). 8. Carrion, A. M., Link, W. A., Ledo, F., Mellstrom, B. & Naranjo, J. R. Nature 398, 80–84 (1999). 9. Buxbaum, J. D. et al. Nat. Med. 4, 1177–1181 (1998). 10. Levitan, I. B. Annu. Rev. Physiol. 56, 193–212 (1994). 11. Zuhlke, R. D., Pitt, G. S., Deisseroth, K., Tsien, R. W. & Reuter, H. Nature 399, 159–162 (1999). 12. Keen, J. E. et al. J. Neurosci. 19, 8830–8838 (1999). 13. Sheng, M. & Kim, E. Curr. Opin. Neurobiol. 6, 602–608 (1996). 14. Gong, J. et al. Science 285, 1565–1569 (1999). 15. Tiffany, A. M. et al. J. Cell Biol. 148, 147–158 (2000).



news and views

Toying with memory in the hippocampus Howard Eichenbaum and Kristen Harris


Mice lacking NMDA receptors in hippocampal area CA1 are deficient in spatial memory. They also have nonspatial memory deficits, which are overcome by environmental enrichment. “Memory being... altogether conditioned on [the ability to excite] brain-paths, its excellence in a given individual will depend partly on the number and partly on the persistence of these paths.” (William James1, p. 659) In this two-factor view of the biological basis of memory, James characterized the “persistence” factor as a physiological property of one’s brain tissue. He envisioned persistence as a “native tenacity,” except for natural variability among individuals and decline with illness or aging. By contrast, James characterized the “number of paths” factor as very much modifiable with experience. He argued that memory could be improved substantially by establishing a large network of linkages through which one could readily associate, and later access, a new memory. As an example, James described the college athlete who was a “dunce at his books” but could astonish with his ability to remember sports statistics precisely because he had worked at creating a rich knowledge framework for this kind of information. James may have been prescient in proposing that enriching one’s memory network can make up for a lesser native persistence, as in findings from Joe Tsien and colleagues2 in the current issue of Nature Neuroscience. This report extends a recent study that used state-of-the-art molecular genetics to knock out the N-methyl-D-aspartate (NMDA) receptor beginning at postnatal weeks three and four selectively within the CA1 region of the hippocampus3. These mutant mice lack NMDA-receptor-dependent long-term potentiation (LTP), a type of physiological plasticity thought to be a cellular substrate of memory. Correspondingly, they have severely impaired spatial learning and memory4. Now Tsien Howard Eichenbaum is in the Department of Psychology, and Kristen Harris is in the Department of Biology, Boston University, Boston, Massachusetts 02215, USA. e-mail: [email protected] or [email protected]

and colleagues2 show that mice with the same mutation are severely impaired across a broad range of nonspatial learning tests. The authors also address James’s “number of paths” factor by exposing these mice to a complex ‘enriched’ environment, in which they could presumably establish many diverse associations through exploration. By electron microscopy, environmental enrichment was shown to increase the number of synaptic connections within hippocampal area CA1 in normal mice, and surprisingly also in the mutant mice, even without NMDA receptors. Furthermore, enrichment improved learning performance in control mice and almost eliminated the memory deficits observed in the CA1 NMDA receptor-knockout mice. How is it possible that enriched experience can support new memory even Standard

without NMDA-receptor-dependent LTP? Does the observed anatomical plasticity compensate for the impaired functional plasticity, and if so, how? Here we consider two interpretations of Tsien and colleagues2 that differ in the critical locus where increased connectivity ameliorates the loss of NMDA-receptor-dependent LTP in area CA1. First, enhanced intrinsic hippocampal connectivity might compensate for the loss of LTP through NMDA-receptor-independent processes. Second, enhanced connectivity outside the hippocampus, specifically within the neocortex where NMDA-receptor-dependent plasticity is intact, might compensate for a dysfunctional hippocampus. According to the first possibility, the hippocampus might use processes that do not require NMDA receptors after exposure to environmental enrichment. In CA1, LTP can be induced by strong patterned stimulation, even when NMDA receptors are pharmacologically blocked5. This form of LTP is NMDA receptor independent. Tsien and colleagues2 show that many new dendritic spines form, and robust synaptogenesis occurs within CA1 after enrichment experience (Fig. 1b) in both mutant and control mice. This synaptogenesis does not depend on NMDA receptors because an equal number of new dendritic spines and synapses formed in mutant and conEnriched

Cortex

Hippocampus

Bob Crimi

Fig. 1. Environmental enrichment increases the connections within the hippocampus and neocortex. (a) Under control conditions, there are fewer synapses within both the hippocampus and cortex, as well as between these areas. The pathway through the hippocampus could be required to connect distinct representations in the neocortex (red and blue), and this capacity could be mediated by strengthening the existing connections within the hippocampus using NMDA receptors (yellow). (b) After exposure to an enriched environment, more connections are formed within both the hippocampus and neocortex, and perhaps between these areas. Strengthening of additional non-NMDA receptor connections (orange) within the hippocampus, or between the hippocampus and cortex, may suffice to improve memory. Alternatively, the additional connections within the neocortex (green) may suffice to link distinct neocortical representations and thereby ‘short circuit’ the hippocampal contribution.


205



news and views

trol mice following environmental enrichment. Other studies have shown robust synaptogenesis in the adult brain when synaptic activity is silenced pharmacologically6,7. The new spines form either when presynaptic release of neurotransmitter is blocked or when postsynaptic glutamate receptors are blocked, and new spines can last for at least eight hours without subsequent activation. Furthermore, induction of NMDA-receptor-dependent LTP in hippocampal area CA1 does not require the formation of new synapses8,9. Together with the findings from Tsien and colleagues, these studies show that dendritic spines can form in the mature brain without NMDA-receptor-dependent processes like LTP, and even without synaptic activity. Perhaps the new spine synapses can facilitate NMDA-receptor-independent processes within the hippocampus to enhance subsequent learning and memory in CA1-NMDA knockout mice. How might the enrichment-induced dendritic spines within the hippocampus facilitate learning and memory? Hebb10 originally suggested that learning and memory occurs by strengthening some connections and weakening other, inappropriate connections. Tsien and collegues show that the enrichment effects are specific to a particular type of spine synapse, causing an increase only in those with a continuous (that is, ‘non-perforated’) postsynaptic surface. There was no change in the frequency of large irregularly shaped synapses, those with ‘perforated’ postsynaptic surfaces. Thus, the non-perforated synapses might enhance some forms of learning and memory via NMDA-receptor-independent mechanisms. Other studies have shown a transient elaboration of a subset of perforated synapses with NMDA-receptor-dependent LTP 11. An open question is whether NMDA-receptor-dependent changes at perforated synapses might be involved in refinement of synaptic connections during more complex learning protocols than those tested by Tsien and colleagues2. Either way, these findings are among the first to demonstrate a possible role for non-perforated synapses in learning and memory. Understanding the function of the small nonperforated synapses is especially important because these are normally the most abundant synapse type (> 75%) in both hippocampus and neocortex. The second possible explanation for the findings of Tsien and colleagues2 is that the hippocampus can be short-circuited altogether during learning and memory if environmental enrichment 206

can induce enough connectivity outside the hippocampus, specifically within the neocortex. Tsien and colleagues did not examine the cortex, but previous evidence indicates that enriched experience increases intrinsic connectivity within the neocortex12. It is clear that memory is not mediated solely by CA1, or even by the entire hippocampus alone. Rather, the hippocampus is part of a memory system that prominently involves its bidirectional connections with diverse and interconnected regions of the cerebral cortex13 (Fig. 1). Within this system, memories are likely ‘stored’ among large cell assemblies widespread across the cortex, and the organization of associations is mediated by the formation of links between the cell assemblies10. The role of the hippocampus may be to facilitate the consolidation of these cortical linkages by storing aspects of new information, or indices pointing to cortical loci of new representations, and using these to temporarily link otherwise separated cortical memories (Fig. 1a). We know that the role of the hippocampus is temporary because it is not necessary for the recall of longestablished memories, suggesting that eventually new intracortical connections form to mediate permanent links14. The increase in synaptic connectivity in neocortex, likely to have occurred as a result of enriched training experience12, might be so effective that lasting plasticity within the hippocampus is not required (Fig. 1b), at least for the relatively simple types of learning examined by Tsien and colleagues2. One way to distinguish the ‘cortical hypothesis’ from the ‘hippocampal

hypothesis’ discussed above would be to determine whether the CA1-NMDA knockout mice after enrichment can tolerate loss of hippocampal area CA1 and still enjoy improved learning and memory. An early study15 found that enriched experience reduced, but did not eliminate, the effects of hippocampal damage on spatial learning. These findings are consistent with the possibility that both putative mechanisms contribute to the effects of enrichment. 1. James, W. The Principles of Psychology (Holt, New York, 1890). 2. Rampon, C. et al. Nat. Neurosci. 3, 238–244 (2000). 3. Tsien, J. Z. et al. Cell 87, 1317–1326 (1996). 4. Tsien, J. Z., Huerta, P. T. & Tonegawa, S. Cell 87, 1327–1338 (1996). 5. Morgan, S. L. & Teyler, T. J. J. Neurophysiol. 82, 736–740 (1999). 6. Bravin, M., Morando, L., Vercelli, A., Rossi, F. & Strata, P. Proc. Natl. Acad. Sci. USA 96, 1704–1709 (1999). 7. Kirov, S. A. & Harris, K. M. Nat. Neurosci. 2, 878–883 (1999). 8. Muller, D. Rev. Neurosci. 8, 77–93 (1997). 9. Sorra, K. E. & Harris, K. M. J. Neurosci. 18, 658–671 (1998). 10. Hebb, D. O. The Organization of Behavior (Wiley, New York, 1949). 11. Toni, N., Buchs, P. A., Nikonenko, I., Bron, C. R. & Muller, D. Nature 402, 421–425 (1999). 12. Klintsova, A. V. & Greenough, W. T. Curr. Opin. Neurobiol. 9, 203–208 (1999) 13. Eichenbaum, H. Annu. Rev. Psychol. 48, 547–572 (1997). 14. Squire, L. R. & Alvarez, P. Curr. Opin. Neurobiol. 5, 169–177 (1995). 15. Hughes, K. R. Can. J. Psychol. 19, 325–332 (1965).

Attention - brains at work! Roger B.H. Tootell and Nouchine Hadjikhani Two new studies use event-related fMRI to reveal a network of brain regions that are activated during different steps in the control of visual spatial attention. The amount of information that is potentially available through our sense organs is far greater than our brains can The authors are at the Magnetic Resonance Imaging Center, Department of Radiology, Massachusetts General Hospital, 149 13th Street, Charlestown, Massaschusetts 02129, USA. e-mail: [email protected]

handle. Much of this information must therefore be discarded, and the brain must select only those stimuli that are of greatest relevance for further processing. Understanding how this occurs is a major challenge for cognitive neuroscience, and two papers1,2 in the current issue of Nature Neuroscience provide the most detailed spatio-tem-



news and views

subjects know where to expect the stimulus, compared to trials in DLPF SPL IPL which they do not know or are misdirected. PS Attention has someLO times been likened to a Cue spotlight, and funca tional neuroimaging has recently allowed researchers to see the M ‘beam’ directly3–8. SevSPL IPL eral studies have conTPJ firmed that attention is mapped topographically in all the early Target (valid) (retinotopic) visual b cortical areas, and that when attention is directed to a particular M location, the part of the SPL IPL cortex that represents TPJ that location becomes increasingly responsive. But where and how are Target (invalid) attention signals first c Bob Crimi generated—in other Fig. 1. Brain areas activated in a 'naked' eye and brain, from a subject words, how is the who was facing a display screen and doing a covert attention task, spotlight controlled? similar that used in Hopfinger et al.1 and Corbetta et al.2 (a) A cue Answering this quesinstructs the subject to attend to a given location: here to the sub- tion is now an imporject's left. Then the attention target appears (here a checkerboard tant goal for the field9. stimulus, but usually a more subtle stimulus change), at either the One problem with expected location (b) prompted by the 'valid' cue, or at an unexmost previous neupected location (c), misdirected by the 'invalid' cue in (a). The actiroimaging studies of vated areas described in Hopfinger et al. and Corbetta et al. include DLPF (dorsolateral prefrontal cortex), IPL (inferior parietal lobule), attention (as well as LO (lateral occipital region of visual cortex), M (supplementary many other cognitive motor region), PS (peri-sylvian), SPL (superior parietal lobule), TPJ processes) is that they (temporal-parietal junction) and VP (ventral parietal region). have used a ‘block’ Additional areas were activated but are not visible from this vantage design, in which the point (see refs. 1 and 2). High levels of activity are shown in red, and hemodynamic signal is lower levels of activity are shown in green. averaged over many similar trials. This generates a static activation map that represents the average activation for a particular task, poral views yet of the brain structures without giving any information about that control the deployment of visual the individual steps involved. Yet spatial spatial attention. attention is inherently dynamic, and our Our focus of attention is constantly brains are constantly choosing new locashifting, either automatically—in response tions of interest, disengaging attention to an ‘attention-grabbing’ stimulus—or and (often) eye position from previously voluntarily. Usually, an attentional shift is attended locations, and shifting attention followed by an eye movement to the and eye position to new targets. It is difnewly attended location, but it is also posficult to resolve these different steps using sible to attend to a location without a block design. looking at it; we are sometimes forced to The new studies1,2 avoid this problem do this during demanding visual tasks (such as driving on a busy road), where it by using ‘event-related’ designs. Eventis impossible to fixate all items of interest related fMRI is a relatively new analytical simultaneously. In the laboratory, these method, in which the hemodynamic sigso-called ‘covert’ attentional shifts nal is analyzed on a trial-by-trial basis to can be detected because reaction identify patterns that occur at fixed times times are shorter for trials in which after a given event, such as cue or target


VIP

VIP


presentation. Unlike block designs, eventrelated designs can reveal the time course of the response during an individual trial, making it possible to identify different patterns of activation associated with different components of the task. Both groups used covert attention tasks, thus avoiding any complications due to eye movements. The subjects were instructed to fixate on the center of a screen and then shift their focus of attention to either the left or the right, as indicated by a cue at the fixation point. A few seconds later, a target appeared either at the cued location (‘valid cue’ condition)1,2 or on the opposite side (‘invalid’ condition)2, and subjects had to respond to it. Both groups confirmed that their subjects really were making attentional shifts during the task. Corbetta et al.2 showed that their subjects’ reaction times were faster after valid than invalid cues, and Hopfinger et al.1 showed that the neural activity evoked by the arrow cue (which, being in the center, could activate both hemispheres) was greater in the hemisphere that represents the cued side. Despite their similar techniques, the two studies addressed different questions and yielded complementary results. Hopfinger et al.1 made few prior assumptions, and simply asked which brain regions were activated in response to the cues (reflecting an attentional shift) and which were activated by the subsequent target presentation (reflecting processing of the attended stimulus). Cues and targets both activated a number of different regions; the surprising finding was that there was relatively little overlap between the two sets of responses, suggesting that the brain structures that control spatial attention are largely distinct from those that participate in the processing of the attended stimulus. In a more hypothesis-driven approach, Corbetta et al.2 tested two specific proposals regarding the role of parietal cortex in attention. Based on studies of brain-damaged patients, it has been suggested that the region around the temporal-parietal junction (TPJ) is involved in reorienting attention toward stimuli at unexpected locations, and that the region around the intraparietal sulcus (IPs) is involved in voluntary orientation and maintenance of attention at cued locations. As described below, their data support both these ideas, and provide a view of parietal function that is complementary to, and largely consistent with, that of the other study. Both groups agree on the role of a posterior parietal region in and around the 207



news and views

intraparietal sulcus; this region is activated in response to the cue and remains active as attention is maintained, but shows a reduced response once the target is presented (regardless of whether it appears at an expected or unexpected location)2. Similar responses have been observed in electrophysiological recordings from alert monkeys (see ref. 2 for references), and the combined evidence from physiology and neuroimaging strongly suggests that the posterior parietal cortex is involved in selecting a location and retaining it in working memory (although other areas may also be involved—see below). Interestingly, the greater region of posterior parietal activation may include the visual cortical area V7, which is retinotopically organized (albeit crudely), suggesting a possible role in the spatial allocation of attention3,9,10. Another popular candidate for storing spatial cues in working memory is the dorsolateral prefrontal cortex (see ref. 1 and references cited therein). Hopfinger et al.1 observed activation of this region during the cueing period, but Corbetta et al.2—using a better analytical method that avoided prior assumptions about the time course of the hemodynamic response—found that the activation in the prefrontal cortex was more transient than that observed in the intraparietal cortex. Thus, the intraparietal cortex seems to be the stronger candidate for storing spatial working memories, although it is possible that both regions are involved, particularly as they are known to be interconnected in monkeys (and presumably in humans too). Most previous models of visual attention have assumed that it is controlled by higher cortical regions, which regulate the processing of sensory inputs in lower regions via top-down projections. This makes sense because decisions about allocating attention are often based on highlevel features that are not represented at the earliest stages of the cortical hierarchy. However, alternative, bottom-up models have also been proposed 11 , in which attention arises as an emergent property from competitive interactions at each level in the hierarchy. The new findings do not completely exclude the latter model, but they are more consistent with a top-down model. In particular, both groups found that a cue instructing subjects to attend to a particular location caused increased activation of the corresponding parts of the early retinotopic visual areas, even before any stimulus appeared. It is difficult to see how this 208

could happen except through top-down signals, and the challenge now will be to identify the anatomical connections that underlie these effects. The Corbetta et al. 2 study has some interesting clinical implications. For many years, it has been known that damage to the right parietal cortex, particularly the temporal-parietal junction, causes a complex syndrome known as unilateral visual neglect (reviewed in ref. 12). Parietal neglect patients have problems attending to and responding to objects on in the left visual field; for instance, they often bump into objects on their left, and when asked to draw what they see, they tend to neglect what is in the left visual field. The syndrome has attracted a great deal of interest, not only for its clinical importance but also because of its implications for normal perceptual mechanisms. One interpretation of parietal neglect is that the TPJ is responsible for disengaging attention from its present focus and redirecting it to a new target. Corbetta et al.2 now provide elegant support for this hypothesis. Unlike other parts of the parietal cortex, the TPJ showed little or no response to the initial cue, but it responded strongly to the subsequent presentation of the target. Moreover, the TPJ response was much stronger for invalid than for valid targets, suggesting that it is specifically involved in reorienting of attention in cases where the target appears at an unexpected location. Finally, the TPJ response was always stronger in the right hemisphere than the left, regardless of the side where the target was presented. This right lateraliza-

tion fits very well with the clinical literature on parietal neglect, and the link seems even more compelling given that the other activations seen in these studies were not lateralized. In conclusion, these two papers demonstrate the power of new imaging techniques to resolve complex cognitive operations into their component steps, and to reveal the neural structures involved in each step. They are likely to stimulate many future studies, and by combining ever-better imaging methods with other approaches such as patient studies and physiology of non-human primates, we can hope to gain a new depth of understanding of how the brain controls attention. 1. Hopfinger, J. B., Buonocore, M. H. & Mangun, G. R. Nat. Neurosci. 3, 284–291 (2000). 2. Corbetta, M., Kincade, J.M., Ollinger, J.M., McAvoy, M.P. & Shulman, G.L. Nat. Neurosci. 3, 292–297 (2000). 3. Tootell, R. B. H. et al. Neuron 21, 1409–1422 (1998). 4. Brefczynski, J. & DeYoe, E. Nat. Neurosci. 2, 370–374 (1999). 5. Somers, D. C. et al. Proc. Natl. Acad. Sci. USA 96, 1663–1668 (1999). 6. Ghandi, S., Heeger, D. & Boynton, G. Proc. Natl. Acad. Sci. USA 96, 3314–3319 (1999). 7. Martinez, A. et al. Nat. Neurosci. 2, 364–369 (1999). 8. Kastner, S. et al. Science 282, 108–111 (1998). 9. Corbetta, M. Proc. Natl. Acad. Sci. USA 95, 831–838 (1998). 10. Culham, J. C. et al. J. Neurophysiol. 80, 2657–2670 (1998). 11. Desimone, R. & Duncan, J. Annu. Rev. Neurosci. 18, 193–222 (1995). 12. Mesulam, M. M. Ann. Neurol. 10, 309–325 (1981).

Signaling dendritic growth in vivo Small GTPases of the Rho family affect cell morphology by regulating the cytoskeleton, and they have been implicated in neurite outgrowth. On page 217 of this issue, Holly Cline and colleagues (Cold Spring Harbor Laboratory, New York) report that RhoA, Rac and Cdc42 regulate different aspects of dendritic growth in vivo. The authors used vaccinia virus to express constitutively active or dominantnegative forms of these GTPases in albino Xenopus tadpoles. Time-lapse imaging of DiI-labeled neurons showed that constitutively active Rac and, to a lesser extent, Cdc42 increased branch addition and retraction. Activation of endogenous RhoA promoted the elongation of existing branches. Cline has previously shown that blocking NMDA receptors reduces dendritic growth, and the dominantnegative form of RhoA prevented this effect, suggesting that RhoA may act downstream of NMDA receptors to control dendritic development.

Sandra Aamodt



brief communications Three-dimensional spatial selectivity of hippocampal neurons during space flight


James J. Knierim1,2, Bruce L. McNaughton1 and Gina R. Poe1 1

University of Arizona, Arizona Research Laboratories Division of Neural Systems, Memory & Aging, 384 Life Sciences North Bldg., Tucson, Arizona, 85724, USA

2

University of Texas-Houston Medical School, Department of Neurobiology & Anatomy and the W. M. Keck Center for the Neurobiology of Learning and Memory, P.O. Box 20708, Houston, Texas 77225, USA Correspondence should be addressed to B.L.M. ([email protected])

‘Place’ cells of the hippocampus and ‘head-direction’ (HD) cells of the thalamus and limbic cortex derive their spatial and directional specificity from a combination of idiothetic (self-motion) cues and external landmarks, which normally reinforce each other to generate a robust neural code for location and direction1. In weightlessness, however, three-dimensional navigation can cause the idiothetic and landmark cues to conflict. Nonetheless, neural recordings on the space shuttle demonstrated that the hippocampus can create a robust spatial code for three orthogonal surfaces in the weightless environment of space flight. The firing properties of place cells and HD cells are coupled2, and one function of the HD system may be to orient the ‘cognitive map’ in the hippocampus3. As an animal explores a novel environment, vestibular input and other idiothetic cues are thought to keep the HD system aligned with external landmarks long enough for the landmarks to form stable associations with, and thereby exert control over, place and HD cells2–7. In normal gravity, HD cells are sensitive to only the horizontal component of head direction; although changes in pitch and roll attitude are not signaled directly by these cells8,9, calculation of head direction in the horizontal plane may involve compensation for such changes. Because the otolith organs—normally, a major source of information about static pitch and roll attitude—are rendered useless in zero gravity, the HD system would be deprived of this input in compensating for changes in pitch and roll (although the semicircular canals would still detect angular accelerations in all three dimensions). Three-dimensional navigation in microgravity might thus lead to inconsistent associations between HD and landmark information and a consequent inconsistency in A the hippocampal place code. In light of the disa orientation frequently reported by astronauts

during space flight10, it was of interest to determine whether the hippocampus can create a stable representation of space during three-dimensional movement in the absence of gravity. During the Neurolab Space Shuttle mission of April, 1998, ensembles of place cells were recorded from three rats implanted with a multi-electrode recording array11,12. The rats were trained to negotiate a three-dimensional track (the ‘Escher staircase’) in which 3 turns of 90° in yaw were interleaved with 3 turns of 90° in pitch (Fig. 1). As a result, the rat completed a full circuit of the track and returned to its starting location/direction after having made only 3 right turns (270° total yaw). The spatial information provided by external landmarks was thus presumably in conflict with the direction information from HD cells, which under normal conditions require a fourth 90° turn (360˚ total yaw) to signal a return to the starting direction. Nonetheless, place cells eventually demonstrated normal, spatially specific firing properties. Spatial firing patterns of 16 active place cells from rat 2 were recorded on the ninth day of flight (FD9; Fig. 2a), which was the second day in which the rats had been exposed to the Escher staircase track in flight. For comparison, we show the spatial firing patterns of 12 representative place cells from the same rat recorded 4 days before launch as the rat ran clockwise on a flat, rectangular track (Fig. 2b). An index of spatial tuning specificity, which quantifies the amount of information about the rat’s position transmitted by the firing of a single spike13, did not significantly differ between cells recorded preflight and those recorded on FD9 (for active cells, defined as having a mean firing rate > 0.05 Hz on the track; mean information per spike ± s.e. preflight, 1.12 ± 0.08 bits, n = 19; FD9, 1.13 ± 0.09 bits, n = 16; not significant by Mann-Whitney). We also determined the spatial tuning of 21 active cells from rat 1 on FD9 (Fig. 2c). The mean spatialinformation content for these cells (1.19 ± 0.09) did not significantly differ from that rat’s preflight data (1.37 ± 0.10; n = 28). The mean spatial-information content for all 7 active place cells from rat 3 on the fourth day of flight (FD4) was 1.12 ± 0.27 bits (Fig. 3a), which was not different from the information content measured for this rat’s place cells before the flight (1.48 ± 0.11 bits; n = 24). Another recording session followed immediately, and most cells maintained the same firing fields in both sessions (Fig. 3b), demonstrating that the spatial tuning was stable across different exposures to the track. The rats occasionally turned around on the track and moved counterclockwise for short periods. Some cells demonstrated place fields when the rat was moving only counterclockwise through a single location, thus demonstrating the direction selectivity that is seen on such tracks under normal terrestrial conditions14. Hippocampal EEG activity was also

B b

Fig. 1. Escher staircase track. (a) To obtain stimulation of the medial forebrain bundle as a reward, the rat moved along the track by grasping the edges of the track and propelling itself forward. (b) Normal place field recorded from rat 3. Red indicates maximal firing rate (> 5 spikes per s), and blue indicates positions sampled for which the cell never fired. Locations indicated outside the black outline of the Escher staircase were sampled when the rat’s head moved off the track. Although all statistical analyses included these off-track data, they were deleted in the remaining figures for clarity of illustration. nature neuroscience • volume 3 no 3 • march 2000

211



brief communications

a

Rat 2, flight day 9

b

Rat 2, preflight

c

Rat 1, flight day 9

may require either a period of adaptation to microgravity or more experience with the environment than is typically required in normal gravity. It remains to be determined whether the hippocampal code in microgravity can fully represent three dimensions, or whether the system adapts by developing independent two-dimensional representations for each orthogonal surface. It is also unknown what cues drive the firing of place cells under these conditions. Although the eventual formation of stable fields suggests that the visual landmarks may be primary, other contributing factors may include adaptive mechanisms that alter the efficacy of idiothetic cues during extended exposure to microgravity (for instance, changes in the vestibular system or learned ability to path integrate in three dimensions). Despite these unanswered questions, our recordings of CNS neurons from freely moving mammals in space demonstrate the feasibility of performing such complex neurophysiological and behavioral experiments in the microgravity environment. Further experimentation in the international space station may yield a better understanding of the effects of prolonged space flight on various components of the nervous system as well as insight into their normal function on Earth.

ACKNOWLEDGEMENTS

Fig. 2. Representative place fields. (a) Normal place fields from rat 2 recorded on FD9. The number inside each map indicates the firing rate coded by red (for instance, > 1 spike per s for the first cell). (b) Preflight place fields from rat 2 as the rat ran on a rectangular track. (c) Normal place fields from rat 1 on FD9.

recorded during baseline and behavioral sessions in all rats on FD4 and FD9. Although the small number of subjects precluded a statistical analysis, they showed normal theta rhythm during active locomotion and normal sharp waves and ripples15 during quiet inactivity in the sleeping pouch on both flight days. It is interesting to note that on the first experience on the Escher staircase on FD4, firing of place cells showed abnormal patterns of spatial selectivity that differed between rats 1 and 2 (J.J.K., B.L.M. and G.R.P., unpublished observations). Thus hippocampal cells can form unique, reliable representations of position on three orthogonal surfaces in microgravity, but they

a

Session2 2 A. Session

5

1

5

1

1

8

1

3

1

1

3

1

b

Session B. Session 33

2

1

Fig. 3. Stability of place fields across sessions. (a) Normal place fields from rat 3 on FD4 recorded on second exposure to the track. (Data from the first exposure were lost because of technical problems.) (b) Place fields for the same seven cells on a third run shortly after the second. Most place cells maintained the same firing locations, although the fourth cell lost its field and a few other cells (not shown) gained a field in session 3; such changes in responses of a minority of place cells are not uncommon. 212

We thank the crew of STS-90 (Scott Altman, Jay Buckey, Alex Dunlap, Kay Hire, Rick Linnehan, Chiaki Mukai, Jim Pawelczyk, Rick Searfoss and Dave Williams), Bryan Roberts, Lisa Baer, Mike Eodice, Laurie Dubrovin, Tom Howerton, Louis Ostrach, Chris Maese, Justine Grove, Ali Werner, Dave Bergner, Tom McCarthy, George Swaiss, Steve Carmen, Jim Cockrell and others at NASA-Ames. We also thank Casey Stengel (who designed and built the recording system), Krzystof Jagiello (who designed and wrote the data acquisition software), Kathy Dillon, Shanda Roberts, Shane Smith, Vince Pawlowski, Carol Barnes, Katalin Gothard, Veronica Fedor-Duys, Mark Bower, Karen Reinke, Chris Duffield, Luann Snyder, Doug Wellington and others at the University of Arizona. Additionally, we thank Bill Skaggs and Matt Wilson, who wrote much of the data analysis software, and science and engineering support and management teams at Johnson Space Center and Kennedy Space Center. Supported by grants NAG 2-949 from NASA; NS33471 and NS20331 from NIH; and N0014-98-1-0180 and N0014-96-1-1082 from ONR.

RECEIVED 3 SEPTEMBER 1999; ACCEPTED 1 FEBRUARY 2000 1. Redish, A. D. Beyond the Cognitive Map (MIT Press, Cambridge, Massachusetts, 1999). 2. Knierim, J. J., Kudrimoti, H. S. & McNaughton, B. L. J. Neurosci. 15, 1648–1659 (1995). 3. O’Keefe, J. & Nadel, L. The Hippocampus as a Cognitive Map (Clarendon, Oxford, 1978). 4. Muller, R. U. & Kubie, J. L. J. Neurosci. 7, 1951–1968 (1987). 5. Goodridge, J. P., Dudchenko, P. A., Worboys, K. A., Golob, E. J. & Taube, J. S. Behav. Neurosci. 112, 749–761 (1998). 6. McNaughton, B. L. et al. J. Exp. Biol. 199, 173–185 (1996). 7. Jeffery, K. J. & O’Keefe, J. M. Exp. Brain Res. 127, 151–161 (1999). 8. Taube, J. S., Muller, R. U. & Ranck, J. B. Jr. J. Neurosci. 10, 420–435 (1990). 9. Taube, J. S. Prog. Neurobiol. 55, 225–256 (1998). 10. Oman, C. M. in Proceedings of the Symposium on Vestibular Organs and Altered Force Environment (eds. Igarashi, M. & Nute, K.) 25–37 (NASA Space Biomedical Research Institute, Houston, Texas, 1988). 11. Wilson, M.A. & McNaughton, B. L. Science 261, 1055–1058 (1993). 12. Gothard, K. M., Skaggs, W. E., Moore, K. M. & McNaughton, B. L. J. Neurosci. 16, 823–854 (1996). 13. Skaggs, W. E., McNaughton, B. L., Wilson, M. A. & Barnes, C. A. Hippocampus 6, 149–172 (1996). 14. McNaughton, B. L., Barnes, C. A. & O’Keefe, J. Exp. Brain Res. 52, 41–49 (1983). 15. Buzsaki, G. Brain Res. 398, 242–252 (1986).



articles

NMDA receptor-mediated control of protein synthesis at developing synapses A. J. Scheetz1,4, Angus C. Nairn2 and Martha Constantine-Paton1,3 1

Department of Molecular, Cellular and Developmental Biology, Yale University, Kline Biology Tower, P.O. Box 208103, New Haven, Connecticut 06520-8103, USA

2

Laboratory of Molecular and Cellular Neuroscience, Rockefeller University, Box 296, 1230 York Avenue, New York, New York 10021, USA

3

Present address: Massachusetts Institute of Technology, Department of Biology, Building 68, Rm 380, 77 Massachusetts Ave., Cambridge, Massachusetts 02139-4307, USA

4

Present address: Department of Molecular Biophysics & Biochemistry, Yale University, 333 Cedar St., P.O. Box 208024, New Haven, Connecticut 06520-8024, USA


Correspondence should be addressed to A.J.S. ([email protected])

We demonstrate a rapid and complex effect of N-methyl-D-aspartate receptor (NMDAR) activation on synaptic protein synthesis in the superior colliculi of young rats. Within minutes of receptor activation, translation of alpha Ca2+/calmodulin dependent kinase II (αCamK II) was increased, whereas total protein synthesis was reduced. NMDAR activation also increased phosphorylation of eukaryotic elongation factor 2 (eEF2), a process known to inhibit protein translation by reducing peptide chain elongation. Low doses of cycloheximide, which reduce elongation rate independently of eEF2 phosphorylation, decreased overall protein synthesis but increased αCaMK II synthesis. These observations suggest that regulation of peptide elongation via eEF2 phosphorylation can link NMDAR activation to local increases in the synthesis of specific proteins during activity-dependent synaptic change.

Activity-dependent synaptic plasticity often involves changing the function of a subset of a neuron’s synapses in response to activation of neurotransmitter receptors. Dendritic synthesis of specific proteins is implicated in many of these modifications1,2. Activitydependent translation of mRNA into protein in dendrites may occur during synaptogenesis, when young neurons must selectively stabilize subsets of inputs and eliminate others based on their pattern of activity3. The presence of polyribosomes at young synapses supports this idea4,5. Furthermore, activation of NMDARs is associated with synaptic phosphorylation of eEF2 (ref. 6), and plasticity of young contacts requires activation of NMDARs. In the rat retinocollicular pathway, blocking NMDAR activation during development disrupts normal synaptogenesis7 and reduces levels of αCamK II (ref. 8), a protein widely implicated in synaptic plasticity9. To examine the role of local protein synthesis in this pathway, we used isolated synaptic preparations from the superficial layers of the postnatal rat superior colliculus (sSC). We found that brief NMDAR stimulation reduced overall protein synthesis but rapidly increased αCaMK II synthesis. The fast NMDAR-mediated regulation of protein synthesis was temporally correlated with increased eEF2 phosphorylation, a modification that inhibits protein synthesis by reducing peptide chain elongation10–13. We also found that inhibition of protein synthesis elongation independent of eEF2 phosphorylation similarly increased αCaMK II synthesis. Taken together, these results suggest that the dendritic translation of αCaMK II transcripts is differentially increased by transient blockade of elongation and that phosphorylation of eEF2 represents a rapid, local and selective mechanism that may increase αCaMK II synthesis in response to NMDAR activation at developing synaptic contacts.

RESULTS To study the effects of NMDAR activation on synaptic protein synthesis, we used synaptoneurosomes14 (fractions enriched in nature neuroscience • volume 3 no 3 • march 2000

isolated, functional pre- and postsynaptic elements) prepared from the superficial visual layers of the developing rat sSC8. By postnatal day (P) 13, the majority of synapses are formed by retinal ganglion cell axons on sSC neurons15. This property makes the sSC well suited for the preparation of relatively homogeneous synaptic fractions for studies of NMDAR-linked signaling pathways during this period of synaptogenesis. The effect of NMDAR activation on protein synthesis in sSC synaptoneurosomes was examined using a 30-second pulse of 10 µM glutamate and 50 µM NMDA (referred to as NMDAR stimulation)16. NMDAR stimulation was terminated by the addition of AP-5 to a concentration of 120 µM. For each interval, we incubated an additional set of samples in 120 µM AP-5 for 5 minutes before NMDAR stimulation and continuously thereafter to serve as the NMDAR-inactivated controls. Protein synthesis in these AP-5 controls showed no consistent variation and remained relatively constant throughout the incubation period. Newly synthesized proteins were pulse labeled with 35S-methionine at staggered intervals after NMDAR stimulation (Fig. 1a) using a pulse–chase protocol. One-minute pulses of 35S-methionine were followed by 10-minute chases in the excess non-radioactive methionine to allow complete synthesis of proteins labeled during the pulse period. All data on protein synthesis following NMDAR activation is presented as a percentage of synthesis measured in these matched AP-5 control samples. Overall protein synthesis was maximally decreased within five minutes of NMDAR stimulation (Fig. 1b). Protein synthesis levels subsequently increased above baseline and remained elevated from 15 minutes after stimulation until the end of the experiment. To monitor the effects of prolonged incubation and AP-5 exposure independent of any exposure to the NMDA-stimulation cocktail, some samples received either no treatment or AP-5 treatment for one hour. No changes in protein synthesis were observed in these samples (Fig. 1b, boxes). 211


articles

a

NMDAR activation – AP-5

35S-methionine

Cold methionine Unstimulated AP-5 Alone

Sample sets

3 min. 5 min. 10 min. 15 min. 30 min. 45 min. 60 min.

Time between stimulation and 35S-methionine pulse

incorporation into protein (percent AP-5 control)

35S-methionine


b

35S-methionine incorporation (percent of value at incubation onset)

Total duration, 75 min

Interval between NMDA stimulation and 35S-methionine pulse

c

To resolve a large number of proteins, we used two-dimensional gel electrophoresis. Under basal conditions, only a few proteins were 35S-methionine labeled (Fig. 1c, panel 1). During the period of maximal inhibition of protein synthesis, incorporation of 35S-methionine into these proteins was qualitatively reduced (Fig. 1c, panel 2). A larger array of proteins in samples were 35Smethionine labeled following 60 minutes of NMDAR stimulation. Although we cannot rule out the possibility that posttranslational modifications account for some of the observed changes in distribution of spots, it seems probable that these changes were primarily due to changes in synthesis of individual proteins for two reasons. First, the two-dimensional gel electrophoresis analysis correlates well with our measures of NMDARstimulated changes in overall protein synthesis. Second, we find no evidence of large-scale changes in protein phosphorylation that could account for dramatic shifts in protein migration with NMDAR stimulation16. We conclude that, even though each of the 35S-methionine-labeled spots observed 60 minutes after NMDAR stimulation does not necessarily reflect new translation, 35S-methionine incorporation into postsynaptic protein shows a complex response following brief NMDAR activation. The mRNA for αCaMK II is present at high concentrations in dendrites17. We therefore examined the effect of NMDAR acti212

Fig. 1. NMDAR activation dynamically regulates protein synthesis in synaptic preparations. (a) Schematic description of the pulse-labeling procedure used to label newly synthesized protein in synaptoneurosomes. All samples (1–7) were incubated for the same total time. At the designated times, samples received NMDAR activation (top) or AP-5 for five min before NMDAR activation (bottom). At the designated times after NMDAR stimulation, both pairs of samples were pulse-labeled with 35S-methionine for one min (black) followed by a ten-min chase with nonradioactive methionine (gray) For 60 minutes preceding the 35S-methionine pulse in control experiments, samples either were untreated or were treated with AP-5. (b) NMDAR activation resulted in a rapid decrease in synaptic protein synthesis followed by a prolonged increase in overall protein synthesis. Data are means and standard errors from five independent determinations. Filled and open boxes represent the means and standard errors for control for either prolonged AP-5 exposure (n = 7) or unstimulated incubation (n = 9), respectively. Data for these samples are expressed as a percentage of 35Smethionine incorporation into freshly prepared samples. (c) Two-dimensional gels from samples at different times after NMDAR activation. The numbers in (b) correspond to the panel numbers (c). Panel 1, unstimulated; panel 2, 5 min NMDAR stimulation; panel 3, 60 min NMDAR stimulation. Examples of proteins whose synthesis was relatively unaffected by NMDAR stimulation are marked with rectangles. Circles mark examples of proteins whose synthesis was dramatically upregulated by NMDAR activation. These data are representative of two independent determinations for each time point. The isoelectric focusing range was between pI 4 (left) and 7 (right). Molecular mass standards are 205 kDa, 116 kDa, 70 kDa, 43 kDa, 36 kDa, 18 kDa and 7.5 kDa.

vation on αCaMK II synthesis. The level of newly synthesized αCaMK II was measured by densitometry after immunoprecipitation and subsequent gel electrophoresis. We detected two radiolabeled bands corresponding to molecular weights of 50 and 48 kDa. In a western blot, the 50 kDa protein comigrated with the αCaMK II protein detected with a different monoclonal antibody (Fig. 2b, inset). The other band may correspond to a midbrain-specific αCaMK II isoform 18. An analysis of variance showed that NMDAR stimulation produced statistically significant changes in αCaMK II synthesis (p < 0.05, F6 = 15.627). In contrast to overall protein synthesis levels, 3 minutes after NMDAR stimulation αCaMK II synthesis was increased (p < 0.05 on a planned post-hoc comparison). Following this rapid increase, 15 minutes after the cessation of NMDAR activation, αCaMK II synthesis was reduced to 40% of the AP-5 control level and subsequently showed a longer-latency increase that lasted for the duration of the experiment (Fig. 2a). Western blots of material from samples taken five minutes after stimulation of NMDARs (sample set #3 in Fig. 1a) and immunoprecipitated 16.5 minutes after NMDAR stimulation confirmed an increase in total synaptic αCaMK II levels (Fig. 2b). Thus, the synthesis as well as the overall expression level of αCaMK II in sSC synaptoneurosomes rapidly increased in response to brief NMDAR stimulation. Both nature neuroscience • volume 3 no 3 • march 2000


articles

a

35S-methionine

No stimulation

Interval between NMDA stimulation and 35S-methionine pulse

d

incorporation into GluR2/3 precipitated protein (percent AP-5 control)

c

35S-methionine


Auto Blot

Optical density

incorporation into αCaMKII protein (percent AP-5 control)

b

No stimulation

5 min after stimulation

the rapid increase in 35S-methionine-labeled αCaMK II (Fig. 2a, three minutes) and the increase in total αCaMK II protein measured 16.5 minutes after NMDAR stimulation (Fig. 2b) are consistent with the interpretation that synaptic activation of NMDARs increases αCaMK II levels via local protein synthesis. To assess the selectivity of NMDAR-induced protein translation, we immunoprecipitated several other synaptic proteins from 35S-methionine-labeled synaptoneurosomes, including GluR2/3 (Fig. 2c), NR1, calcineurin, spinophilin and eEF2 kinase (data not shown). None of these proteins showed any NMDAR-dependent increase in 35S incorporation. At P11, the retinocollicular map within the developing rat sSC completes its refinement19,20 and NMDA receptor currents are downregulated en masse 21 . To determine if changes in NMDAR-stimulated αCaMK II synthesis correlated with in vivo alterations in synaptic organization and NMDAR function, we analyzed sSC preparations from both P8 and P13 rats. NMDARmediated changes in αCaMK II synthesis showed the same temporal pattern at both ages. However, studies suggest that αCaΜΚ ΙΙ associates with other key synaptic proteins such as actin22 and the 2B subunit of the NMDAR23,24. Non-denaturing immunoprecipitation five minutes after NMDAR stimulation revealed that αCaMK II complexes in the developing sSC contain multiple newly synthesized proteins. Furthermore, a consisnature neuroscience • volume 3 no 3 • march 2000

16.5 min after stimulation

Fig. 2. Alpha CaMK II synthesis is increased by NMDAR activation. (a) We observed a significant increase in 35S-methionine incorporation into αCaMK II 3 min after termination of NMDAR stimulation (n = 5, p < 0.05, post-hoc planned comparison). A second phase of αCaMK II synthesis was observed between 30 and 60 min after NMDAR activation (n = 3). The numbers correspond to autoradiographs shown in the inset (1, unstimulated; 2, three minutes after NMDAR stimulation; 3, five min after NMDAR stimulation). (b) Steady-state αCaMK II protein levels were monitored from the 5-min stimulated samples. Including times for pulse-chase, these samples were collected 16.5 min after NMDAR stimulation. Overall levels of αCaMK II protein were significantly increased by the treatment (n = 3, p < 0.05). Inset shows the electrophoretic mobility of radiolabeled proteins (Auto) compared to αCaMK II protein detected with a different antibody via western blot (Blot). The slower migrating band from the autoradiograph corresponds to the αCaMK II protein, the faster migrating band corresponds to the band marked in (d) with an arrowhead. (c) Immunoprecipitation using an antibody directed against GluR2/3 revealed no NMDA stimulation-induced increase in 35S-methionine incorporation within five min of stimulation. Representative autoradiographs from either unstimulated (above, left) or NMDAR stimulated (right) samples are shown. (d) CaMK II immunoprecipitates several additional 35Smethionine labeled proteins (αCaMK II, white arrow). These gels are representative of three independent determinations of samples labeled five min after NMDAR stimulation ended. Molecular mass standards are as in Fig. 1.

tent qualitative difference in the composition of these complexes was observed between P8 and P13 synaptoneurosomes (Fig. 2d). Newly synthesized proteins from P8 sSC synaptoneurosomes that co-immunoprecipitated with CaMK II included two proteins with masses of approximately 25 and 30 kDa (arrows on left). Immuncomplexes derived from P13 synaptoneurosomes contained a different set of labeled proteins (arrows on right), including proteins of 10, 15, 35 and 42 kDa. NMDAR activation in the developing tadpole retinotectal projection induces an increase in synaptic phosphorylation of eEF2 that is localized to the immediate postsynaptic cytoplasm 6. Because eEF2 phosphorylation is both mediated by a Ca2+-dependent kinase and is associated with decreased protein synthesis via inhibition of peptide elongation10–13, it is possible that NMDARinduced eEF2 phosphorylation in sSC synaptoneurosomes could account for the observed rapid depression of overall synaptic protein synthesis. NMDAR activation of sSC synaptoneurosomes resulted in a fivefold increase in phospho-eEF2 levels within one minute of stimulation. Phospho-eEF2 levels remained elevated at least threefold for 5 minutes and then decreased, reaching baseline levels after 15 minutes (Fig. 3a). Phospho-eEF2 levels were not increased by NMDAR stimulation in samples treated before and during NMDA stimulation with either AP-5 or EGTA (data not shown). Thus, the rapid NMDAR-induced depression of 213


articles

Fig. 3. Inhibition of synaptic translation elongation. (a) Quantitation of phospho-eEF2 levels from NMDA-stimulated synaptoneurosomes. Immediately after 30 s of NMDAR activation, phospho-eEF2 levels were significantly increased (t-test, p < 0.05). Phospho-eEF2 levels returned to baseline 15 min after the addition of AP-5 (n = 5, mean ± s.d.). The hatched bar indicates the interval during which we observed opposing effects of NMDA stimulation on total protein and αCaMK II synthesis (compare Figs. 1b and 2a). (b) The effect of cycloheximide on αCaMK II synthesis depended on both time and concentration. The potentiation of αCaMK II synthesis was maximal after 15 min of cycloheximide treatment at a concentration of 0.5 µg per ml. (c) After 15 min of treatment with 0.5 µg per ml of cycloheximide, αCaMK II synthesis was significantly increased, whereas overall protein synthesis was decreased to 60% of untreated levels. Higher concentrations of cycloheximide (5 µg per ml and 50 µg per ml) decreased overall protein synthesis to 20% and 5% of untreated levels, respectively. Treatment with 5 µg per ml of cycloheximide resulted in modest increases in αCaMK II synthesis (40% over baseline), whereas 50 µg per ml severely decreased αCaMK II synthesis (50% below baseline). Data are presented as means and standard deviations from three independent determinations.

eEF2 phosphorylation (fold increase over AP-5 control)

a

incorporation into αCaMKII protein (percent untreated)

0.5 µg/ml cycloheximide 5.0 µg/ml cycloheximide 50 µg/ml cycloheximide

Time after cyclohexamide application (min)

5.0 µg/ml

n ei

I

ot

KI

pr

M To t

Ca

al

I

n

50 µg/ml

ei

KI

ot

M

pr al

Ca

ot

ei

To t

To t

al

pr

M

KI

I

n

0.5 µg/ml

Ca

c

incorporation into protein 15 min after cycloheximide application (percent untreated)

35S-methionine

b

35S-methionine


Time after NMDA stimulus (min)

synaptic protein synthesis at developing sSC synapses correlates with calcium-dependent phosphorylation of eEF2. We next examined if a generalized reduction of polypeptide elongation via eEF2 phosphorylation could also account for the rapid and relatively selective increase in αCaMK II synthesis induced by NMDAR stimulation. Studies of mRNA competition indicate that the efficiency of translation of certain transcripts is increased by slowing elongation, whereas translation of the majority of transcripts is decreased25–28. For example, doses of cycloheximide that reduce overall protein synthesis by 50% potentiate the translation of several transcripts by as much as 60% in normal fibroblasts29. Thus, it seemed possible that transiently reducing 214

overall protein synthesis via eEF2 phosphorylation could account for the observed increase in αCaMK II synthesis if αCaMK II translation shared this paradoxical response to mild elongation inhibition. We therefore treated synaptoneurosomes with low doses of cycloheximide, an agent that reduces elongation rate via a mechanism that does not require eEF2 phosphorylation. After 15 minutes in the presence of cycloheximide (0.5 µg per ml), overall protein synthesis decreased by 60%, whereas αCaMK II synthesis increased by 150% over baseline (Fig. 3c). In other cell-based systems, this concentration of cycloheximide reduces elongation to rates roughly equivalent to those observed in the presence of phospho-eEF212 (A.C.N., unpublished observations). The effect of cycloheximide on αCaMK II synthesis was time-dependent, occurring at a latency that was longer than the NMDA-induced effect, probably because of the time necessary for the drug to diffuse across the synaptoneurosome membranes (Fig. 3b). Higher cycloheximide concentrations (5 and 50 µg per ml) reduced overall protein synthesis by 80% and 95%, respectively (Fig. 3c). Synthesis of αCaMK II was slightly increased by addition of cycloheximide at 5 µg per ml, but showed a 50% decrease with cycloheximide at 50 µg per ml (Fig. 3c). Our interpretation of these data is that αCaMK II transcripts exhibit an increase rather than a decrease in translation efficiency in response to mild inhibition of elongation. The relative resistance of αCaMK II synthesis (50% reduction) to levels of cycloheximide that reduced total protein synthesis in the synaptoneurosomes to 5% is also consistent with previous studies of proteins that respond paradoxically to blockade of translation elongation. In fibroblasts, doses of cycloheximide that reduce overall protein synthesis by 50% leave all of the remaining synthetic activity accounted for by the synthesis of just 7 proteins29. Here, αCaMK II synthesis may account for much of the residual 5% of control synthetic activity we observed following 50 µg per ml cycloheximide. Clearly, we cannot formally rule out alternative explanations for our 35S- methionine data. Our results, however, indicate significant similarities between the effects of NMDAR stimulation and low doses of cycloheximide on translation in synaptoneurosome fractions. Both treatments decrease total protein synthesis but increase αCaMK II synthesis, and for both of these treatments, the opposing effects on translation are coupled in time. Thus, these data suggest that eEF2 phosphorylation and the consequent decrease in elongation rate can account for the rapid increase in synaptic αCaMK II synthesis induced by NMDAR stimulation. nature neuroscience • volume 3 no 3 • march 2000


articles


DISCUSSION We have demonstrated that brief NMDAR activation produced rapid and dynamic changes in the synthesis of numerous proteins in synaptoneurosomes from the developing sSC. Shortly after NMDAR activation, both the synthesis and the absolute levels of αCaMK II protein were increased, whereas synthesis of most other proteins was severely depressed. Correlated with the increase in αCaMK II synthesis, we observed an increase in the Ca2+-dependent phosphorylation of eEF2, a response known to reduce elongation rate. Additionally, low doses of cycloheximide, which reduce elongation rate via an eEF2-independent mechanism, produce a similar, temporally coupled decrease in total protein and an increase in αCaMK II synthesis. Taken together, these observations suggest the following sequence of events. NMDAR-mediated Ca2+ influx into dendrites activates Ca2+-dependent eEF2 kinase, which then phosphorylates eEF2. This phosphorylation might slow the local rate of protein translation, and elongation, rather than initiation, would consequently become the rate-limiting step in protein synthesis. Such a shift should favor upregulation of translation of abundant but poorly initiated transcripts such as αCaMK II in dendrites. Other proteins are probably selectively translated, but their identities as well as the downstream implications of such signaling remain unexplored. We propose that this pathway constitutes a mechanistic link between the activation of a neurotransmitter receptor and the rapid and local control of protein production. A number of other studies suggest that dendritic protein synthesis may be regulated locally by activation of receptors on dendrites. In the hippocampus, dendritic protein synthesis is increased by simultaneous stimulation of afferents and activation of acetylcholine receptors30; isolated processes of Aplysia sensory neurons stimulated with serotonin respond by increasing protein synthesis levels1 and stimulation of metabotropic glutamate receptors in isolated synaptic contacts from neonatal rat brain increases protein synthesis31 as well as the synthesis of proteins from exogenous mRNA in isolated neurites32. In addition, local application of neurotrophins to isolated hippocampal dendrites produces an enhancement of synaptic transmission that is blocked by inhibitors of protein synthesis2. Earlier data also show that regulation of translation can be exerted on specific proteins. For example, synthesis of the fragile X protein is increased by activation of metabotropic glutamate receptors in synaptoneurosomes33, and several unidentified proteins are synthesized in response to depolarization of isolated synaptic fractions34. Rapid increases in dendritic αCaMK II levels have also been demonstrated after synaptic activation, both in vitro35,36 and in vivo37, suggesting that local synthesis of αCaMK II in dendrites could have an appreciable effect on the synaptic concentration of αCaMK II. Despite these compelling observations, the mechanism(s) involved in the regulation of dendritic protein synthesis remain unknown. Here we provide a plausible explanation for the speed and selectivity of responses in dendritic translation, particularly in relationship to NMDARs, Ca2+ and αCaMK II. However, there are probably additional mechanisms of synaptic translational control in neurons. For example, cytoplasmic polyadenylation of αCaMK II mRNA in rat visual cortex is increased by 30 minutes of light exposure after dark rearing38, and this polyadenylation is correlated with an increase in αCaMK II protein levels at synapses, suggesting that, as in oocytes39, polyadenylation increases translational efficiency. Interestingly, the time course observed with a dark rearing/light exposure protocol correlates well with the long-latency increase in αCaMK II we observed in isolated neonatal sSC synapses. nature neuroscience • volume 3 no 3 • march 2000

The implications of our observations for brain maturation remain to be explored. Both NMDAR as well as eEF2 kinase show pronounced developmental regulation, and changes in the function of either of these proteins would significantly affect the mechanism we propose. In addition, the targeting of specific mRNAs to dendrites may change during development. This idea is supported by the developmental change in the array of newly synthesized proteins that interact with αCaMK II. Thus, rapid and selective local regulation of dendritic protein synthesis may constitute an important and previously unrecognized aspect of neuronal differentiation.

METHODS Preparation of synaptoneurosomes. Synaptoneurosomes were prepared using a described method14 with modifications. Unless otherwise noted, postnatal day 13 rats were used. Rats were anesthetized with carbon dioxide and decapitated, and the superficial, retinorecipient layers of the SC were dissected as previously described8. Samples were then homogenized in ice-cold oxygenated buffer (118 mM NaCl, 4.7 mM KCl, 1.2 mM MgSO4, 2.5 mM CaCl2, 1.53 mM KH2PO4 212.7 mM glucose) supplemented with 0.002 µl per ml of Complete protease inhibitor cocktail (Boehringer-Mannheim, Indianapolis, Indiana), 0.04 units per ml of human placental RNase inhibitor (Ambion, Austin, Texas) and 200 µg per ml of chloramphenicol (Sigma). All subsequent steps were carried out at 4°C. Samples were passed through a series of nylon filters of descending pore size. The final pass was through a MLCWP 047 Millipore filter with a 10-µm pore size. Samples were then centrifuged for 15 min at 1,000 × g. The supernatant was discarded and the pellet resuspended to a final protein concentration of 0.5 mg per ml. A protocol with characterized specificity6,16 for NMDAR stimulation consisted of a 30-s exposure to a cocktail of 10 µM glutamate and 50 µM NMDA added from a concentrated stock solution. Synaptoneurosomes (500 µl, containing 0.5 mg per ml protein) were maintained at 37°C for at least 10 min before NMDAR stimulation. Stimulation was terminated by addition of AP-5 from a concentrated stock to a final concentration of 120 µM. Five min before NMDAR stimulation, negative control samples received AP-5. 35S-Methionine pulse–chase labeling. Synaptoneurosomes were first warmed to 37°C for 10 min. For each time point, two samples were prepared, one sample was treated with AP-5 for five min before NMDAR stimulation, and another sample received NMDAR stimulation followed by AP-5 treatment. At designated times after NMDAR stimulation, 50 µCi of 35S-methionine was added to each sample. After 1 min, nonradioactive methionine was added to a final concentration of 200 µM. Samples were incubated for an additional 10 min to allow completion of synthesis of proteins labeled during the pulse period.

Assessment of overall protein synthesis. To measure overall protein synthesis, synaptoneurosome samples were treated with an equal volume of ice-cold 10% trichloroacetic acid (TCA) for 1 h. Insoluble material was then pelleted by centrifugation and washed 3 times in ice-cold 5% TCA (for 1 h total). Pellets were washed 3 times with ice-cold methanol and allowed to dry at room temperature. Samples were solubilized in either 0.1N NaOH for scintillation counting or a buffer (8.0 M urea, 0.5% SDS and 0.4% 2-mercaptoethanol) for two-dimensional isoelectric focusing24. Gels were treated with Amplify (Amersham Life Sciences, Piscataway, New Jersey) before drying and exposure. Radioactive proteins were detected either using X-ray film (Kodak, Rochester, New York) or phosphorimager plates (Fuji, Chicago, Illinois). αCaMK II immunoprecipitation. CaMK II was immunoprecipitated under non-denaturing conditions as previously described40. Synaptoneurosome samples were labeled as described above before immunoprecipitation. Samples were pelleted and solubilized in cold immunoprecipitation buffer (20 mM Tris-HCl, pH 7.4, 5.0 mM EDTA, 1.0% Triton X-100, 153 mM NaCl, 20 mg per ml BSA, 0.03% sodium azide and Complete protease inhibitors; Boehringer Mannheim). Samples were mixed with either 1 µl of the beta sub215


articles


unit-specific antibody CBβ1 (Gibco, Rockville, Maryland) or IgG and incubated overnight at 4oC. Immuncomplexes were collected by absorption with 50 µl of a 1:1 slurry of protein A sepharose beads (Pharmacia, Piscataway, New Jersey) in immunoprecipitation buffer. Pellets were washed 3 times in wash buffer (PBS, 0.25% NP-40, 0.05% sodium azide) before addition of 4× sample buffer. Samples were boiled and centrifuged. Samples (5 µl) were removed for scintillation counting, and the remainder was loaded on 8%–18% gradient acrylamide gels. Gels were exposed as described above. Scintillation counting or densitometric scanning of phosphorimager plates generated comparable results. Western blots using a monoclonal antibody (6G9, Boehringer Mannheim) were used to detect αCaMK II protein. Measurement of phospho-eEF2 levels. At designated times after NMDAR activation, samples were centrifuged and the supernatant removed. Pellets were suspended in 4× sample buffer and applied directly to 8%–18% gradient acrylamide gels. Separated protein was transferred to nitrocellulose and phospho-eEF2 detected with the polyclonal antibody cc81 (refs. 6, 41) visualized using SuperSignal chemiluminescence detection (Pierce, Rockford, Illinois). Total levels of eEF2 were estimated by stripping the blot and reprobing it using an affinity-purified anti-eEF2 antibody that does not discriminate between phospho- and dephospho-eEF2. Signal was quantitated as previously described8 using NIH Image 1.60. Cycloheximide treatment. A concentrated stock of cycloheximide (50 mg per ml) was prepared in 0.2% DMSO. Synaptoneurosomes prepared from the sSC of P13 rat pups were treated with cycloheximide at final concentrations of 0.5 µg per ml, 5.0 µg per ml and 50.0 µg per ml. At designated times, samples were pulse-labeled as above for one min and chased with non-radioactive methionine for ten min. Samples were then subjected to either TCA precipitation or CaMK II immunoprecipitation. Data were analyzed as above.

ACKNOWLEDGEMENTS This work was supported by U.S. Public Health Service Grants EY 06039 to M.C.P and GM 50402 to A.C.N.

RECEIVED 11 JANUARY; ACCEPTED 24 JANUARY 2000 1. Martin, K. C. et al. Synapse-specific, long-term facilitation of Aplysia sensory to motor synapses: a function for local protein synthesis in memory storage. Cell 91, 927–938 (1997). 2. Kang, H. & Schuman, E. M. A requirement for local protein synthesis in neurotrophin-induced hippocampal plasticity. Science 273, 1402–1406 (1996). 3. Constantine-Paton, M., Cline, H. T. & Debski, E. A. Patterned activity, synaptic convergence and the NMDA receptor in developing visual pathways. Annu. Rev. Neurosci. 13, 129–154 (1990). 4. Steward, O. & Falk, P. M. Protein-synthetic machinery at postsynaptic sites during synaptogenesis: a quantitative study of the association between polyribosomes and developing synapses. J. Neurosci. 6, 412–423 (1986). 5. Steward, O. & Falk, P. M. Selective localization of polyribosomes beneath developing synapses: a quantitative analysis of the relationships between polyribosomes and developing synapses in the hippocampus and dentate gyrus. J. Comp. Neurol. 314, 545–557 (1991). 6. Scheetz, A. J., Nairn, A. C. & Constantine-Paton, M. N-methyl-D-aspartate receptor activation and visual activity induce elongation factor-2 phosphorylation in amphibia tecta: a role for N-methyl-D-aspartate receptors in controlling protein synthesis. Proc. Natl. Acad. Sci. USA 94, 14770–14775 (1997). 7. Simon, D. K., Prusky, G. T., O’Leary, D. D. M. & Constantine-Paton, M. NMDA receptor antagonists disrupt the formation of a mammalian neural map Proc. Natl. Acad. Sci. USA 89, 10593–10597 (1992). 8. Scheetz, A. J., Prusky, G. T. & Constantine-Paton, M. Chronic NMDA receptor blockade during retinotopic map formation decreases the CaM kinase II differentiation in rat superior colliculus. Eur. J. Neurosci. 8, 1322–1328 (1996). 9. Kelly, P. T. Calmodulin-dependent protein kinase II. Multifunctional roles in neuronal differentiation and synaptic plasticity. Mol. Neurobiol. 5, 153–177 (1991). 10. Nairn, A. C. & Palfrey, H. C. Identification of the major Mr 100,000 substrate for calmodulin-dependent protein kinase III in mammalian cells as elongation factor-2. J. Biol. Chem. 262, 17299–17303 (1987). 11. Nairn, A. C. & Palfrey, H. C. in Translational Control (eds. Hershey, J. W. B., Mathews, M. B. & Sonenberg, N.) 295–318 (Cold Spring Harbor Press, Plainview, New York, 1996). 12. Redpath, N. T. & Proud, C. G. The tumour promoter okadaic acid inhibits reticulocyte-lysate protein synthesis by increasing the net phosphorylation of elongation factor 2. Biochem. J. 262, 69–75 (1989).

216

13. Ryazanov, A. G., Shestakova, E. A. & Natapov, P. G. Phosphorylation of elongation factor 2 by EF-2 kinase affects rate of translation. Nature 334, 170–173 (1988). 14. Hollingsworth, E. B. et al. Biochemical characterization of a filtered synaptoneurosome preparation from guinea pig cerebral cortex: cyclic adenosine 3′:5′-monophosphate-generating systems, receptors and enzymes. J. Neurosci. 5, 2240–2253 (1985). 15. Lund, R. & Lund, J. Development of synaptic patterns in the superior colliculus of the rat. Brain Res. 42, 1–20 (1972). 16. Scheetz, A. J. & Constantine-Paton, M. NMDA receptor activation-responsive phosphoproteins in the developing tectum. J. Neurosci. 15, 1460–1469 (1996). 17. Burgin, K. E. et al. In situ hybridization histochemistry of Ca++/calmodulindependent protein kinase II in developing brain. J. Neurosci 10, 1788–1798 (1990). 18. Brocke, L., Srinivasan, M. & Schulman, H. Developmental and regional expression of multifunctional Ca2+/calmodulin-dependent protein kinase in rat brain. J. Neurosci 15, 6797–6808 (1995). 19. Simon, D. K. & O’Leary, D. D. M. Limited topographic specificity in the targeting and branching of mammalian retinal axons. Dev. Biol. 137, 125–134 (1990). 20. Simon, D. K. & O’Leary, D. D. M. Development of topographic order in the mammalian retinocollicular projection. J. Neurosci 12, 1212–1232 (1992). 21. Shi, J., Aamodt, S. M. & Constantine-Paton, M. Temporal correlations between functional and molecular changes in NMDA receptors and GABA neurotransmission in the superior colliculus. J. Neurosci. 17, 6264–6276 (1997). 22. Shen, K. & Meyer, T. Dynamic control of CaMKII translocation and localization in hippocampal neurons by NMDA receptor stimulation. Science 284, 162–166 (1999). 23. Strack, S. & Colbran, R. J. Autophosphorylation-dependent targeting of calcium/ calmodulin- dependent protein kinase II by the NR2B subunit of the N-methylD- aspartate receptor. J. Biol. Chem 273, 20689–20692 (1998). 24. Leonard, A. S., Lim, I. A., Hemsworth, D. E., Horne, M. C. & Hell, J. W. Calcium/calmodulin-dependent protein kinase II is associated with the N- methyl-D-aspartate receptor Proc. Natl. Acad. Sci. USA 96, 3239–3244 (1999). 25. Brendler, T., Godefroy-Colburn, T., Carhill, R. D. & Thach, R. E. The role of mRNA competition in regulating translation II. Development of a quantitative in vitro assay. J. Biol. Chem. 256, 11747–11754 (1981). 26. Brendler, T., Godefroy-Colburn, T., Yu, S. & Thach, R. E. The role of mRNA competition in regulating translation III. Comparison of in vitro and in vivo results. J. Biol. Chem. 256, 11755–11761 (1981). 27. Godefroy-Colburn, T. & Thach, R. E. The role of mRNA competition in regulating translation IV. Kinetic model. J. Biol. Chem. 256, 11762–11773 (1981). 28. Walden, W. E., Godefroy-Colburn, T. & Thach, R. E. The role of mRNA competition in regulating translation I. Demonstration of competition in vivo. J. Biol. Chem. 256, 11739–11746 (1981). 29. Walden, W. E. & Thach, R. E. Translational control of gene expression in a normal fibroblast. Characterization of a subclass of mRNAs with unusual kinetic properties. Biochemistry 25, 2033–2041 (1986). 30. Feig, S. & Lipton, P. Pairing the cholinergic agonist carbacol with patterned schaffer collateral stimulation initiates protein synthesis in hippocampal pyramidal cell dendrites via a muscarinic, NMDA-dependent mechanism. J. Neurosci. 13, 1010–1021 (1993). 31. Weiler, I. J. & Greenough, W. T. Metabotropic glutamate receptors trigger postsynaptic protein synthesis. Proc. Natl. Acad. Sci. USA 90, 7168–7171 (1993). 32. Crino, P. B. & Eberwine, J. Molecular characterization of the dendritic growth cone: regulated mRNA transport and local protein synthesis. Neuron 17, 1173–1187 (1996). 33. Weiler, I. J. et al. Fragile X mental retardation protein is translated near synapses in response to neurotransmitter activation. Proc. Natl. Acad. Sci. USA 94, 5395–5400 (1997). 34. Leski, M. L. & Steward, O. Protein synthesis within dendrites: ionic and neurotransmitter modulation of synthesis of particular polypeptides characterized by gel electrophoresis. Neurochem. Res. 21, 681–690 (1996). 35. Ouyang, Y., Rosenstein, A., Kreiman, G., Schuman, E. M. & Kennedy, M. B. Tetanic stimulation leads to increased accumulation of Ca2+/calmodulindependent protein kinase II via dendritic protein synthesis in hippocampal neurons. J. Neurosci. 19, 7823–7833 (1999). 36. Ouyang, Y., Kantor, D., Harris, K. M., Schuman, E. M. & Kennedy, M. B. Visualization of the distribution of autophosphorylated calcium/calmodulindependent protein kinase II after tetanic stimulation in the CA1 area of the hippocampus. J. Neurosci. 17, 5416–5427 (1997). 37. Steward, O. & Halpain, S. Lamina-specific synaptic activation causes domainspecific alterations in dendritic immunostaining for MAP2 and CAM kinase II. J. Neurosci. 19, 7834–7845 (1999). 38. Wu, L. et al. CEPB-mediated cytoplasmic polyadenylation and the regulation of experience-dependent translation of α-CaMKII mRNA at synapses. Neuron 21, 1129–1139 (1998). 39. McGrew, L. L. & Richter, J. D. Translational control by cytoplasmic polyadenylation during Xenopus oocyte maturation: characterization of cis and trans elements and regulation by cyclin/MPF. EMBO J. 9, 3743–3751 (1990). 40. Baitinger, C., Alderton, J., Poenie, M., Schulman, H. & Steinhardt, R. A. Multifunctional Ca2+/calmodulin-dependent protein kinase is necessary for nuclear envelope breakdown. J. Cell Biol. 111, 1763–1773 (1990). 41. Marin, P. et al. Glutamate-dependent phosphorylation of elongation factor-2 and inhibition of protein synthesis in neurons. J. Neurosci. 17, 3445–3454 (1997).



articles

Rho GTPases regulate distinct aspects of dendritic arbor growth in Xenopus central neurons in vivo Zheng Li1,2, Linda Van Aelst1 and Hollis T. Cline1,2 1

Cold Spring Harbor Laboratory, Beckman Bldg., 1 Bungtown Rd., Cold Spring Harbor, New York 11724, USA

2

Department of Neurobiology and Behavior, SUNY Stony Brook, New York 11794, USA


Correspondence should be addressed to H.T.C. ([email protected])

The development and structural plasticity of dendritic arbors are governed by several factors, including synaptic activity, neurotrophins and other growth-regulating molecules. The signal transduction pathways leading to dendritic structural changes are unknown, but likely include cytoskeleton regulatory components. To test whether GTPases regulate dendritic arbor development, we collected time-lapse images of single optic tectal neurons in albino Xenopus tadpoles expressing dominant negative or constitutively active forms of Rac, Cdc42 or RhoA. Analysis of images collected at twohour intervals over eight hours indicated that enhanced Rac activity selectively increased branch additions and retractions, as did Cdc42 to a lesser extent. Activation of endogenous RhoA decreased branch extension without affecting branch additions and retractions, whereas dominant-negative RhoA increased branch extension. Finally, we provide data suggesting that RhoA mediates the promotion of normal dendritic arbor development by NMDA receptor activation.

The structure of the dendritic arbor critically determines what synaptic inputs a neuron receives and how they are integrated1. For instance, a neuron within the visual system whose dendritic arbor has a large tangential spread can receive inputs from more visual afferents, leading to a larger receptive field over which information is processed. Similarly, neurons that extend their dendritic arbors into superficial laminae can receive inputs and process information from afferents in those laminae2. Consequently, factors that regulate development and plasticity of the dendritic arbor control both the neuron’s structure and function and may ultimately affect circuit properties. The molecular mechanisms that underlie development of the dendritic arbor are not clear yet. In vivo time-lapse imaging permits direct observation of dendritic arbor development. Timelapse images of optic tectal neurons collected at intervals ranging from minutes to days in living Xenopus tadpoles show that dendritic branches are very dynamic during arbor formation3–6. The dynamic processes include addition of new branches, retraction of branches, and selective extension or shortening of existing branches. These events can be observed and quantified by collecting repeated images over several hours3–6. The data suggest that the net growth of dendritic arbor occurs as a result of several distinct events: emergence of a new branch, selective maintenance of the new branch, and extension of branch length. Each of these events may be individually regulated. Furthermore, it is very likely that machinery controlling the actin cytoskeleton is involved, because cytoskeleton reorganization accompanies structural changes in cells. Members of the Rho family of small GTPases, Rac, Cdc42 and RhoA, are required components of signal transduction pathways through which extracellular signals cause morphological changes in various cell types7–9. Rho GTPases mediate these changes by nature neuroscience • volume 3 no 3 • march 2000

regulating the cytoskeleton. Investigations of their function have been aided by mutations that result in a constitutively active, GTP-bound form or a dominant-negative, GDP-bound form. Studies of neurite outgrowth from cultured neurons have given us the first insights into the role of the Rho family of GTPases in regulating neuronal process growth9–11. Because neurites in cell culture often fail to take on the characteristics of dendrites and axons, few such studies have been able to distinguish effects of Rho GTPases on axonal and dendritic outgrowth. Rac, Cdc42 and RhoA influence dendrite number in dissociated cortical neurons12. Given that neuronal arbor elaboration in vivo is governed by activity-dependent and activity-independent factors13, which may operate through GTPases14, we wanted to determine whether Rho GTPases regulate dendritic arbor formation in the live animal with the normal pattern of synaptic inputs and local environment. Our previous studies showed that NMDA receptor activity is required for the initial phase of dendritic arbor growth in tectal neurons5,6. A potential link between synaptic activity and the Rho GTPases remains to be defined. As opposed to cultured cells, studies in intact animals provide the opportunity to investigate dendritic and axonal development in their normal complex environment. In vivo experiments in transgenic flies, worms, Xenopus and mice all point to a crucial role of Rac and Cdc42 in axonal growth and target recognition15–19. The roles of the Rho GTPases in regulating dendritic arbor structural plasticity in vivo are relatively unexplored20. Although the dendrites of Purkinje cells in transgenic mice expressing constitutively active Rac in the cerebellum branch normally, dendritic spines are reduced in size and increased in number16. In Xenopus, retinal ganglion cell dendritic arbor elaboration is inhibited by expression of constitutively active RhoA and constitutively active Cdc42, but promoted by constitutively 217


articles


Fig. 1. Expression of Rho GTPases and their effect on the actin cytoskeleton. (a) Myc immunostaining shows the distribution of infected, myc-tagged, GTPaseexpressing neurons in a horizontal section through the midbrain of a stage 47 tadpole. Most cells near the tectal ventricle (those imaged in later experiments) are infected. (b) Propidium iodide staining of a section neighboring the one shown in (a). Optic tectal cells (TC) are stained and appear gray. The neuropil (NP) region is unstained. (c) Phalloidin staining (left), GTPase expression (middle) in infected neurons, detected by myc immunoreactivity for constitutively active Rac (RacV12), dominant-negative Rac (RacN17) and constitutively active Cdc42 (Cdc42V12) or EGFP for dominant-negative Cdc42 (Cdc42N17), and the overlay of the two images (right) in sections from animals infected at low titer. Constitutively active forms of Rac and Cdc42 specifically increase actin polymerization in GTPaseexpressing neurons. Scale bars, 50 µm.

a

c

active Rac18. In transgenic Drosophila, constitutively active and dominant-negative Rac both block axonal growth of sensory neurons, without affecting dendrites, whereas constitutively active Cdc42 inhibits both axonal and dendritic growth15. As in the cell culture studies, some of the phenotypic outcomes resulting from expression of Rac and Cdc42 vary between the different studies, likely because of differences in neuron type15,17. Nevertheless, these studies provide growing evidence that Rac and Cdc42 can regulate dendritic arbor morphology. These studies do not indicate the potential function of Rho GTPases in dendritic arbor development, nor do they reveal how activity of different GTPases might cooperate during dendrite elaboration. To address these open issues, we investigated whether Rho GTPases regulate branch dynamics and dendritic arbor growth in optic tectal neurons in live Xenopus tadpoles, by collecting in vivo time-lapse images of single DiIlabeled tectal neurons. We used vaccinia virus-mediated gene transfer to express constitutively active and dominant-negative forms of RhoA, Rac and Cdc42 in tectal cells. We found that the three Rho GTPases have distinct effects in dendritic arbor development: Rac and Cdc42 regulate dynamic branch additions and retractions, whereas RhoA regulates elongation of existing branches. These results support the idea that dendritic arbor growth occurs through a multi-step process in which Rac and Cdc42 regulate the addition of short branches. RhoA regulates the selective extension of a subset of the added branches.

RESULTS Effects of Rho family GTPases on the actin cytoskeleton Tectal neurons were infected by ventricular injection of recombinant vaccinia viruses expressing constitutively active and dominant-negative forms of Rho family GTPases. The constitutively active mutants we used were RacV12 and Cdc42V12, and the dominant-negative mutants were RacN17, Cdc42N17 and RhoN19. The GTPases were also tagged with myc or enhanced green fluorescent protein (EGFP; Methods). The myc-tagged GTPases were expressed as IRES (internal ribosome entry site)-EGFP constructs that allowed us to identify infected regions of tectum for DiI labeling. Cryostat sections of infected animals immunostained for the 218

b

myc-tagged GTPases showed that after injection of high-titer virus, most tectal cells near the brain ventricle were infected (Fig. 1a). Neurons in this region were DiI labeled for imaging experiments described below. Expression of foreign protein was detected starting six hours after injection and was maintained throughout the experiment. Retinal ganglion cells are not infected when virus is injected into the ventricle, so retinal axons do not express foreign protein (Fig. 1), consistent with our previous data21. To test whether virally expressed Rho GTPases regulate the actin cytoskeleton in infected tectal neurons, we studied the distribution of polymerized actin using double labeling with phalloidin to visualize polymerized actin and either anti-myc antibody or EGFP to visualize Rho GTPase-expressing cells. The animals were infected with low-titer (106 pfu) virus to infect only a few tectal cells per animal, so that we could detect the actin filaments in individual GTPase-expressing neurons. In both uninfected animals and those infected with vaccinia virus expressing β-gal, filamentous actin (F-actin) was enriched in the neuropil region, with relatively little F-actin found in the cell body region of the tectum (Fig. 1b and c). Actin polymerization was increased dramatically in neurons expressing constitutively active Rac and constitutively active Cdc42, as indicated by strong phalloidin staining in infected cells (Fig. 1c). In neurons expressing dominant-negative Rac, dominant-negative Cdc42 or dominant-negative RhoA, the F-actin staining pattern was similar to that observed in control nature neuroscience • volume 3 no 3 • march 2000


articles

a

b

Activated RhoA inhibits dendritic branch extension Expression of dominant-negative RhoA significantly increased arbor growth rate compared to control cells (p < 0.05, Fig. 2; growth rate in control cells, 213.7 ± 18.8 µm per 20 h; in neurons from animals infected with dominant-negative RhoA vacnature neuroscience • volume 3 no 3 • march 2000

d TDBL growth (µm per 12 h)

cells. F-actin staining was also indistinguishable in lysophosphatidic acid (LPA)-treated animals and control animals, consistent with reports that neuronal cells do not form stress fibers after LPA exposure22. These experiments confirm viral expression of the GTPases. They further show that constitutively active Rac and constitutively active Cdc42 caused changes in actin cytoskeleton comparable to those reported previously7,9. One possible scenario is that these actin cytoskeleton alterations affect dendritic arbor growth. c In previous experiments, we collected time-lapse images of tectal cell dendrites at three-minute intervals over about an hour, at thirty-minute intervals over two hours, at two-hour intervals over eight hours and at daily intervals over five days3–6. These experiments demonstrated that short branches are continually added and retracted from the arbor and that arbor growth results from selective stabilization of a small fraction of the newly added branches and their subsequent elongation. To assess the effects of Rho GTPases on branch additions and retractions (which we call branch dynamics) as well as branch elongation, we chose an imaging protocol that permits quantitation of parameters relating to branch dynamics and net dendritic arbor growth. One day after injecting virus, we labeled single tectal cells with DiI and collected an initial image of the labeled tectal neurons two hours after DiI labeling. The following day, 12 hours after collecting the first image, we found the same single cell and imaged it at 2-hour intervals over the 8 hours from the 12-hour to 20-hour time points. To determine the effect of GTPase activity on overall dendritic arbor growth, we compared total dendritic branch length at the 0-hour and 20-hour timepoints. An increase in growth rate indicates an increase in branch elongation. To determine the effect of GTPase activity on dendritic arbor dynamics, we compared branch additions and retractions in sequential two-hour time points over eight hours. This imaging interval permits us to follow the fate of every branch we image within the arbor over time3–5. This protocol underestimates the rates of branch additions and retractions, because branches are continuously added and retracted within the two-hour intervals. Nevertheless, it does provide a relative measure of arbor dynamics between control and experimental conditions in the same neurons where we can also quantify arbor growth due to branch elongation (see Methods for details). TDBL growth (µm per 20 h)


Fig. 2. RhoA activity blocks dendritic arbor elaboration. (a) Drawings of 2 representative tectal cells from control and LPA-treated animals imaged over 12 h in vivo. (b) Drawings of 2 representative tectal cells from control and dominant-negative RhoA vaccinia virus-infected animals imaged over 20 h. (c) Change in total dendritic branch length (TDBL) over 20 h for neurons imaged from animals infected with various viruses. (d) Change in TDBL over 12 h for neurons from controls, LPAtreated animals and LPA-treated animals expressing dominantnegative RhoA. For each condition, 18–47 cells were analyzed. Scale bar, 50 µm, applies to (a) and (b). *p < 0.05; **p < 0.002.

cinia virus, 286.0 ± 33.2 µm per 20 h). We used LPA (10 µM in rearing solution) to activate endogenous RhoA because we were not able to generate a virus expressing constitutively active RhoA, most likely because the protein is toxic. LPA has been shown to activate RhoA specifically in neuronal cell lines11. Activation of RhoA by LPA treatment significantly reduced dendritic arbor growth rate from the control value of 160.8 ± 19.0 µm per 12 h to 67.4 ± 20.9 µm per 12 h (p < 0.002; Fig. 2). To assure that the effect of LPA on growth rate is due to activating RhoA, we showed that the growth-inhibiting effect of LPA could be counteracted by expression of dominant-negative RhoA. Cells from animals infected with dominant-negative RhoA vaccinia virus and treated with LPA had a growth rate of 151.6 ± 22.1 µm per 12 h, comparable to control neurons (p = 0.85; Fig. 2). These data indicate that LPA is acting through RhoA to control branch elongation in tectal neurons. In contrast to RhoA, Rac and Cdc42 did not alter dendritic growth rates (Fig. 2). Although RhoA is essential for dendritic branch extension, data shown below indicate that neither LPA nor dominant-negative RhoA changed branch number and branch dynamics (Fig. 3). Together, these data suggest that RhoA regulates the growth of the dendritic arbor by affecting the elongation of pre-existing branches. Rac regulates arbor dynamics To study the effects of Rho family GTPases on dendritic arbor dynamics, we imaged the same neuron every two hours for a total of eight hours. During dendritic arbor formation, the structure 219


articles

b

c

Total branch additions

Total branch retractions


a

Fig. 3. Rac activity promotes dendritic branch dynamics. (a) Drawings of 2 representative tectal cells from control, constitutively active Rac virus-infected and dominant-negative Rac virus-infected animals imaged over 20 h. Note the rapid changes in fine branch tips in cells expressing constitutively active Rac. (b, c) Total branch additions (b) and retractions (c) for neurons imaged every two hours over eight hours. For each condition, 18–47 cells were analyzed. Scale bar, 50 µm. *p < 0.05; **p < 0.002; ***p < 0.001.

of the arbor is very dynamic, characterized by continuous branch additions and retractions (Fig. 3). In animals infected with constitutively active Rac or dominant-negative Rac vaccinia virus, the dendritic arbor was more dynamic than in control neurons. As a result of these dynamics, the arbor structure was quite different from one time point to the next. Images of neurons infected with vaccinia virus expressing constitutively active Cdc42, dominant-negative Cdc42 and dominant-negative RhoA or neurons exposed to LPA did not show obvious changes in arbor dynamics during the eight-hour imaging period (see Fig. 3). To quantify whether an increase or decrease in RhoA, Rac or Cdc42 activity had significant effects on arbor dynamics, we counted the numbers of newly added branches and retracted branches that were imaged during the eight-hour period. All groups had a comparable number of branch tips and total dendritic branch length at the first imaging observation. Neurons from animals infected with constitutively active Rac vaccinia virus added significantly more dendritic branches than control cells 220

(60.9 ± 7.8, constitutively active Rac; 38.5 ± 3.0, control, p < 0.002; Fig. 3). Constitutively active Rac cells also retracted significantly more dendrites than controls (49.9 ± 7.1, constitutively active Rac; 29.4 ± 2.5, control; p < 0.001). Because constitutively active Rac enhanced both additions and retractions, the final number of branch tips was not significantly different from control cells. In contrast, dominant-negative Rac did not significantly alter rates of branch additions. Expression of dominantnegative Rac, however, did cause a significant increase in branch retractions compared to controls, but to a lesser extent than constitutively active Rac (39.8 ± 3.7, p < 0.05). Experiments were also done using constitutively active Cdc42, dominant-negative Cdc42, dominant-negative RhoA and LPA. Comparable analysis indicated that neither Cdc42 nor RhoA affected branch dynamics (Fig. 3). This analysis demonstrated that Rac activity affects rates of branch additions and retractions and suggested that the branches may be more transient. To test directly whether Rac, Cdc42 or RhoA alters the fate of dendritic branches, we designed an analysis to identify changes in the proportion of transient branches in an arbor. We divided all dendritic branches into four categories: stable branches, lost branches, new branches and transient branches (Fig. 4a). Stable branches are branches that are present throughout the imaging period. Lost branches are those that are present at the first image, but retracted over the eight-hour period. Any branches that were added after the first image and were still there at the last time point were categorized as new branches. The branches that were added after the first image and then retracted before the last image were categorized as transient branches. This analysis showed that among all the branches ever present during imaging, 34.8 ± 1.2% of them were transient in control cells. This number was significantly increased by the expression of constitutively active Rac, and to a lesser extent by constitutively active Cdc42 (Table 1; Fig. 4). Notably, dominantnegative Rac triggered an increase in transient branches (Fig. 4). In addition, dominant-negative Rac significantly decreased the relative number of new branches, that is, those that were added to the arbor and maintained to the end of the observation period. When we evaluated the fraction of stable branches, we noticed that only constitutively active Rac significantly affected this category, causing a 50% reduction of stable branches. No effects on arbor dynamics were observed when we expressed dominantnegative Cdc42 or dominant-negative RhoA or added LPA. Taken together, the analysis of arbor dynamics indicated that Rac and to a lesser extent Cdc42, but not RhoA are crucial in regulating arbor dynamics. To test whether constitutively active Rac increases arbor dynamics in the time frame of minutes, we imaged single labeled neurons every 3 minutes over periods up to 30 minutes. Six cells were imaged from animals infected with either EGFP vaccinia virus (control) or constitutively active Rac vaccinia virus. The rapid branch dynamics are most easily recognized in a time-lapse movie of the cells (see http://neurosci.nature.com/web_specials/). The movie also demonstrates that dendritic arbors of neurons from animals infected with constitutively active Rac vaccinia virus appear more dynamic. Branches within an arbor can have a variety of lifetimes, ranging from several minutes to several days3,4,23,24. The lifetimes of branches can be shifted to longer or shorter times during development4, by increased CaMKII activity3 or by blocking NMDA receptor activity6. Quantitation of the lifetimes of each branch in the arbors show that 67% of branches in constitutively active Rac neurons have lifetimes less than 9 minutes, whereas only 31% of branches in control neurons have similarly short lifetimes. The shift toward shorter nature neuroscience • volume 3 no 3 • march 2000


articles


branch lifetimes observed with expression of constitutively active Rac is consistent with the increase in the fraction of transient branches observed with the two-hour imaging protocol (Fig. 4). These data are consistent with the hypothesis that Rac affects cytoskeletal stability in neurons. Furthermore, they suggest that Rac affects arbor growth by affecting the rates of branch additions and retractions. GTPases affect arbor complexity The constitutively active Rac neurons shown in Fig. 3 appear to have more densely branched dendritic arbors than control neurons. We used Sholl analysis, in which the number of branches crossing concentric rings around the cell body are counted (see Methods) to analyze dendritic arbor complexity. Expression of constitutively active Rac significantly altered dendritic arbor complexity compared to controls (Fig. 5). The distal dendritic arbors of neurons from animals infected with constitutively active Rac vaccinia virus were more complex than in controls, whereas proximal dendritic arbors were significantly less complex than in controls. Sholl analysis also confirmed the impression from the drawings in Fig. 3 that decreased RhoA activity enhanced arbor complexity, whereas LPA treatment caused simpler dendritic arbors.

a Stable branch Lost branch New branch Transient branch

b

Stable branch New branch Lost branch Transient branch

Fig. 4. Rac and Cdc42 increase transient branches. (a) Schematic drawing of the types of branch categories analyzed (see text for details). (b) Plots of the fraction of stable, transient, new and lost branches in arbors from each of the designated treatments. *p < 0.05; **p < 0.01; ***p < 0.001.

RhoA is involved in NMDAR-mediated arbor growth We have previously shown that NMDA receptor activity is (p = 0.55). These data support a model in which NMDA receprequired for normal dendritic arbor elaboration in tectal neutor-mediated control of dendritic arbor elaboration operates rons5,6. Normal arbor elaboration is prevented by exposing tectal through a pathway that decreases RhoA activity. neurons to the NMDA receptor antagonist 3-amino-phosphonovaleric acid (APV) early during dendritic arbor development5, when the glutamatergic synaptic transmission from the retina is DISCUSSION mediated predominantly by the NMDA receptor25. By contrast, The structure of the neuronal dendritic arbor determines the the stability of dendritic arbors is less affected by APV in more inputs received by the neuron as well as its integrative properties1,2. mature neurons, when their retinotectal glutamatergic transChildren with mental retardation have severely reduced dendritic mission is mediated predominantly by the AMPA-type glutamate arbors, demonstrating a fundamental connection between neureceptor25. Here we report that increasing RhoA GTPase activity ronal structure and cognitive ability26. Consequently, mechanisms inhibits dendritic arbor development, whereas decreasing RhoA activity Table 1. Relative distribution of branches in each of the categories shown in Fig. 4. enhances dendritic growth. These data Stable branch New branch Lost branch Transient branch suggest that signals that decrease (Percent total) (Percent total) (Percent total) (Percent total) endogenous RhoA activity promote Control 7.3 ± 0.8 37.6 ± 1.7 20.3 ± 1.4 34.8 ± 1.3 dendritic branch elongation. An (n = 47) intriguing model is that NMDA receptor activity may promote dendritic 3.7 ± 1.1 34.1 ± 1.7 19.5 ± 1.8 42.7 ± 2.3 arbor development by decreasing RacV12 (n = 20) p < 0.013 p < 0.002 endogenous RhoA activity. To test this possibility, we determined whether expression of dominant-negative RhoA RacN17 5.7 ± 1.0 22.8 ± 7.8 21.8 ± 2.7 49.7 ± 5.5 could prevent the reduced dendritic (n = 18) p < 0.008 p < 0.001 growth rate observed as a result of blocking the NMDA receptor. As Cdc42V12 7.4 ± 1.4 28.2 ± 6.5 22.0 ± 1.9 42.4 ± 4.2 reported previously5, exposing animals (n = 22) p < 0.03 to 100 µM APV in their rearing solution inhibits dendritic arbor growth 8.4 ± 1.3 40.9 ± 2.3 16.5 ± 1.9 34.2 ± 2.1 compared to control neurons (APV, Cdc42N17 (n = 21) 160 ± 15 µm per 20 h; control, 214 ± 17 µm per 20 h; p < 0.05; Fig. 6). 9.3 ± 0.9 40.6 ± 2.2 18.5 ± 1.3 31.6 ± 2.2 Cells from animals infected with dom- LPA inant-negative RhoA and treated with (n = 27) APV grow at 256 ± 32.5 µm per 20 h, which is significantly faster than in RhoN19 10.4 ± 1.4 40.1 ± 2.4 16.4 ± 1.4 33.2 ± 2.2 APV-treated neurons (p < 0.005), and (n = 20) comparable to neurons infected with dominant-negative RhoA vaccinia virus Values for numbers of cells analyzed also apply to data in Figs. 2, 3 and 5. nature neuroscience • volume 3 no 3 • march 2000

221


articles


a

b

suggests that endogenous RhoA activity in developing tectal neurons is high and that dendritic arbor growth is actively promoted under conditions that inhibit RhoA activity. Because neither LPA nor dominant-negative RhoA affects rates of branch additions or retractions, these experiments indicate that RhoA selectively affects extension or retraction of existing branches. We previously showed that blocking the NMDA type of glutamate receptor c d decreases the rate of branch additions and decreases the extension of existing branches during early stages of dendritic arbor formation5,6. Later, as the neurons mature, their dendritic arbor structure becomes more stable. These mature neurons express calcium/calmodulin-dependent protein kinase (CaMKII), and enhanced CaMKII activity decreases rates of dendritic branch additions and retractions3. One intriguing scenario is that glutamatergic synaptic activity may promote dendritic arbor elaboration by decreasFig. 5. Rac and RhoA affect dendritic arbor complexity. Sholl analysis of constitutively active Rac (a), ing endogenous RhoA activity in dendominant-negative Rac (b), dominant-negative Rho (c) and LPA-treated (d) neurons compared to drites of immature tectal cells. Indeed, we controls. Expression of constitutively active Rac significantly decreases the number of dendritic find that the decreased arbor growth rate branches close to the cell body, but significantly increases arbor complexity distal to the cell body. observed when NMDA receptors are Decreasing RhoA activity increases arbor complexity, whereas increasing RhoA activity decreases blocked is counteracted by expression of complexity. *p < 0.05; **p < 0.001; ***p < 0.0001. dominant-negative RhoA. These data suggest a mechanism in which RhoA activity and dendritic branch extension may be controlled locally by synaptic inputs. Furthermore, as neuronal structure matures and becomes that influence the development of neuronal structure are likely to more stable, glutamatergic synaptic inputs and CaMKII activity significantly affect brain function. Mutations in a Rho-GTPase may operate through Rac and Cdc42 to control structural plasactivating protein are found in patients with X-linked mental retarticity. The demonstration of a Ras GTPase-activating protein dation27, suggesting that GTPase signaling is required for the develthat is regulated by NMDA receptor and CaMKII activity28,29 opment of normal brain function. Here we have shown that different members of the Rho famcombined with evidence of crosstalk between Ras and Rac in ily of GTPases regulate different aspects of dendritic arbor elabcontrolling cell morphology30 further suggest that such a reguoration in the intact animal. By taking sequential images of latory pathway may exist. individual dendritic arbors with the confocal microscope, we found that Rac, and to a lesser extent Cdc42, regulate branch Rac regulates dendritic arbor dynamics dynamics. RhoA specifically controls branch elongation, withData on the function of Rac in regulating dendritic arbor strucout directly affecting branch dynamics. These data suggest that ture in the intact animal is limited16,18,31. In Drosophila and mice, dendritic arbor elaboration occurs as a result of Rac-mediated Rac does not affect the overall development of dendritic morbranch dynamics followed by RhoA-mediated branch extension. phology15,16, whereas in Xenopus retinal cells, Rac does affect denTogether the activities of the GTPases contribute to the net dritic arbor outgrowth18. Our data show that Rac is involved in growth of dendritic arbors in the intact animal. Because many the formation and turnover of dendrites, but not dendritic branch tectal neurons are infected and express ectopic GTPases in these extension. At a single time point, the total number and length of experiments, an interesting possibility, which we cannot exclude, dendrites in animals expressing constitutively active or domiis that the observed effects of the GTPases are in part due to nant-negative Rac are comparable to the control values, but timeGTPase expression in cells other than the ones imaged. However, lapse imaging clearly demonstrates that constitutively active Rac Rho affects dendritic arbor development in a cell-autonomous increases branch additions and retractions. The final branch fashion in Drosophila20. numbers and total branch length are the same as control neurons because enhanced rates of branch additions are balanced by increased branch retractions. One conceivable interpretation for RhoA regulates dendritic branch extension the constitutively active Rac phenotype is that increased Rac activExposure to LPA, an activator of RhoA, severely impairs denity triggers the initiation of dendritic branches, but Rac activity dritic arbor elaboration, whereas expression of dominant-negaalone is not sufficient to maintain the newly formed branches. tive Rho enhances dendritic arbor growth, consistent with the Consequently, they are retracted. Rac activity is also required to findings in neuronal cell lines, where LPA causes neurite retracmaintain dendritic branches, because more dendrites are retracttion and inhibition of RhoA induces neurite outgrowth10,11. This 222



c ed when Rac activity is reduced by expression of dom- a inant-negative Rac. In neurons from animals infected with dominant-negative Rac vaccinia virus, the enhanced rates of branch retractions contribute to the increased fraction of transient branches in these arbors. These experiments indicate the increased Rac activity promotes branch additions. They further suggest that Rac activity is necessary but not sufficient for branch maintenance. In addition to regulating the rate of branch additions, Rac is also involved in shaping the architecture of the dendritic arbor. Sholl analysis indicates that enhanced Rac activity increases the overall complexity of the dendritic arbor. Constitutively active Rac increased the complexity of the distal dendritic arbor, d b whereas it decreased the complexity of the proximal dendritic tree. The elaboration of proximal and distal dendrites are differentially regulated2, possibly as a result of differences in lamina-specific input activity and the distributions of environmental signals such as adhesion molecules32 and neurotrophins33. Indeed, GTPase activity has been suggested to differentially regulate the elaboration of basal and apical dendrites in dissociated cortical neurons12. Because the Rho family GTPases are regulated by extracellular stimuli9, and also seem to mediate neuronal responses to substratebound guidance cues, including myelin and integrins34–36, normal dendritic arbor elaboration may Fig. 6. NMDA receptor-mediated arbor growth operates through RhoA. Drawings of reflect the complex composition of environmental cues two representative neurons from controls (a), APV-treated animals (b) and APVin intact animals. Consequently, in our experiments, treated animals expressing dominant-negative Rho (c). (d) Change in total dendritic introduction of constitutively active Rac may have had branch length (TDBL) for neurons from the designated groups. For each condition, different outcomes in distal and proximal dendrites 12–24 neurons were analyzed. Scale bar, 50 µm. *p < 0.05. because of local variations in endogenous GTPase activity within the arbor. Increased Cdc42 activity in tectal neurons increased mechanisms, with respect to the cytoskeleton. The cytoskeleton the proportion of transient dendrites. Active Cdc42 induces of transient dendritic branches may be entirely actin-based38 filopodia in growth cones37, which serve a sensory role in axon pathfinding. Cdc42 may serve a similar function to promote and their addition likely due to actin polymerization39. The filopodia in the dendritic arbor, but does not seem to regulate maintenance of a newly added branch may be due to the invaeither the stabilization or extension of dendritic branches. sion of the new branch by microtubules 39, as described for growth cones40. Finally branch extension may be due to assembly of microtubules. Although all three GTPases are reported to Interaction of GTPases with the cytoskeleton regulate the actin cytoskeleton, RhoA may also regulate tubuOur data suggest that the GTPases modify specific aspects of lin assembly41, supporting the idea that RhoA’s principal effect dendritic arbor morphology. Rac seems to govern the rates of additions and retractions of new branches without affecting subon dendritic arbor elaboration is through the extension or sequent regulatory events that determine whether the branch retraction of existing dendritic branches, which are enriched in will extend. The increased actin filaments we observed by phalmicrotubules38. In addition, GTPases can regulate cell–cell conloidin staining in constitutively active Rac and constitutively tacts through presentation and clustering of cadherins and inteactive Cdc42 cells suggest that these GTPases likely mediate their grins on the cell surface7. Evidence that cadherins can function effects on the dendritic arbor dynamics by affecting the actin in synapse stabilization42 suggests an intriguing connection cytoskeleton. Although constitutively active Cdc42 expression between GTPases and synaptogenesis. increased filamentous actin according to the phalloidin stainThe GTPases are positioned in the midst of several signal ing, it had only a modest effect on the parameters of dendritic transduction pathways, which govern cell shape and polarity7,9. dynamics assayed here. It is possible that Cdc42 affects differThe activity of each GTPase may be altered downstream of cell ent aspects of cytoskeletal structure in neurons that were not surface receptors, including growth factors, cytokines and neuassessed in this study. Neither LPA nor dominant-negative RhoA rotransmitters7,14,37. We provide evidence that the NMDA-type had a significant effect on branch dynamics, although they clearglutamate receptor may be upstream of the RhoA GTPase in regly regulated branch shortening and extension, respectively. This ulating dendritic arbor elaboration. Indeed, it seems that NMDA suggests that RhoA activity influences the extension or retracreceptor activity decreases RhoA activity and thereby increases tion of existing branches but does not govern the emergence of branch elongation. This observation is surprisingly complenew branches. Addition and retraction of short branches, mainmentary to a report that activation of the p75 neurotrophin tenance of branches and their extension are distinct cell biologreceptor can increase retinal axon elongation by decreasing ical events that are likely controlled by different regulatory endogenous RhoA activity14. In addition, each of the GTPases TDBL growth (µm per 20 h)


articles


223


articles

affects the cytoskeleton through downstream effectors. It would be of interest to identify the interacting partners through which the GTPases regulate dendritic arbor growth. Potential downstream candidates for RhoA are the serine/threonine kinase, ROCK, and mDia, which regulate the formation of different types of actin fibers43. Rac may act through LIM kinase and cofilin44,45. Filopodia formation induced by Cdc42 seems to depend on N-WASP, a ubiquitously expressed Cdc42 binding protein46. Finally, Rac and Cdc42 share common effectors, including the serine/threonine kinase PAK, which is involved in the morphological changes mediated by Rac and Cdc42 (refs. 9, 47).


METHODS Construction of recombinant vaccinia virus. Human RacV12, RacN17 and Cdc42V12 were myc tagged at the amino (N) terminal. Myc-RacV12, myc-RacN17 and myc-Cdc42V12 were cloned into the Sal I/Spe I site upstream of IRES-EGFP in the pBluescript vector. The myc-GTPaseIRES-EGFP constructs were cut out from pBluescript vector and cloned into the vaccinia virus vector pSC65 downstream of a strong synthetic early/late vaccinia virus promoter. The cDNA encoding EGFP-Cdc42N17 and EGFP-RhoN19 N-terminal fusion proteins were cloned into the Sal I /Sma I sites of pSC65 downstream of the strong synthetic early/late vaccinia virus promoter. The EGFP fusion proteins are active in vitro in a variety of cell lines assayed for cell morphology and cell adhesion (L. Van Aelst, Cold Spring Harbor Laboratory, Cold Spring Harbor, New York, personal communication). A virus expressing only EGFP driven by the early/late promotor was used as a control for the effects of viral infection. All viruses also express β-galactosidase (β-gal) behind a weaker p7.5 viral promotor, which is used for plaque selection of recombinant viruses48. Constructs were confirmed by sequencing. Recombinant vaccinia viruses, obtained by homologous recombination of pSC65 and wild-type vaccinia virus as reported3, were purified and titered before use48. High-titer virus (over 107 plaque-forming units, pfu) was used to infect animals in imaging experiments. Viral infection. Albino Xenopus laevis tadpoles were obtained by mating induced by human chorionic gonadotropin injections. Purified virus (100–150 nl mixed with 0.1% fast green) was injected into the tectal ventricle of stage 46 tadpoles anesthetized with 0.02% 3-aminobenzoic acid ethyl ester (MS222). Immunostaining. Animals were fixed in 4% paraformaldehyde in 0.1 M phosphate buffer (PB, pH7.4) overnight at 4oC, rinsed in PB and cryoprotected in 30% sucrose before brains were cut into 20-µm cryostat sections. For myc immunostaining, sections were preincubated with blocking solution containing 5% goat serum and 0.3% Triton X-100 in PB for 1 h, followed by an overnight incubation at 4oC in anti-myc antibody (Calbiochem), diluted 1:100 in blocking solution. After rinsing, sections were incubated in cy5-goat anti-mouse secondary antibody (diluted 1:100 in blocking solution; Jackson Immunoresearch Laboratories, West Grove) for 30 minutes. We did not observe differing levels of myc immunoreactivity in neurons, suggesting that expression levels of the GTPases did not vary over a detectable range. Rhodamine-phalloidin (Molecular Probes Eugene, Oregon; final dilution 1:800) was added to the incubation solution together with secondary antibody for double labeling. Image acquisition. Single tectal neurons were labeled by iontophoresis of DiI (0.02% 1,1′-dioctadecyl-3,3,3′3′-tetramethylindocarbocyanine perchlorate, Molecular Probes) in ethanol. Positive current (1–10 nA) was used in 3–10 pulses of 200-ms duration. In animals infected with EGFP fusion proteins or IRES-EGFP constructs, DiIlabeled neurons were located within an infected region of the tectum; however, it was not always possible to verify that each DiI-labeled neuron was infected. We selected animals with single brightly labeled neurons two hours after DiI labeling. Images were collected at 2-µm steps through the entire z dimension of labeled neurons with a Noran Instruments XL laser scanning confocal attachment mounted on an upright Nikon Optiphot through a 40× Nikon oil immersion lens (1.30 NA). 224

Each optical section was an average of 8–16 frames. Animals were anesthetized with 0.02% MS222 during DiI labeling, screening and imaging. Animals recovered from anesthesia between imaging sessions, except for the three-minute imaging experiment, where animals were anesthetized throughout the imaging protocol. In experiments using LPA to activate RhoA or APV to block NMDA receptors, the drug was added to the rearing solution immediately after the first image was collected. Image analysis. Dendritic arbors were reconstructed by tracing the portion of the neuron in each optical section onto an acetate sheet until the entire neuron was drawn. This method provides a more detailed representation of the morphology than the three-dimensional reconstruction generated by computer, because fine processes visible in the optical sections are lost in the computer-generated reconstruction. Total dendritic branch length was measured from scanned drawings of cells with NIH Image 1.61. The number of branch tips was counted manually. To analyze the arbor dynamics, drawings of cells from sequential time points were superimposed to identify added and retracted branches. The magnitude of arbor dynamics determined in this and previous studies3,4, in terms of numbers of branches added or retracted over a two-hour period, may be underestimated by five- to tenfold24,49, because unobserved branches are both added and retracted during the two-hour intervals. The shortest interval over which we can collect images is three minutes because this is about the time it takes to collect and save a single z series through the optic tectum. Observations collected at such frequent time points over our 20-hour imaging period would provide a more accurate value of branch additions and retractions; however, neither the DiI-labeled neurons nor the animals can survive such prolonged imaging session or sustained anesthesia49. Branch additions, branch retractions and the change in total dendritic branch length in uninfected control animals and animals infected with EGFP vaccinia virus were comparable, consistent with previous observations that vaccinia virus does not affect the development of tectal cell morphology3. We therefore pooled these two groups of control cells together. Sholl analysis was done with Object-Image software (NIH Image). A scanned drawing of the neuron was overlaid on a series of concentric circles, spaced every 5 µm with the cell body in the center. The number of dendritic branches crossing each concentric circle was marked on the composite image and counted. Statistical analysis was done with two-tailed t-test. Note: Time-lapse movies can be found on the Nature Neuroscience web site (http://neurosci.nature.com/web_specials/).

ACKNOWLEDGEMENTS We thank Wun Chey Sin for providing the EGFP-Cdc42N17 and EGFP-RhoN19 viruses, Kim Bronson for technical assistance and Neil Mahapatra for help with the Sholl analysis. Support for the work was provided by the NIH (H.T.C., L.V.A.) and the Eppley Foundation (H.T.C.).

RECEIVED 14 DECEMBER 1999; ACCEPTED 27 JANUARY 2000 1. Stuart, G., Spruston, N. & Hausser, M. Dendrites (Oxford Univ. Press, Oxford, 1999). 2. Koester, S. E. & O’Leary, D. D. Development of projection neurons of the mammalian cerebral cortex. Prog. Brain Res. 102, 207–215 (1994). 3. Wu, G.-Y. & Cline, H. T. Stabilization of dendritic arbor structure in vivo by CaMKII. Science 279, 222–226 (1998). 4. Wu, G.-Y., Zou, D. J., Rajan, I. & Cline, H. T. Dendritic dynamics in vivo change during neuronal maturation. J. Neurosci. 19, 4472–4483 (1999). 5. 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). 6. 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, 2089–2096 (1999). 7. Hall, A. Rho GTPases and the actin cytoskeleton. Science 279, 509–514 (1998). 8. Narumiya, S. The small GTPase Rho: cellular functions and signal transduction. J. Biochem. (Tokyo) 120, 215–228 (1996).




articles

9. Van Aelst, L. & D’Souza-Schorey, C. Rho GTPases and signaling networks. Genes Dev. 11, 2295–2322 (1997). 10. Van Leeuwen, F. N. et al. The guanine nucleotide exchange factor Tiam1 affects neuronal morphology; opposing roles for the small GTPases Rac and Rho. J. Cell Biol. 139, 797–807 (1997). 11. Gebbink, M. F. et al. Identification of a novel, putative Rho-specific GDP/GTP exchange factor and a RhoA-binding protein: control of neuronal morphology. J. Cell Biol. 137, 1603–1613 (1997). 12. Threadgill, R., Bobb, K. & Ghosh, A. Regulation of dendritic growth and remodeling by Rho, Rac, and Cdc42. Neuron 19, 625–634 (1997). 13. Katz, L. C. & Shatz, C. J. Synaptic activity and the construction of cortical circuits. Science 274, 1132–1138 (1996). 14. Yamashita, T., Tucker, K. L. & Barde, Y.-A. Neurotrophin binding to the p75 receptor modulates Rho activity and axonal outgrowth. Neuron 24, 585–605 (1999). 15. Luo, L., Liao, Y. J., Jan, L. Y. & Jan, Y. N. Distinct morphogenetic functions of similar small GTPases: Drosophila Drac1 is involved in axonal outgrowth and myoblast fusion. Genes Dev. 8, 1787–1802 (1994). 16. Luo, L. et al. Differential effects of the Rac GTPase on Purkinje cell axons and dendritic trunks and spines. Nature 379, 837–840 (1996). 17. Kaufmann, N., Wills, Z. P. & Van Vactor, D. Drosophila Rac1 controls motor axon guidance. Development 125, 453–461 (1998). 18. Ruchhoeft, M. L. et al. The neuronal archicture of Xenopus retinal ganglion cells is sculpted by Rho-family GTPases in vivo. J. Neurosci. 19, 8454–8463 (1999). 19. Zipkin, I. D., Kindt, R. M. & Kenyon, C. J. Role of a new Rho family member in cell migration and axon guidance in C. elegans. Cell 90, 883–894 (1997). 20. Lee, T. et al. Essential roles of Drosophila RhoA in the regulation of neuroblast proliferation and dendritic but not axonal morphogenesis. Neuron (in press). 21. Wu, G.-Y., Zou, D.-J., Koothan, T. & Cline, H. T. Infection of frog neurons with vaccinia virus permits in vivo expression of foreign proteins. Neuron 14, 681–684 (1995). 22. Jalink, K. et al. Inhibition of lysophosphatidate- and thrombin-induced neurite retraction and neuronal cell rounding by ADP ribosylation of the small GTP-binding protein Rho. J. Cell Biol. 126, 801–810 (1994). 23. O’Rourke, N. A., Cline, H. T. & Fraser, S. E. Rapid remodeling of retinal arbors in the tectum with and without blockade of synaptic transmission. Neuron 12, 921–934 (1994). 24. Witte, S., Stier, H. & Cline, H. T. In vivo observations of timecourse and distribution of morphological dynamics in Xenopus retinotectal axon arbors. J. Neurobiol. 31, 219–234 (1996). 25. Wu, G.-Y., Malinow, R. & Cline, H. T. Maturation of a central glutamatergic synapse. Science 274, 972–976 (1996). 26. Purpura, D. P. Dendritic differentiation in human cerebral cortex: normal and aberrant developmental patterns. Adv. Neurol. 12, 91–134 (1975). 27. Billuart, P. et al. Oligophrenin-1 encodes a rhoGAP protein involved in Xlinked mental retardation. Nature 392, 923–926 (1998). 28. Chen, H. J., Rojas, S. M., Oguni, A. & Kennedy, M. B. A synaptic Ras-GTPase activating protein (p135 SynGAP) inhibited by CaM kinase II. Neuron 20, 895–904 (1998).


29. Kim, J. H., Liao, D., Lau, L.-F. & Huganir, R. L. SynGAP: a synaptic RasGAP that associates with the PSD-95/SAP90 protein family. Neuron 20, 683–691 (1998). 30. Ridley, A. J. et al. The small GTP-binding protein rac regulates growth factorinduced membrane ruffling. Cell 70, 401–410 (1992). 31. Luo, L., Liao, Y. J., Jan, L. Y. & Jan, Y. N. Distinct morphogenetic functions of similar small GTPases: Drosophila Drac1 is involved in axonal outgrowth and myoblast fusion. Genes Dev. 8, 1787–1802 (1994). 32. Miskevich, F., Zhu, Y., Ranscht, B. & Sanes, J. R. Expression of multiple cadherins and catenins in the chick optic tectum. Mol. Cell Neurosci. 4, 240–255 (1998). 33. McAllister, A. K., Lo, D. C. & Katz, L. C. Neurotrophins regulate dendritic growth in developing visual cortex. Neuron 15, 791–803 (1995). 34. Kuhn, T. B. et al. Myelin and collapsin-1 induce motor neuron growth cone collapse through different pathways: inhibition of collapse by opposing mutants of rac1. J. Neurosci. 19, 1965–1975 (1999). 35. Jin, Z. & Strittmatter, S. M. Rac1 mediates collapsin-1-induced growth cone collapse. J. Neurosci. 17, 6256–6263 (1997). 36. Kuhn, T. B., Brown, M. D. & Bamburg, J. R. Rac1-dependent actin filament organization in growth cones is necessary for beta1-integrin-mediated advance but not for growth on poly-D-lysine. J. Neurobiol. 37, 524–540 (1998). 37. Kozma, R., Sarner, S., Ahmed, S. & Lim, L. Rho family GTPases and neuronal growth cone remodeling: relationship between increased complexity induced by Cdc42Hs, Rac1, and acetylcholine and collapse induced by RhoA and lysophosphatidic acid. Mol. Cell Biol. 17, 1201–1211 (1997). 38. 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). 39. Ziv, N. E. & Smith, S. J. Evidence for a role of dendritic filopodia in synaptogenesis and spine formation. Neuron 17, 91–102 (1996). 40. Bentley, D. & O’Connor, T. P. Cytoskeletal events in growth cone steering. Curr. Opin. Neurobiol. 4, 43–48 (1994). 41. Hirose, M. et al. Molecular dissection of the Rho-associated protein kinase (p160ROCK)-regulated neurite remodeling in neuroblastoma N1E-115 cells. J. Cell Biol. 141, 1625–1636 (1998). 42. 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). 43. Maekawa, M. et al. Signaling from Rho to the actin cytoskeleton through protein kinases ROCK and LIM-kinase. Science 285, 895–898 (1999). 44. Arber, S. et al. Regulation of actin dynamics through phosphorylation of cofilin by LIM-kinase. Nature 393, 805–809 (1998). 45. Yang, N. et al. Cofilin phosphorylation by LIM-kinase 1 and its role in Racmediated actin reorganization. Nature 393, 809–812 (1998). 46. Miki, H., Sasaki, T., Takai, Y. & Takenawa, T. Induction of filopodium formation by a WASP-related actin-depolymerizing protein N-WASP. Nature 391, 93–96 (1998). 47. Obermeier, A. et al. PAK promotes morphological changes by acting upstream of Rac. EMBO J. 17, 4328–4339 (1998). 48. Mackett, M., Smith, G. L. & Moss, B. DNA Cloning: A Practical Approach (IRL Press, Oxford, 1985). 49. Cline, H. T. et al. in Imaging: A Laboratory Manual (eds. Yuste, R., Lanni, F. & Konnerth, A.) 13.1–13.12 (Cold Spring Harbor Laboratory Press, 1999).

225


articles

Anatomical and physiological evidence for D1 and D2 dopamine receptor colocalization in neostriatal neurons


Oleg Aizman1, Hjalmar Brismar1, Per Uhlén1, Eivor Zettergren1, Allan I. Levey2, Hans Forssberg1, Paul Greengard3 and Anita Aperia1 1

Department of Woman and Child Health, Karolinska Institutet, Astrid Lindgren Children’s Hospital, Q2:09, 171 76 Stockholm, Sweden

2

Department of Neurology, Emory University School of Medicine, Atlanta, Georgia, Georgia 30322, USA

3

Laboratory of Molecular and Cellular Neuroscience, The Rockefeller University, New York, New York 10021-6399, USA Correspondence should be addressed to A.A. ([email protected]) O.A. and H.B. contributed equally to this work.

Despite the importance of dopamine signaling, it remains unknown if the two major subclasses of dopamine receptors exist on the same or distinct populations of neurons. Here we used confocal microscopy to demonstrate that virtually all striatal neurons, both in vitro and in vivo, contained dopamine receptors of both classes. We also provide functional evidence for such colocalization: in essentially all neurons examined, fenoldopam, an agonist of the D1 subclass of receptors, inhibited both the Na+/K+ pump and tetrodotoxin (TTX)-sensitive sodium channels, and quinpirole, an agonist of the D2 subclass of receptors, activated TTX-sensitive sodium channels. Thus D1 and D2 classes of ligands may functionally interact in virtually all dopamine-responsive neurons within the basal ganglia.

Of the vast number of biogenic amines and peptide neurotransmitters acting in the mammalian brain, dopamine receives by far the most attention1. This is partly explained because abnormalities of dopaminergic neurotransmission are implicated in the etiology of several major neurological and psychiatric disorders, including Parkinson’s disease2,3, schizophrenia4,5, attention deficit hyperactivity disorder6,7 and drug abuse8,9. There is considerable evidence from immunocytochemical as well as in situ hybridization studies that the two major subclasses of dopamine receptors are enriched in different projection neurons in the neostriatum10–15. Thus, the D1 subclass of dopamine receptors is enriched in striatal neurons containing substance P and dynorphin that project to the substantia nigra, pars reticulata and entopeduncular nucleus16,17. Conversely, the D2 subclass of dopamine receptors is enriched in enkaphalin-containing striatal neurons projecting to the external segment of the globus pallidus16,17. In contrast to these anatomical data, electrophysiological measurements as well as measurements of mRNA in single cells suggest that many striatal neurons contain both classes of dopamine receptors18. The advent of confocal microscopy led us to re-examine the possible colocalization of D1 and D2 subclasses of receptors. Moreover, on-line recording of intracellular sodium concentration in these cells permitted functional evaluation of the co-expression of the two subclasses of dopamine receptors.

RESULTS After two to three weeks in culture, virtually all cells from embryonic rat striatum seemed to be mature neurons, as they resembled typical medium spiny neurons and stained positively for three neuronal markers: β-tubulin III (Fig. 1a), the alpha 3 isoform of Na+/K+-ATPase (Fig. 1b) and neuron-specific nuclear 226

protein (Fig. 1c). DIC imaging verified the absence of unlabeled (glial) cells. Cultured neurons were labeled using antibodies against either D1-subclass or D2-subclass receptors, each raised in two species (Fig. 1d–l). The cells were examined by confocal microscopy. The improved signal-to-noise ratio for the determination of dopamine receptors, made possible by the smaller depth of focus and the high lateral resolution, permits detection of substances present in low abundance. In all cases, we detected immunofluorescence from virtually all cells. With both antibodies against the D1 receptor subclass, the signal was randomly distributed in the cytoplasm and in the vicinity of the plasma membrane; with both antibodies against the D2 receptor subclass, the signal was more distinct in the region of the plasma membrane and in the perinuclear region. Both D1 and D2 signals were present in dendrites. Double labeling confirmed the co-existence of D1- and D2- like receptors in virtually all cells. DARPP-32 (dopamine- and cyclic AMP-regulated phosphoprotein, 32 kDa) is present in all medium spiny neurons and only in medium spiny neurons in the striatum and nucleus accumbens19,20. We used this selective distribution to identify cells containing both D 1 and D 2 subclasses of receptors. DARPP-32 immunofluorescence was observed in virtually all cultured neurons. Double labeling of cultured neurons for D1-subclass receptors and DARPP-32 (Fig. 2a–c) or D2-subclass receptors and DARPP-32 (Fig. 2d–f) indicated colocalization of both subclasses of receptors with DARPP-32 in virtually all cells; no neurons were singly labeled for D1- or D2-subclass receptors. These results identify the dopamine receptor-labeled cultured cells as medium spiny neurons. To evaluate the possibility that the high degree of colocalization of D1 and D2 subclasses of dopamine receptors resulted from an nature neuroscience • volume 3 no 3 • march 2000


articles


Fig. 1. Primary culture of medium spiny neurons from rat embryonic striatum stained for neuronal markers and for D1 and D2 receptors. (a–c) Combined DIC and confocal immunofluorescence images. (a) Alexa 488 labeling of β-tubulin III. (b) Alpha 3 isoform of Na+/K+ATPase labeled with Alexa 488. (c) Neuron-specific nuclear protein labeled with Alexa 488. (d) D1 receptors labeled with rabbit anti-human D1 receptor primary antibody and Alexa 488-labeled goat anti-rabbit secondary antibody. (e) D2 receptors labeled with goat anti-human D2 receptor primary antibody and Cy3-labeled donkey anti-goat secondary antibody. (f) D1 and D2 receptors, double labeled using same antibodies as for (d, e). (g–i) Immunocytochemical labeling with antibodies used in (d–f) at higher magnification. (j) D1 receptors labeled with rat monoclonal anti-human D1 receptor primary antibody and Alexa 488-labeled goat anti-mouse secondary antibody. (k) D2 receptors labeled with rabbit anti-human D2 receptor primary antibody and Alexa 546-labeled goat anti-rabbit secondary antibody. (l) D1 and D2 receptors double labeled using same antibodies as for (j, k). Green signal, D1; red signal, D2. Yellow indicates overlap.

alteration in the properties of neurons under culture conditions, we examined the distribution of dopamine receptors in ten slices freshly prepared from neostriatum as well as from accumbens shell of each of three adult rats. In these slices, D1 and D2 subclasses of receptors were colocalized as in the cultured neurons (Fig. 3). Confocal recordings of the slices were inspected by an observer blind to the experimental procedure. In 10 randomly selected fields of view (250 µm × 250 µm), a total of 519 cells were identified. We found that 517 of 519 cells (> 99% of the neurons) were positive for both D1 and D2 receptor subclasses. We also examined the colocalization of D1 or D2 subclasses with DARPP-32 in these slices (Fig. 2g–l). Receptors were detected only in DARPP-32-labeled cells, demonstrating that dopamine receptors were present only on neurons, not on glial cells.

We obtained evidence of a functional role for the ubiquitously expressed D1- and D2-subclass dopamine receptors in studies of Na+ influx and efflux in individual cells. This required measurement of initial rates of TTX-sensitive Na+ influx and ouabainsensitive Na+ outflux21 (Fig. 4a; see Methods). Cells were loaded with the sodium-sensitive dye, SBFI-AM. After recording the basal level, extracellular Na+ was replaced by choline chloride. When the intracellular Na+ concentration approached zero, extracellular Na+ was restored, and K+ was removed from the medium to prevent pump-mediated Na+ efflux. This resulted in a rapid influx of Na+. In all protocols, TTX decreased the initial rate of Na+ influx by approximately 60–70%. When the concentration of intracellular Na+ started to level off, the pump was reactivated by adding 4 mM K+ to the medium and replacing Na+ by choline chloride. To prevent influx of Na+ via channels, both TTX and the NMDA- and AMPA-receptor inhibitors MK801 and CNQX were added to the medium. The rapid efflux of Na+ from the cells was almost completely ouabain dependent. The specific D1 subclass agonist fenoldopam caused a pronounced and highly significant decrease in TTX-sensitive Na+

Fig. 2. D1 and D2 subclasses of receptors colocalize with DARPP-32 in primary cultures and slices of rat neostriatal neurons. (a–f) Primary cultures stained for D1 receptor (a), DARPP-32 (b), D1 receptor and DARPP-32 (c), DARPP-32 (d), D2 receptor (e) or DARPP-32 and D2 receptor (f). (g–l) Slices stained for DARPP-32 (g), D1 receptor (h), DARPP-32 and D1 receptor (i), DARPP-32 (j), D2 receptor (k) or DARPP-32 and D2 receptor (l). D1 and D2 receptor antibodies were as described in Fig. 1d and e in all frames except (h); here Alexa 546labeled goat anti-rabbit was used. DARPP-32 was stained with mouse monoclonal anti-bovine DARPP-32 primary antibody and TexasRedX(b, c), Oregon Green- (d, f) and and Alexa 488- (g, i, j, l) labeled goat anti-mouse secondary antibodies. nature neuroscience • volume 3 no 3 • march 2000

227


articles

virtually all neurons provides functional evidence for colocalization of the two classes of dopamine receptors.

DISCUSSION Our immunocytochemical studies indicated that virtually all striatal projection neurons contained both D1 and D2 subclasses of dopamine receptors. Moreover, electrophysiological measurements of Ca2+ and Na+ currents18,22, together with our b 0

a 60

0

Fig. 3. D1-subclass receptors colocalize with D2-subclass receptors in neurons of rat neostriatal slices. (a–c) Neostriatum, high magnification. (d–f) Neostriatum, low magnification. (g–i) Accumbens shell, low magnification. (a, d, g) D1 receptor labeling. (b, e, h). D2 receptor labeling. (c, f, i) D1 and D2 receptor labeling. Antibodies were as described in Fig. 1d and e.

0 Ratio/min

40 30

0 0

20

0

10

0 0

0 0.0 140 mM Na+ 4 mM K+

5.0

10.0

15.0

20.0 min

b 0.140

228

Efflux

0.100 0.080 0.060 *

0.040

*

Fenoldopam +SCH23390

Fenoldopam

Control

Fenoldopam +SCH23390

Control

0.000 0.0 min

Fenoldopam

0.020

0.200 *

0.180 0.160 0.140

Influx

0.120

Efflux

0.100 0.080 0.060 0.040

Quinpirole + sulpiride

Quinpirole

Control

Quinpirole + sulpiride

0.000

Quinpirole

0.020 Control

Fig. 4. Effects of D1 and D2 agonists and antagonists on initial rates of Na+ influx and efflux in primary cultures of striatal cells. (a) Experimental design to study the initial rates of Na+ influx and efflux under Vmax conditions21. (b) Fenoldopam (10–5 M; a D1 agonist), SCH23390 (10–5 M; a D1 antagonist), quinpirole (10–5 M; a D2 agonist) and sulpiride (10–5 M; a D2 antagonist) were present in the incubation mixture as indicated. Images were taken every 20 s. From each coverslip, a field of view containing 5–15 cells was examined. Values (means ± s.e.) for each coverslip were calculated as rate of change in sodium (340:380 ratio) per min. Each group included 6–12 experiments. For both influx and efflux studies, 50–120 cells were used. The control and treatment experiments were carried out on the same day on paired coverslips from the same batch of cells. *Significant differences between control and treated cells, p < 0.01, Student’s t-test.

Influx

0.120

Ratio/min

influx (Fig. 4b, upper panel). The effect was reversed in the presence of the D1 subclass antagonist, SCH 23390. Fenoldopam also significantly inhibited Na+/K+-ATPase-dependent Na+ efflux, and this effect of fenoldopam was also abolished by SCH23390. The D2 subclass agonist quinpirole caused a pronounced and highly significant stimulation of TTX-sensitive Na+ influx, an effect that was abolished by the D2 subclass antagonist sulpiride (Fig. 4b, lower panel). Quinpirole had no significant effect on Na+/K+ATPase-dependent Na+ efflux. In resting cells, free intracellular Na+ was 16.0 mM. When quinpirole was added to the medium, intracellular Na+ rapidly increased (Fig. 5b); this occurred in virtually all cells and was attributable to the activation of TTX-sensitive Na+ channels (Fig. 4b, lower panel). Virtually all cells also showed a slow increase in intracellular Na+ in response to fenoldopam (Fig. 5a), indicating that the effect of this agonist of D1-subclass receptors on pump inhibition masked its effect on channel inhibition (Fig. 4b, upper panel). The finding that sodium levels were altered both by stimulation of D1 and by stimulation of D2 receptors in

Ratio/min


[Na+]i (mM)

50




articles

Na+/K+-ATPase indicate that the α1 isoform of the sodium pump has a higher affinity for Na+ than the α3 isoform31,32. The effects of D1 and D2 subclasses of receptors on sodium fluxes and intracellular sodium concentration have important physiological implications. Regulation of the TTX-sensitive sodium channels directly affects the onset of the action potential. The hyperpolarization between bursts of action potentials results from sodium extrusion via the electrogenic sodium pump; hence inhibition of the sodium pump affects the firing rate33. In addition, sodium may serve as a signaling ion in postsynaptic neurons: intracellular sodium regulates NMDA receptor activity34. Thus opposing effects of D1 and D2 receptors on sodium fluxes and their net effect on intracellular sodium are probably important in the physiological and pathophysiological roles of dopamine receptors in various complex behaviors. Our results provide compelling evidence for the presence of both classes of dopamine receptors on virtually all dopaminoceptive neurons in the striatum and indicate physiological activity of these ubiquitously distributed D1 and D2 subclasses. This suggests intracellular rather than intercellular mechanisms for many of the synergistic and antagonistic actions exerted by activation of D1 and D2 subclasses of dopamine receptors on numerous signal-transduction pathways.

METHODS Fig. 5. Effects of D1 and D2 agonists on [Na+]i in primary cultures of rat striatal cells. At time 0, SBFI-loaded cells were incubated with 5 × 10–5 M fenoldopam (a) or 5 × 10–5 M quinpirole (b), and the ratio images were recorded every 30 s. Values represent ratio changes from the basal level. The results were corrected for SBFI bleaching and leakage. Curves were fit by using the method of least squares. Digital images of SBFI fluorescence under control conditions and eight min after fenoldopam (a) or quinpirole (b) treatment are shown in insets. Changes in color from blue to red on the pseudo-color scale used in these images indicate increases in [Na+]i.

results on Na+ influx and efflux, provide physiological evidence for a high degree of colocalization of D1- and D2-subclass receptors. Contradictory evidence suggesting a low degree of overlap of D1 and D2 subclasses of receptors10–13 can be reconciled with our conclusion by assuming that the striatonigral projection neurons contain high levels of D1-subclass receptors and low levels of D2subclass receptors, and that the converse is true for the striatopallidal pathway. Thus, previous failure to substantially colocalize the two subclasses using immunocytochemistry23 and in situ hybridization24 is probably attributable to inadequate sensitivity of these analytical procedures. Conversely, the high degree of overlap of D1 and D2 subclasses of receptor mRNA seen using PCR25 is attributable to the extremely high sensitivity of this technique. Furthermore, our data suggest that almost undetectable levels of D1-subclass receptors in striatopallidal neurons and of D2-subclass receptors in striatonigral neurons were adequate to produce physiological responses. Intracellular Na+ in cultured striatal neurons was 16 mM, considerably higher than in non-neuronal cells, in which [Na+]i is generally 3–7 mM26–28. We attribute the high [Na+]i in neurons to the presence of the neuron-specific catalytic subunit of the sodium pump, α3. Most non-neuronal cells express only the more widespread α1 isoform29,30. Studies on purified Na+/K+ATPase and on cells transfected with different isoforms of nature neuroscience • volume 3 no 3 • march 2000

Cell culture. Cultures of striatal neurons were prepared from 18–19 dayold rat embryos as described35, with modifications. The cells were plated on poly-D-lysine-coated glass coverslips and cultured in Eagle’s minimal essential medium (MEM) plus F-12 (1:1). The cultures were grown in the presence of 5% FBS. To suppress the growth of glial cells, FBS was exchanged with N2 (1%) two days after plating, and cytosine arabinoside (5 µM) was added to the culture medium from day 5 to day 7. Cells were maintained in culture for two to three weeks before experiments. Intracellular Na+ measurement with SBFI. Cells were loaded with the sodium-sensitive fluorescent probe, SBFI-AM (sodium-binding benzofuran isophthalate; 10 µM). Coverslips were then transferred to an experimental chamber (FCS2, Bioptechs, Butler, Pennsylvania) with laminar flow that allowed instantaneous exposure to the indicated drugs. The cells were excited at wavelength 340/10 nm and 380/10 nm every 20–30 s with a 400-ms exposure time. Emission fluorescence was selected and collected with a 510/30 nm band-pass filter. Recordings were made from cell soma. Collected data were then processed by an image acquisition program from Inovision Corporation, Raleigh, North Carolina. Calibration of intracellular [Na+] was carried out in living cells. Immunostaining. Cells were fixed for 20 min in ice-cold 2% paraformaldehyde, permeabilized with 0.1% saponin, blocked with 7% normal goat (NGS) or donkey (NDS) serum and then incubated overnight at 4°C with primary antibody in PBS containing 2.8% serum and 0.1% saponin. The cells were washed, incubated with fluorescent secondary antibody in PBS with 2.8% serum and 0.1% saponin and washed again. For double labeling, the cells were incubated again overnight at 4°C with the second primary antibody, followed by incubation with the second fluorescent secondary antibody. The slides were subsequently mounted in Prolong antifade (Molecular Probes, Eugene, Oregon). Similar protocols were used for the immunostaining of adult rat striatal slices with minor modifications. D1 subclass dopamine receptors were probed with an affinity-purified rabbit polyclonal antibody against human D1 dopamine receptor (1:100; ref. 36) or with a rat monoclonal antibody against the human D1 dopamine receptor (1:100; ref. 23). D2-subclass dopamine receptors were probed with polyclonal goat affinity-purified anti-human D2 dopamine-receptor antibody (1:200; Santa Cruz Biotechnology, Santa Cruz, California) or with an affinitypurified rabbit polyclonal antibody against human D2 (1:100; ref. 23). The two D1 antibodies used are both reported to be specific for the D1 receptor in the D1 subclass of dopamine receptors and not to crossreact 229


articles


with the D5 receptor23,36. Likewise, the two D2 antibodies are both reported to be specific for the D2 receptor in the D2 subclass of dopamine receptors and not to crossreact with the D3 and D4 receptor23 (statement from Santa Cruz Biotechnology). DARPP-32 was detected with a monoclonal mouse antibody against bovine DARPP-32 (1:1000; ref. 37). The β-tubulin III was detected with a monoclonal mouse antibody against human βtubulin III (1:400; Sigma), the neuron-specific nuclear protein was detected with a mouse monoclonal antibody against neuronal nuclei (1:50; Chemicon, Temecula, California). The α3 isoform of Na+/K+ATPase was probed with polyclonal rabbit affinity-purified anti-rat antibody against the α3 isoform of Na+/K+-ATPase (gift from M. Caplan, Yale University, New Haven, Connecticut). Donkey-anti-goat Cy3 (Jackson ImmunoResearch, West Grove, Pennsylvania), goat-anti-rabbit Oregon green, goat-anti-rabbit TexasRedX, goat-anti-rabbit Alexa 488, goat-anti-rabbit Alexa 546, goat-anti-mouse Oregon green, goat-antimouse TexasRedX and goat-anti-mouse Alexa 488 (all from Molecular Probes) were used at 1:200 as secondary antibodies. Confocal microscopy. The immunolabeled cells were recorded with a Zeiss LSM410 or a Leica TCS SP inverted confocal scanning laser microscope using 63×/1.4 N.A. and 20×/0.75 N.A. objectives. Green fluorescence was excited at 488 nm and detected with a 515–540 nm band-pass filter. Red fluorescence were excited at 543 nm and detected with a 570 nm long-pass filter. Preparations in which the primary antibody was omitted from the staining protocol were used as negative controls.

ACKNOWLEDGEMENTS We thank Ann-Christine Eklöf for experimental assistance. This work was supported by grants from the Swedish Medical Research Council (H.B. and A.A.), Märta and Gunnar V. Philipsons Foundation (H.B.), NIH grants MH40899 and DA 10044 (P.G.) and a grant from Stiftelsen Frimurare Barnhuset I Stockholm (O.A.).

RECEIVED 29 NOVEMBER 1999; ACCEPTED 3 JANUARY 2000 1. Missale, C., Nash, S. R., Robinson, S. W., Jaber, M. & Caron, M. G. Dopamine receptors: from structure to function. Physiol. Rev. 78, 189–225 (1998). 2. Chase, T. N., Oh, J. D. & Blanchet, P. J. Neostriatal mechanisms in Parkinson’s disease. Neurology 51, S30–35 (1998). 3. Grace, A. A., Gerfen, C. R. & Aston-Jones G. Catecholamines in central neurvous system. Overview. Adv. Pharmacol. 42, 655–670 (1998). 4. Knable, M. B. & Weinberger, D. R. Dopamine, the prefrontal cortex and schizophrenia. J. Psychopharmacol. 11, 123–131 (1997). 5. Egan, M. F. & Weinberger, D. R. Neurobiology of schizophrenia. Curr. Opin. Neurobiol. 7, 701–707 (1997). 6. Castellanos, F. X. Toward a pathophysiology of attention-deficit/hyperactivity disorder. Clin. Pediatr. (Phila.) 36, 381–393 (1997). 7. Swanson, J., Castellanos, F. X., Murias, M., LaHoste, G. & Kennedy J. Cognitive neuroscience of attention deficit hyperactivity disorder and hyperkinetic disorder. Curr. Opin. Neurobiol. 8, 263–271 (1998). 8. Self, D. W., Barnhart, W. J., Lehman, D. A. & Nestler, E. J. Opposite modulation of cocaine-seeking behavior by D1- and D2-like dopamine receptor agonists. Science 27, 1586–1589 (1996). 9. Koob, G. F. & Le Moal, M. Drug abuse: hedonic homeostatic dysregulation. Science 278, 52–58 (1997). 10. Gerfen, C. R. & Keefe, K. A. Neostriatal dopamine receptors. Trends Neurosci. 17, 2–3 (1994). 11. Bloch, B. & Le Moine, C. Neostriatal dopamine receptors. Trends Neurosci. 17, 3–4 (1994). 12. Surmeier, D. J., Reiner, A., Levine, M. S. & Ariano, M. A. Neostriatal dopamine receptors. Trends Neurosci. 17, 4–5 (1994).

230

13. Surmeier, D. J., Reiner, A., Levine, M. S. & Ariano, M. A. Are neostriatal dopamine receptors co-localized? Trends Neurosci. 16, 299–305 (1993). 14. Gerfen, C. R. The neostriatal mosaic: multiple levels of compartmental organization. Trends Neurosci. 15, 133–139 (1992). 15. Le Moine, C. & Bloch, B. D1 and D2 dopamine receptor gene expression in the rat striatum: sensitive cRNA probes demonstrate prominent segregation of D1 and D2 mRNAs in distinct neuronal populations of the dorsal and ventral striatum. J. Comp. Neurol. 355, 418–426 (1995). 16. Gerfen, C. R. et al. D1 and D2 dopamine receptor-regulated gene expression of striatonigral and striatopallidal neurons. Science 250, 1429–1432 (1990). 17. Yung, K. K. et al. Immunocytochemical localization of D1 and D2 dopamine receptors in the basal ganglia of the rat: light and electron microscopy. Neuroscience 65, 709–730 (1995). 18. Surmeier, D. J. et al. Dopamine receptor subtypes colocalize in rat striatonigral neurons. Proc. Natl. Acad. Sci. USA 89, 10178–10182 (1992). 19. Ouimet, C. C. & Greengard, P. Distribution of DARPP-32 in the basal ganglia: an electron microscopic study. J. Neurocytol. 19, 39–52 (1990). 20. Anderson, K. D. & Reiner, A. Immunohistochemical localization of DARPP32 in striatal projection neurons and striatal interneurons: implications for the localization of D1-like dopamine receptors on different types of striatal neurons. Brain Res. 568, 235–243 (1991). 21. Borin, M. L. Dual inhibitory effects of dopamine on Na+ homeostasis in rat aorta smooth muscle cells. Am. J. Physiol. 272, C428–438 (1997). 22. Cepeda, C., Buchwald, N. A. & Levine, M. S. Neuromodulatory actions of dopamine in the neostriatum are dependent upon the excitatory amino acid receptor subtypes activated. Proc. Natl. Acad. Sci. USA 90, 9576–9580 (1993). 23. Levey, A. I. et al. Localization of D1 and D2 dopamine receptors in brain with subtype-specific antibodies. Proc. Natl. Acad. Sci. USA 90, 8861–8865 (1993). 24. Le Moine, C., Normand, E. & Bloch, B. Phenotypical characterization of the rat striatal neurons expressing the D1 dopamine receptor gene. Proc. Natl. Acad. Sci. USA 88, 4205–4209 (1991). 25. Surmeier, D. J., Song, W. J. & Yan, Z. Coordinated expression of dopamine receptors in neostriatal medium spiny neurons. J. Neurosci. 16, 6579–6591 (1996). 26. Belusa, R. et al. Mutation of the protein kinase C phosphorylation site on rat alpha1 Na+,K+-ATPase alters regulation of intracellular Na+ and pH and influences cell shape and adhesiveness. J. Biol. Chem. 272, 20179–20184 (1997). 27. Borin, M. L., Goldman, W. F. & Blaustein, M. P. Intracellular free Na+ in resting and activated A7r5 vascular smooth muscle cells. Am. J. Physiol. 264, C1513–1524 (1993). 28. Satoh, H. et al. Regulation of [Na+]i and [Ca2+]i in guinea pig myocytes: dual loading of fluorescent indicators SBFI and fluo 3. Am. J. Physiol. 266, H568–576 (1994). 29. Peng, L., Martin-Vasallo, P. & Sweadner, K. J. Isoforms of Na,K-ATPase alpha and beta subunits in the rat cerebellum and in granule cell cultures. J. Neurosci. 17, 3488–3502 (1997). 30. Pietrini, G., Matteoli, M., Banker, G. & Caplan M. J. Isoforms of the Na,KATPase are present in both axons and dendrites of hippocampal neurons in culture. Proc. Natl. Acad. Sci. USA 89, 8414–8418 (1992). 31. Jewell, E. A. & Lingrel, J. Comparison of the substrate dependence properties of the rat Na,K-ATPase alpha1, alpha 2 and alpha 3 isoforms expressed in HeLa cells. J. Biol. Chem. 266, 16925–16930 (1991). 32. Zahler, R., Zhang, Z.-T., Manor, M. & Boron, W. F. Sodium kinetics of Na,KATPase α isoforms in intact transfected HeLa cells. J. Gen. Physiol. 110, 201–213 (1997). 33. Johnson, S. W., Seutin, V. & North, R. A. Burst firing in dopamine neurons induced by N-methyl-D-aspartate: role of electrogenic sodium pump. Science 258, 665–667 (1992). 34. Yu, X. M. & Salter, M. W. Gain control of NMDA-receptor currents by intracellular sodium. Nature 396, 469–474 (1998). 35. Bockaert, J. et al. Primary culture of striatal neurons: a model of choice for pharmacological and biochemical studies of neurotransmitter receptors. J. Physiol. (Paris) 81, 219–227 (1986). 36. O’Connell, D. P. et al. Localization of dopamine D1A receptor protein in rat kidneys. Am. J. Physiol. 268, F1185–1197 (1995). 37. Hemmings, H. C. Jr. & Greengard, P. DARPP-32, a dopamine- and adenosine 3′:5′-monophosphate-regulated phosphoprotein: regional, tissue, and phylogenetic distribution. J. Neurosci. 6, 1469–1481 (1986).



articles

Growth cone and dendrite dynamics in zebrafish embryos: early events in synaptogenesis imaged in vivo James D. Jontes, JoAnn Buchanan and Stephen J. Smith Department of Molecular and Cellular Physiology, Stanford University School of Medicine, Stanford, California 94305-5435, USA


Correspondence should be addressed to S.J.S. ([email protected])

We used time-lapse fluorescence microscopy to observe the growth of Mauthner cell axons and their postsynaptic targets, the primary motor neurons, in spinal cords of developing zebrafish embryos. Upon reaching successive motor neurons, the Mauthner growth cone paused briefly before continuing along its path. Varicosities formed at regular intervals and were preferentially associated with the target regions of the primary motor neurons. In addition, the postsynaptic motor neurons showed highly dynamic filopodia, which transiently interacted with both the growth cone and the axon. Both Mauthner cell and motor neurons were highly active, each showing motility sufficient to initiate synaptogenesis.

During development of the central nervous system, a mass of undifferentiated cells develops into a precisely organized system that acquires information, processes that information and uses it to generate appropriate behavior. Formation of central chemical synapses, an important part of this self-organization, is poorly understood. Here we take advantage of the small size, optical transparency and rapid development of the zebrafish (Danio rerio) to study central synaptogenesis in an intact vertebrate embryo. The Mauthner cells (M-cells) are large, paired, identified hindbrain neurons that are well characterized in the teleost fishes1–7 and can be identified on the basis of size, position and morphology4,5 (Fig. 1a and b). The M-cells mediate the escape response, a rapid movement away from auditory, vibrational or tactile stimuli1,2,7. Each M-cell axon crosses the ventral midline to extend down the contralateral spinal cord to synapse on primary motor neurons (Pmns) in each spinal hemisegment2,7 (Fig. 1a and c). Serially repeating, bilaterally paired sets of three primary motor neurons, the RoP (rostral primary), MiP (middle primary) and CaP (caudal primary)8–11, are activated during the escape reflex and fast swimming12. In goldfish and tench, the M-cell–Pmn synapse forms between a dorsal collateral of the M-cell axon and a spine-like structure on the ‘ventral dendrite’ of the Pmn2,13,14 (Fig. 1d). Time-lapse imaging in vivo is used to study growth cone dynamics and axonal pathfinding of the retinotectal projections in Xenopus laevis15–19 and zebrafish20, innervation of axial muscle by the Pmns11 and embryonic olfactory neuron pathfinding in zebrafish21, as well as the effect of collapsin 1 on sensory neuron growth cones22; however, these studies do not address the dynamics of pre- or postsynaptic cells as they contact one another. During synapse formation in vitro, dendritic filopodia of central neurons show rapid and extensive dynamics23,24, suggesting an active role in synaptogenesis. The number of filopodia gradually decreases as the number of dendritic spines increases, suggesting that filopodia may represent precursors to the more stable spines23,24. However, even in a tissue slice, the similarity of events in culture to those in an intact, developing organism is unknown. nature neuroscience • volume 3 no 3 • march 2000

Analysis of brain development in vivo involving mainly electron microscopy (EM) reveals many synapses on dendritic filopodia25–27; however, such static studies do not reveal cellular dynamics or detailed interactions between cells. We used time-lapse confocal laser-scanning microscopy and two-photon laser-scanning microscopy of neurons labeled in living zebrafish embryos to characterize the cell motility events involved in making initial contacts. We found that the Mauthner growth cone moved rapidly and relatively smoothly down the spinal cord and seemed to interact transiently with successive target cells. Primary motor neurons interacted with the nearby axon via dynamic filopodia. These findings are consistent with an active involvement of both pre- and postsynaptic elements in establishing synaptic contacts. In addition, regularly spaced varicosities formed along the Mauthner axon in close proximity to motor neurons; these varicosities may represent nascent presynaptic boutons.

RESULTS Growth of the Mauthner cell axon To observe migration of the M-cell growth cone down the spinal cord, lipophilic dye, DiD (confocal imaging) or DiO (two-photon imaging)28, was iontophoretically applied to the cell bodies of the M-cell 18–20 hours after fertilization (Fig. 1b). Consistent with estimations from fixed embryos5,9, the Mauthner growth cone moved at a mean rate of 99.6 ± 22.7 µm per hour (n = 12, ∼28°C; Fig. 2). In general, the Mauthner growth cone moved smoothly and steadily, but time-lapse sequences variably showed slowing or brief pauses every 10–15 µm; these were more pronounced in some cases (Fig. 2b, top and middle) than in others (Fig. 2b, bottom). In all sequences examined, varicosities formed with variable timing along the nascent axon (Fig. 2c), often just behind the growth cone but not always in sequence. Thus, as it moved, the growth cone seemed to leave a trail of axonal thickenings spaced at 13.5 ± 4.3 µm (67 varicosities from 6 axons). 231



articles

a

b

c

d

Because the varicosities we observed may have been nascent presynaptic termini, we investigated their structure using EM. Micrographs of cross-sectioned embryonic spinal cord revealed long, narrow (∼0.6 µm) segments of Mauthner axon largely devoid of synaptic vesicles (Fig. 2d). In addition, short areas of the axon were of larger diameter (∼1.2 µm) and contained small (∼40 nm), clear synaptic vesicles (Fig. 2e). Accumulation of synaptic vesicles in Mauthner axon swellings suggests that these varicosities were sites of early synaptic contact. Labeling of primary motor neurons We injected the ventrolateral myotomes of segments 12–15 with DiD or DiO to label motor neurons. An injection was generally confined to a single hemisegment, labeling only the neurons innervating it. Because of the injection placement, the RoP and CaP motor neurons were preferentially labeled, although occasionally the MiP motor neuron could be identified (Fig. 1c). Before each time-lapse sequence, we used migration of the lateral line primordium to assess embryonic stage to prim-stages 5–18 (ref. 29), roughly corresponding to 24–33 hours post-fertilization (hpf; at 28.5°C). Labeled motor neurons consistently showed variable numbers of characteristic filopodia extending from the cell bodies and ventral dendrites (Figs. 1c and 3). Time-lapse analysis revealed extremely dynamic and variable protrusive/retractive behavior of filopodia, occurring on a time scale of seconds to minutes (Fig. 3). Filopodia were analyzed in 19 primary motor neurons for several basic properties, length, lifetime, density and rates of extension and of retraction. The lifetimes of individual filopodia seemed exponentially distributed with a time constant, τ, of approximately three minutes (Fig. 3b). The filopodium length (mean, 2.7 ± 0.7 µm; n = 571; Fig. 3c) typically varied from 1.3 to 4.0 µm, but could reach lengths of 13 µm. Density of filopodia on the ventral dendrites was highly variable, both over time in the same cell and from cell to cell, and varied from 0.1 to 0.7 filopodia per µm of dendrite. Rates of extension essentially matched those of retraction (∼1.2 ± 0.3 µm per min). These parameters did not change significantly over the developmental peri232

Fig. 1. Illustration of the Mauthner cell–primary motor neuron system. (a) Schematic diagram of the embryonic zebrafish hindbrain and spinal cord showing the overall relationship between the Mauthner cells and the serially repeating primary motor neurons. OV, otic vesicle; Pms, primary motor neurons; My, myotome; M, Mauthner cell; SC, spinal cord. (b) Confocal image of bilaterally symmetric Mauthner neurons in rhombomere 4, which have been labeled with the lipophilic dye, DiD. The image is a superposition of the labeled cells and a DIC image of a more dorsal plane, chosen to give a clearer view of the anatomical relationships. An additional, unidentified neuron with an ascending growth cone has also been labeled on the upper side of the hindbrain. (c) Confocal image of three DiD-labeled primary motor neurons, RoP, MiP and CaP. (d) Schematic representation of the mature Mauther-primary motor neuron synapse, based on ultrastructural analyses in adult goldfish and tench. The basic geometry holds true in the zebrafish: the axon of the motor neuron passes medially and ventrally to the Mauthner axon before leaving the spinal cord through the ventral root. DS, dendritic spine; MA, Mauthner axon; MC, Mauthner collateral; Mn, motor neuron axon; MnVD, motor neuron ventral dendrite.

od investigated (data not shown). However, we cannot rule out short-term changes corresponding to the growth cone pauses we observed. Double-label time-lapse analysis To examine initial contact and possible responses between cells during synaptogenesis, we labeled both Mauthner axons and motor neurons and collected time-lapse sequences. Both the Mauthner growth cone and the motor neurons behaved essentially the same as in single-label experiments; motor neurons showed dynamic filopodia, and growth cones moved rapidly and smoothly. As the Mauthner growth cone first approached and became apposed to a primary motor neuron, its path crossed close to the motor neuron’s ventral dendrites. In some cases, the growth cones seemed to transiently interact with the target regions of the motor neurons (Fig. 4a). As in single-label experiments, analysis of motility of the M-cell growth cone revealed brief and periodic pauses. Such behavior could either represent the specific recognition of targets, or it could simply represent the stochastic nature of growth cone progress. To investigate this, we analyzed growth cone dynamics relative to labeled motor neurons (Fig. 4b–d). Averaging the growth cone motilities from nine suitable time-lapse data sets smoothed away much of the temporal variability, leaving only a marked discontinuity when the growth cone approached a motor neuron (Fig. 4d). This supports the conclusion that the Mauthner growth cone interacts transiently with each successive motor neuron, and that these interactions have a subtle but detectable effect on growth cone dynamics. However, we observed no overt change in growth cone structure, such as collapse or loss of filopodia. The axonal varicosities observed in time-lapse analysis of labeled M-cell axons were also observed in double-label timelapse experiments. Importantly, they formed preferentially near the ventral dendrites of motor neurons, over time associating with them very closely (Figs. 5 and 6). In 5 of 13 observations, an axonal varicosity formed within 5 minutes of the passing of the growth cone, suggesting that the varicosity had arisen from nature neuroscience • volume 3 no 3 • march 2000


a

Distance (µm)

b

Distance (µm)

Fig. 2. Time-lapse imaging of the Mauthner cell growth cone. (a) DiO-labeled growth cone imaged with two-photon microscopy. Image stacks (dorsal view) were acquired every minute for one hour. This sequence shows the emergence of an axonal varicosity that occurred well after the growth cone had moved on. Each displayed image is a projection from a stack of 16 optical sections. (b) Quantitative analysis of the growth cones reveals transient pauses not readily apparent from direct observation of time-lapse sequences. The occurrence and periodicity of these pauses is variable, as shown in this analysis of three different growth cones. Changes in velocity of two growth cones are indicated by an apparent staircase pattern (b, top and middle), whereas another growth cone seemed to move more smoothly (b, bottom). (c) A dorsal view of a DiD labeled M-cell axon imaged with a confocal microscope showing periodicity of the varicosities. (d) An electron micrograph of a Mauthner axon from a cross-sectioned zebrafish spinal cord at ∼28 hpf. The axon (arrow) is narrow and contains only a small number of vesicles. (e) An electron micrograph of another section from the same embryo shown in (d), revealing an axonal varicosity in the same Mauthner axon. The axon has a larger diameter and contains many synaptic vesicles (arrow), as well as a dense core vesicle (arrowhead) and a mitochondrion (M). Dorsal is to the right, and scale bar is 0.25 µm in (d, e).

Distance (µm)


articles

c the growth cone. However, in the remaining eight time-lapse sequences, either no varicosd ity formed or one appeared long after the growth cone had passed. An axonal varicosity formed near the ventral dendrite of a CaP motor neuron (Fig. 6). Dendritic filopodia were quite active near the M-cell axon and interacted with it frequently, touching both the passing growth cone and the axonal shaft (Fig. 7). These interactions were transient, lasting only a few minutes before retraction of the filopodia. The relatively short distance (≤ 3 µm) between the M-cell axon and motor neurons, placed the axon well within reach of the filopodia (mean length, 2.7 µm; maximum, ∼13 µm; Fig. 7).

DISCUSSION Mauthner growth cone The large Mauthner axon growth cone moved rapidly (∼100 µm per hour at ∼28°C; Fig. 2a). Considering that the Mauthner growth cone is believed to pioneer the medial longitudinal fasciculus, surprisingly little exploratory behavior was observed; growth cones largely proceeded along a direct path with few, if any, deviations. This behavior is characteristic of a growth cone moving down a permissive corridor rather than exploring a complex environment. Brief, periodic pauses presented an intriguing feature of growth cone movement (Fig. 2c). These pauses were spaced at 10–15 µm, roughly the distance between successive Pmns in the spinal cord, suggesting interaction of the growth cone with targets. Analysis of double-labeling experiments reveals correlation of these brief reductions in growth cone velocity with proximity to a motor neuron (Fig. 4). nature neuroscience • volume 3 no 3 • march 2000

Time (min)

e

Events in synaptogenesis A conventional expectation is that an active axonal growth cone initiates synaptogenesis with a passive postsynaptic target. This mechanism is most vividly illustrated by the neuromuscular junction, at which the motor growth cone grows up to and terminates on a muscle fiber before developing into a presynaptic bouton. In contrast with this configuration en terminale, many synapses in the CNS, including the M-cell–Pmn synapse, are formed en passant. Like the en-terminale synapse, the en-passant synapse might develop from a motile axonal growth cone. Alternatively, filopodia from the dendrite could be responsible for establishing contact between two neurons. If the Mauthner growth cone were responsible for establishing each synapse as it descended the spinal cord, one might expect the growth cone to show some overt evidence of such an interaction. In the M-cell–Pmn system, we found some evidence for growth cone–motor neuron interactions. The growth cone migrated along a path that approached the motor neuron ventral dendrite, but our analysis revealed only a brief pause when it 233


articles

a

came into contact with successive motor neurons (Fig. 4). In addition, growth cone structure or dynamics seemed to sustain little or no perturbation, in contrast with growth cone behavior affected by guidance cues such as the collapsins/semaphorins, which induce growth cone collapse and temporary paralysis of motility30,31. If the M-cell growth cone is responsible for establishing synaptic contact, the early events must occur very rapidly and without significant disruption of growth cone structure or dynamics. In time-lapse images, we consistently observed formation of varicosities on Mauthner axons in the wake of growth cone migration (Figs. 2, 5 and 6). Electron micrographs showing synaptic vesicles within varicosities along the Mauthner axon (Fig. 2d) support the idea that varicosities represent nascent synapses. Similar swellings are observed on developing corticorubral neurons; EM reveals these varicosities as sites of synaptic contact25. The formation of varicosities along the Mauthner axon may, therefore, represent the early organization of presynaptic boutons. Perhaps our most provocative and striking observation was the presence of highly dynamic filopodia extending from the ventral dendrites of primary motor neurons, which interact with the M-cell axon (Fig. 3). Present throughout the period of synaptogenesis, these filopodia made contact with Mauthner axons and were enriched in regions of the motor neurons that ultimately receive large numbers of synaptic inputs. The varicosity shown in Fig. 6 formed after filopodia–axon interactions and was frequently contacted by dendritic filopodia during the time-lapse sequence. However, it is unclear whether varicosity formation was actually induced by contact with filopodia. Our observations are consistent with proposals that dendritic filopodia may be involved in initiating synaptogenesis and might represent developmental precursors of spines23,24. Electron microscopy demonstrates dendritic filopodia as frequent sites of early synaptic contact25,26, although spine synapses may develop primarily from shaft synapses26. If synaptogenesis does not result directly from growth cone motility, then dendritic filopodia may actively help to bridge the gap between the two cells and to initiate contact. However, it may be unrealistically simplistic to assume that synaptogenesis results exclusively from the actions of either the growth cone or of the dendritic filopodia. Contact between the cells may be the critical event, and it may be irrelevant if this contact results primar234

Percent filopodia

b

Time (min)

c Percent filopodia


Fig. 3. Time-lapse imaging of primary motor neurons and filopodia dynamics. (a) Brief timelapse sequences of two RoP motor neurons. Each image is a projection of 2 (left column) or 14 (right column) sections collected by confocal microscopy of a DiD-labeled cell and two-photon microscopy of a DiO-labeled cell, respectively. (b) Distribution of lifetimes of 571 filopodia from 19 cells imaged by confocal microscopy (∼28°C). Fit to a single exponential reveals a filopodia lifetime of ∼3 min. (c) Distribution of filopodium lengths for the same data set as in (b). The mean length is 2.7 ± 0.7 µm.

Length (µm)

ily from motility of the growth cone or motility of the dendrite. Synaptogenesis, even in what seems a ‘hardwired’ system like the Mauthner–motor neuron synapse, is probably a stochastic process, depending on neither the growth cone nor dendritic filopodia exclusively. Moreover, possible redundancy in capacities of both the pre- and postsynaptic cell to initiate synaptogenesis may be important for ensuring the formation of such a stereotyped synapse. Thus, further investigations may reveal examples of synaptogenesis resulting from action of both growth cones and dendritic filopodia. Timing of synaptogenesis Currently, our only marker for the presence of a nascent synapse is the formation of an axonal varicosity. By this criterion, there can be a considerable and variable lag between the interaction of the growth cone with the motor neuron and synapse formation. Zebrafish embryos respond to touch by ∼21 hpf; hindbrain lesions destroy this touch sensitivity32. Based on the timing of axon extension by the reticulospinal neurons, the Mauthner cell was suggested to participate in this early behavior32. The Mauthner growth cone reaches the rostral spinal cord by ∼20 hpf, indicating a lag of ∼1 h between growth cone passage and initiation of synaptic function. The time frame for synapse formation established by the behavioral studies is consistent with our observations of formation of axonal varicosities with a variable lag following passage of Mauthner growth cones. nature neuroscience • volume 3 no 3 • march 2000


METHODS Labeling of neurons. Zebrafish embryos were collected from a laboratory colony33 and were allowed to develop at ∼28.5°C in a beaker of fresh tank water. At times ranging from 24–35 hpf (motor neuron labeling) or 18–21 hpf (Mauthner labeling), embryos were de-chorionated and mounted on glass depression slides in 1–2% agarose with 0.05% tricaine (3amino benzoic acid ethyl ester; Sigma). DiD (1,1′-dioctadecyl-3,3,3′,3′-tetramethylindocarbocyanine; Molecular Probes, Eugene, Oregon; 0.5–1% in dimethylformamide) or DiO (3,3′-dioctadecyloxacarbocyanine perchorate; Molecular Probes; 0.5–1% in dimethylformamide) was pressure injected into the ventral myotome of anesthetized embryos or iontophoretically applied (0.5–1% in 50% dimethylformamide/50% ethanol) to the cell bodies of Mauthner cells. Embryos were then allowed to develop for several hours, allowing time for growth cones to invade the spinal cord. Well-labeled growth cones were further selected either for single-label time-lapse analysis, or for motor neuron labeling and double-label time-lapse analysis. Electron microscopy. Embryos (24–30 hpf) were fixed for 1 h at room temperature in 3% glutaraldehyde, 2% formaldehyde, 1% acrolein and 1% DMSO in PBS and then further processed for electron microscopy in a Pelco 3450 Laboratory Microwave Processor (Ted Pella, Redding, California). Embryos were washed with 0.1 M cacodylate buffer for 5 min at room temperature and post-fixed with 1% osmium tetroxide in 0.1 M cacodylate buffer containing 0.8% potassium ferricyanide on the bench for 5 min on ice, then given 2 short (10 s) pulses in the microwave. The embryos were washed in distilled water for 5 min, stained en bloc with 2% aqueous uranyl acetate for 20 min and given another short pulse (10 s) in the microwave. They were then dehydrated in an ascending ethanol series (50%, 70%, 95%, 100% × 3) in the microwave oven for 10 s each. The embryos were infiltrated in the microwave in 1:1, 2:1 (resin: ethanol) for 5 min each, and in 100% resin for 5 min using Embed 812 resin (Electron Microscopy Sciences, Fort Washington, Pennsylvania). The embryos were flat embedded and polymerized in a 60°C oven overnight. Thin sections were then cut using a Reichert-Jung Ultracut E ultramicrotome (Leica, Deerfield, Illinois). After staining with 6% uranyl acetate and 1% lead citrate, the sections were imaged in a Philips CM12 transmission electron microscope (FEI/Philips, Hillsboro, Oregon) operating at an accelerating voltage of 80 kV. nature neuroscience • volume 3 no 3 • march 2000

b

Distance (µm)

a

c Distance (µm)

Fig. 4. Simultaneous two-photon imaging of the Mauthner growth cone and primary motor neurons. (a) Maximum-intensity projections from a representative time-lapse sequence during which an M-cell growth cone migrated past a CaP motor neuron. The growth cone was highly active and transiently interacted with the CaP cell, one of its synaptic partners. The growth cone moved past the CaP cell without collapsing, stalling or significantly altering its morphology. (b–d) Analysis of growth cone migration with respect to primary motor neurons. To assess whether growth cone deceleration was related to growth cone–target interactions, we measured rate of growth cone motility relative to the positions of labeled motor neurons. In single-label experiments, growth cone motility varied in overall speed as well as the degree of saltation. (b) This growth cone distinctly paused as it passed the labeled motor neuron. (c) The growth cone represented in this graph (shown in a) moved without clear periodic changes in rate and did not pause near the motor neuron. (d) To filter out stochastic fluctuations and thus reveal whether the rates of growth cone motility indicated any target cell interactions, data from nine time-lapse sequences were averaged after alignment with respect to the labeled motor neurons. The rate of growth cone progress distinctly fluctuates near the motor neurons.

d Distance (µm)


articles

Time (min)

Imaging. Labeled embryos were remounted on cover slips in 1–2% agarose with 0.05% tricaine and imaged by confocal or two-photon microscopy. The embryos were imaged either on a confocal laser-scanning microscope using a Zeiss 20×/NA 0.75 dry objective, or on a twophoton laser-scanning microscope using a Zeiss 63×/NA 0.9 water-immersion objective. (Both microscopes were designed and built by S.J.S., software by N.E. Ziv.) The 633-nm line of a helium-neon laser (0.5–1 mW at the specimen) was used to excite the DiD, and fluorescence passed through a 650-nm long-pass filter before detection. Z-stacks Fig. 5. Projections of the Mauthner axon and sets of primary motor neurons. In both panels, an image stack was collected a few segments rostral to the caudally directed growth cone. In each case, distinct varicosities were associated with the ventral dendrites of primary motor neurons. Secondary motor neurons (asterisks; characterized by their small size, round cell bodies and ventral position) are present in both panels. In the bottom panel, one varicosity does not seem to be associated with a primary motor neuron. However, it is in the correct position to be associated with an unlabeled MiP motor neuron. 235



articles

Fig. 7. Interaction of dendritic filopodia with the Mauthner axon. This two-photon time-lapse sequence illustrates the interaction of a Mauthner growth cone with a RoP motor neuron as these two cells first came into contact. The top panel shows the start of this time-lapse, as the Mauthner growth cone (gc) first reached the RoP ventral dendrite. The bottom panels are a segment of the time-lapse sequence (collected at 1-min intervals) beginning 40 min after the initial cell–cell contact. Each image is a maximum-intensity projection of 14 sections collected at 1-µm steps. Numerous dendritic filopodia extended and retracted from the RoP ventral dendrite, many of which interacted transiently with the Mauthner axon. During this experiment, no stable cell–cell contacts were observed. Because a synapse will ultimately form between the Mauthner axon and the RoP ventral dendrite, it is possible that a filopodium may be responsible for initiating synaptogenic contact.

(6 sections spaced at 1 µm) were collected every min over periods of 30 min–2 h. To image DiO with two-photon microscopy, the Ti:sapphire laser (Mira 900, Coherent, Santa Clara, California) was tuned to 900–910 nm. Every min, we collected 10–18 sections spaced at 1 µm.

Fig. 6. Formation of an axonal varicosity. This two-photon time-lapse sequence shows the formation of an axonal varicosity in close association with a CaP motor neuron. The images are maximum-intensity projections of image stacks displayed at five-min intervals. Initially, the axon swelled, forming a large, elongated varicosity close to the CaP cell. Over time, the varicosity shrank and became more spherical. The arrow in the 5-min panel marks the future site of the stable varicosity, emphasized by a second arrow in the 45-min panel. Throughout the time lapse, there was extensive filopodia activity on the CaP cell near the developing varicosity. 236

Deconvolution. Data sets were deconvolved using the commercial software AutoDeblur (AutoQuant Imaging, Watervliet, New York) using an algorithm for maximum-likelihood estimation that did not require measurement of the point spread function. Maximum-intensity projections resulting from five iterations of deconvolution were used to make the figures. In general, image processing produced neither new specimen features nor loss of observed image detail. However, background noise was strongly suppressed, and image features such as axons and filopodia were sharpened. Analysis of filopodia dynamics. The highest quality time-lapse sequences were selected for analysis. A mouse-driven program (written by N.E. Ziv) was used to measure the lengths of filopodia in each frame of each timelapse sequence. nature neuroscience • volume 3 no 3 • march 2000


articles

Filopodia lengths. Lengths were defined as the distance from the base of the filopodium to its tip. Only filopodia ≥ 3 pixels (∼0.7 µm) were scored, as shorter ‘filopodia’ could not be identified reliably. Lengths were measured from maximum-intensity projections. Thus the values we present slightly underestimate (by 13%) the actual three-dimensional lengths34. Rates. The rate of extension/retraction for a given filopodium was defined as the change in unit length between successive frames. Filopodial extension/retraction rate for a given motor neuron was calculated as the mean rate of extension or retraction for all filopodia observed on that neuron.


Lifetimes. The lifetime of an individual filopodium was defined as length of time for which its length was measureable. A filopodium extending from a position at which one had previously disappeared was counted as a second filopodium. Lifetime data were pooled and plotted in a histogram (bin size, 2 min). Population lifetime was calculated from the time constant, τ, of a single-exponential fit to the histogram. Densities. Filopodia density was calculated as the number of filopodia per unit length for each motor neuron averaged over the duration of the time-lapse sequence. Analysis of growth cone motility. For each frame of a time-lapse sequence, we recorded the leading edge of the growth cone’s lamellipodium. In double-label experiments, growth cone velocities and movement analyses were generated from the position of the growth cone relative to the position of the ventral dendrite of the labeled motor neuron. After aligning to labeled motor neurons, suitable data sets were averaged. As only those with ≥ 15 time points on either side of the motor neuron were included, 9 of 13 data sets were used. Although two-photon-imaged growth cones had reduced and more variable velocities (68.5 ± 29.8 µm per h) that may be attributed to the ∼6C°-cooler ambient temperature of the two-photon microscope stage (∼22°C) compared with that of the warmer confocal microscope (∼28°C), this temperature difference probably did not significantly affect our results. All embryos were incubated at a constant ∼28°C except during injections and during the 1–2 h periods of the time-lapse experiments. Additionally, apart from the enhanced resolution and reduced growth cone rate observed with the two-photon microscope, all the phenomena we report here were observed on both microscopes at both temperatures. Note: Movies of these data can be found on the Nature Neuroscience web site (http://neurosci.nature.com/web_specials/).

ACKNOWLEDGEMENTS Susan Pike conducted initial work on this project. We thank S. Pike and faculty members of the MBL Neural Development and Genetics of the Zebrafish course (Woods Hole, Massachusetts) for help and advice. J.D.J. is a fellow of the Helen Hay Whitney Foundation. This work was supported by NIH grants to S.J.S.

RECEIVED 29 NOVEMBER; ACCEPTED 23 DECEMBER 1999 1. Yasargil, G. M. & Diamond, J. Startle-response in teleost fish: an elementary circuit for neural discrimination. Nature 220, 241–243 (1968). 2. Diamond, J. in Fish Physiology Vol. 5 (eds. Hoar, W. S. & Randall, D. J.) 265–346 (Academic, New York, 1971). 3. Kimmel, C. B., Powell, S. L. & Metcalfe, W. K. Brain neurons which project to the spinal cord in young larvae of the zebrafish. J. Comp. Neurol. 205, 112–127 (1982). 4. Mendelson, B. Development of reticulospinal neurons of the zebrafish. I. Time of origin. J. Comp. Neurol. 251, 160–171 (1986). 5. Mendelson, B. Development of reticulospinal neurons of the zebrafish. II. Early axonal outgrowth and cell body position. J. Comp. Neurol. 251, 172–184 (1986).


6. Metcalfe, W. K., Mendelson, B. & Kimmel, C. B. Segmental homologies among reticulospinal neurons in the hindbrain of zebrafish larva. J. Comp. Neurol. 251, 147–159 (1986). 7. Fetcho, J. R. & Faber, D. S. Identification of motoneurons and interneurons in the spinal network for escapes initiated by the Mauthner cell in goldfish. J. Neurosci. 8, 4192–4213 (1988). 8. Myers, P. Z. Spinal motoneurons of the larval zebrafish. J. Comp. Neurol. 236, 555–561 (1985). 9. Myers, P. Z., Westerfield, M. & Eisen, J. S. Development and axonal outgrowth of identified motoneurons in the zebrafish. J. Neurosci. 6, 2278–2289 (1986). 10. Westerfield, M., McMurray, J. V. & Eisen, J. S. Identified motoneurons and their innervation of axial muscles in the zebrafish. J. Neurosci. 6, 2267–2277 (1986). 11. Eisen, J. S., Myers, P. Z. & Westerfield, M. Pathway selection by growth cones of identified motoneurones in live zebra fish embryos. Nature 320, 269–271 (1986). 12. Liu, D. W. & Westerfield, M. Function of identified motoneurones and coordination of primary and secondary motor systems during zebra fish swimming. J. Physiol. (Lond.) 403, 73–89 (1988). 13. Celio, M. R., Gray, E. G. & Yasargil, G. M. Ultrastructure of the Mauthner axon collateral and its synapses in the goldfish spinal cord. J. Neurocytol. 8, 19–29 (1979). 14. Yasargil, G. M. & Sandri, C. Topography and ultrastructure of commisural interneurons that may establish reciprocal inhibitory connections of the Mauthner axons in the spinal cord of the tench, Tinca tinca L. J. Neurocytol. 19, 111–126 (1990). 15. Harris, W. A., Holt, C. E. & Bonhoeffer, F. Retinal axons with and without their somata, growing to and arborizing in the tectum of Xenopus embryos: a timelapse video study of single fibres in vivo. Development 101, 123–133 (1987). 16. O’Rourke, N. A. & Fraser, S. E. Dynamic changes in optic fiber terminal arbors lead to retinotopic map formation: an in vivo confocal microscopic study. Neuron 5, 159–171 (1990). 17. O’Rourke, N. A., Cline, H. T. & Fraser, S. E. Rapid remodeling of retinal arbors in the tectum with and without blockade of synaptic transmission. Neuron 12, 921–934 (1994). 18. Witte, S., Stier, H. & Cline, H. T. In vivo observations of timecourse and distribution of morphological dynamics in Xenopus retinotectal axon arbors. J. Neurobiol. 31, 219–234 (1996). 19. Wu, G. Y. & Cline, H. T. Stabilization of dendritic arbor structure in vivo by CaMKII. Science 279, 222–226 (1998). 20. Kaethner, R. J. & Stuermer, C. A. Dynamics of terminal arbor formation and target approach of retinotectal axons in living zebrafish embryos: a timelapse study of single axons. J. Neurosci. 12, 3257–3271 (1992). 21. Dynes, J. L. & Ngai, J. Pathfinding of olfactory neuron axons to stereotyped glomerular targets revealed by dynamic imaging in living zebrafish embryos. Neuron 20, 1081–1091 (1998). 22. Shoji, W., Yee, C. S. & Kuwada, J. Y. Zebrafish semaphorin Z1a collapses specific growth cones and alters their pathway in vivo. Development 125, 1275–1283 (1998). 23. Dailey, M. E. & Smith, S. J. The dynamics of dendritic structure in developing hippocampal slices. J. Neurosci. 16, 2983–2994 (1996). 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. Saito, Y. et al. Developing corticorubral axons of the cat form synapses on filopodial dendritic protrusions. Neurosci. Lett. 147, 81–84 (1992). 26. 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). 27. Saito, Y., Song, W.-J. & Murakami, F. Preferential termination of corticorubral axons on spine-like dendritic protrusions in developing cat. J. Neurosci. 17, 8792–8803 (1997). 28. Honig, M. G. & Hume, R. I. Fluorescent carbocyanine dyes allow living neurons of identified origin to be studied in long-term cultures. J. Cell Biol. 103, 171–187 (1986). 29. Kimmel, C. B., Ballard, W. W., Kimmel, S. R., Ullmann, B. & Schilling, T. F. Stages of embryonic development of the zebrafish. Dev. Dyn. 203, 253–310 (1995). 30. Luo, Y., Raible, D. & Raper, J. A. Collapsin: a protein in brain that induces the collapse and paralysis of neuronal growth cones. Cell 75, 217–227 (1993). 31. Kolodkin, A. L. et al. Fasciclin IV: sequence, expression, and function during growth cone guidance in the grasshopper embryo. Neuron 9, 831–845 (1992). 32. Saint-Amant, L. & Drapeau, P. Time course of the development of motor behaviors in the zebrafish embryo. J. Neurobiol. 37, 622–632 (1998). 33. Westerfield, M. The Zebrafish Book (Univ. of Oregon Press, 1995). 34. Maletic-Savatic, M., Malinow, R. & Svoboda, K. Rapid dendritic morphogenesis in CA1 hippocampal dendrites induced by synaptic activity. Science 283, 1923–1927 (1999).

237


articles

Enrichment induces structural changes and recovery from nonspatial memory deficits in CA1 NMDAR1-knockout mice Claire Rampon, Ya-Ping Tang, Joe Goodhouse, Eiji Shimizu, Maureen Kyin and Joe Z. Tsien Department of Molecular Biology, Princeton University, Washington Road, Princeton, New Jersey 08540-1014, USA The first two authors contributed equally to this work.


Correspondence should be addressed to J.Z.T. ([email protected])

We produced CA1-specific NMDA receptor 1 subunit-knockout (CA1-KO) mice to determine the NMDA receptor dependence of nonspatial memory formation and of experience-induced structural plasticity in the CA1 region. CA1-KO mice were profoundly impaired in object recognition, olfactory discrimination and contextual fear memories. Surprisingly, these deficits could be rescued by enriching experience. Using stereological electron microscopy, we found that enrichment induced an increase of the synapse density in the CA1 region in knockouts as well as control littermates. Therefore, our data indicate that CA1 NMDA receptor activity is critical in hippocampus-dependent nonspatial memory, but is not essential for experience-induced synaptic structural changes.

Comparative anatomical studies of hippocampal cytoarchitecture reveal a selective expansion of the CA1 region in the hippocampus in primates with respect to rodents and in humans with respect to monkeys, suggesting that this subregion is important in human hippocampal function1. Damage to the hippocampus or the CA1 subregion in humans leads to deficits in memories of people, objects, places and events (‘declarative memory’)2–4. Similar spatial and nonspatial memory deficits are also observed in a variety of laboratory animals with hippocampal lesions4,5. Recordings of neuronal activity in the CA1 subregion of rodents during behavioral tests show the importance of this subregion in the encoding of nonspatial information6. However, the molecular mechanisms underlying these processes remain unknown. N-methyl-D-aspartate (NMDA) receptors are widely distributed in the brain7 and are essential for the induction of major forms of long-term potentiation (LTP) and long-term depression (LTD)8,9. Enhanced NMDA receptor function in the forebrain improves learning and memory10, indicating its crucial role in these processes. Using the Cre/loxP-mediated recombination system, we developed a region-specific gene-knockout technique to generate conditional-knockout mice in which NMDAR1, a key subunit of NMDA receptor, was selectively deleted in the CA1 subregion11. These CA1-specific NMDAR1 knockout mice lacked NMDA-mediated currents and plasticity in the CA1 region and were profoundly impaired in spatial memory tasks12,13. In this study, our first goal was to examine the role of CA1 NMDA receptor activity in the formation of nonspatial memory and the effects of enriched experience on these memory functions. Our second goal was to examine whether enriched experience during adulthood could lead to structural changes in the CA1 region, and whether CA1 NMDA receptor-mediated responses are required for such structural modifications. Synaptic structural changes are assumed to be the anatomical substrate of long-term storage of learned experience14,15. For example, numbers of dendritic spines in the hippocampus increase after spatial learn238

ing in adult rats16. Based on findings of NMDA receptor function in the developing brain17,18, it is postulated that NMDA receptors might be required for behavioral experience-induced structural plasticity in adult brain. However, new dendritic spines can form on mature hippocampal neurons in vitro in the absence of synaptic activity19, and LTP has no effect on synapse number in the CA1 region20. These findings imply that the role of the NMDA receptor in structural changes in adult brain might differ from that in the developing brain. By using unbiased stereological electron microscopy, we examined the structural changes in our CA1-KO mice before and after enriched experience.

RESULTS NMDAR1 knockout in the CA1 region We confirmed the complete deletion of the NMDAR1 gene in CA1-KO mice using in-situ hybridization histochemistry. No NMDAR1 mRNA was detected in the CA1 subregion (Fig. 1), indicating that the genetic deletion of NMDAR1 in young adult mice was heritable after three years of breeding and was complete. Nonspatial learning and memory deficits CA1-KO and control mice used in all behavioral tests were 2–4 month-old littermates. We used three hippocampus-dependent behavioral tasks to assess associative, nonspatial memory functions. The hippocampus is important in the formation of recognition memory in both human patients and animals 3–5. However, little is known about its anatomical basis and its molecular and cellular mechanisms21. We evaluated recognition memory with a novelobject recognition-memory task10. During the training session, the total amount of time spent exploring two objects was 28.62 ± 3.94 seconds in control mice (n = 17) and 34.71 ± 3.56 seconds in CA1KO mice (n = 12), and no significant exploratory preference was found between control and CA1-KO mice (data not shown). These observations indicate that these two types of mice have the same levels of motivation, curiosity, and interest in exploring novel objects. nature neuroscience • volume 3 no 3 • march 2000


articles

b

Exploratory preference (%)

between control and CA1-KO mice (1.723 ± 0.248 g versus 1.584 ± 0.378 g, respectively). Because these two types of scented food were counterbalanced throughout the experiments, these results indicated that the deficits observed in CA1-KO mice were due neither to abnormal food intake nor to a scent bias per se. We then examined nonspatial memory using a contextual fearFig. 1. Cre/loxP-mediated deletion of NR1 gene in CA1-KO mice (2.5 months old) is complete. An anti10 sense 42-mer oligonucleotide recognizing the NR1 gene was used for in situ hybridization. CTX, cortex; ST, conditioning task . In this hippocampus-dependent task 25 , striatum; DG, dentate gyrus. animals learn to fear an environment by associating it with an aversive stimulus (foot shock). Control and CA1-KO mice significantly differed in contextual For retention tests, at 30 minutes, 2 hours and 24 hours after freezing, but not in freezing responses immediately after the the training (Fig. 2), one object was replaced by a novel object. shocks, in retention tests at 2 hours and 24 hours (p < 0.05, As expected, control mice showed a significant preference for p < 0.01, respectively; Fig. 4a), indicating that CA1-KO mice were exploring the novel object during each retention test. In contrast, impaired in contextual fear memory. However, in a cued-condiCA1-KO mice showed only marginal preference for the novel tioning test, which is hippocampus independent 25, freezing object (Fig. 2). A self-ANOVA between training and retention tests revealed a significant difference in control mice (F3,64 = 7.744, responses of CA1-KO mice to a tone were similar to those of control mice in retention tests at 2 hours (data not shown) and 24 p < 0.001) but not in CA1-KO mice. Repeated measures of hours (Fig. 4b). In addition, no abnormal nociceptive responses ANOVA on retention tests showed a highly significant difference were found in CA1-KO mice: current required to elicit flinchbetween control and CA1-KO mice (F2,27 = 19.435, p < 0.001). A ing/running, jumping or vocalization in the CA1-KO mice was post-hoc analysis using Dunnett’s test demonstrated significance the same as in control mice (data not shown). These data clearof this difference in retention at 30 minutes, 2 hours and 24 hours ly showed that hippocampus-dependent, but not hippocampus(p < 0.05, p < 0.01 and p < 0.01, respectively; Fig. 2). These results independent, fear memory was impaired in CA1-KO mice. indicate profound deficits in novel object recognition memory in CA1-KO. We then examined olfactory discrimination memory using a social transmission of food preference 22. Rodents develop a Control preference for foods they have recently smelled on the breath of CA1 - KO other individuals 23. Lesions restricted to the hippocampus impair this kind of memory 24 hours after social interaction (training)24. However, the molecular mechanisms underlying this kind of memory remain unknown22,24. The task used here consisted of three stages. First, animals became accustomed to eating from a food cup placed on the cage floor. Second, in training sessions, observer mice were allowed to interact with demonstrator mice fed with either cinnamon-scented (1% per weight) or cocoa-scented (2% per weight) food. Third, observer mice were then tested by presentation of both scented foods, Training session and consumption of each food was recorded. Control mice showed a significantly higher preference for food smelled during the training session over unsmelled food than did CA1-KO b mice (F1,26 = 5.291; Fig. 3), indicating that olfactory-discrimination memory was impaired in CA1-KO mice. To ensure that Control this observation was not due to a difference in feeding behavCA1 - KO iors, we measured the total amount of food taken during the ** ** retention session; this quantity did not significantly differ *

Fig. 2. Impaired novel-object recognition memory in CA1-KO mice. Recognition memory is expressed in terms of exploratory preference in the retention tests. The memory for control (n = 17) or CA1-KO (n = 12) mice was measured at three different time intervals after training. The data are expressed as mean ± s.e.; *p < 0.05, **p < 0.01, determined by post-hoc analysis. We observed similar results in separate groups of CA1-KO and control mice. nature neuroscience • volume 3 no 3 • march 2000

Exploratory preference (%)


a

30 min

2h

24 h

Retention interval

239


articles

Contextual freezing (%)

a Control CA1 - KO

**

Immediate freezing

24-hour retention

b

Food preference (%)

Control CA1 - KO

*

24-hour retention

b Control CA1 - KO

enriched control and enriched CA1-KO mice (F2,24 = 11.619, p < 0.01), indicating that enriching experience could only partially rescue the deficits observed in the CA1-KO mice. In the test of social transmission of food preference, we found that enriched control animals (n = 15) did not increase food preference more than naive animals. This possibly reflected either a ceiling effect or a delicate balance between choosing the preferred food and eating any food after 24 hours of food deprivation. However, food preference of enriched CA1-KO mice (n = 12) was dramatically increased over that of naive CA1-KO animals (F1,22 = 4.746, p < 0.05, Fig. 5b), indicating that the enriched experience completely rescued the memory deficits in the CA1-KO mice. We also observed that the enrichment training significantly enhanced contextual freezing in control mice (F1,25 = 7.366, p < 0.05) and completely rescued the deficits observed in the CA1-KO mice (F1,23 = 14.559, p < 0.001). No significant difference was found between enriched control and enriched CA1-KO mice (Fig. 5c). In addition, cued freezing also dramatically increased following enriched experience in both control (F1,25 = 6.213, p < 0.05) and CA1-KO mice (F1,23 = 4.372, p < 0.001; Fig. 5d). These results clearly indicate that the enriched experience could enhance these two kinds of fear memory in control animals and could rescue the deficits in CA1-KO mice.

Control CA1 - KO

Pre-CS

240

a

Total food consumption (grams)

Enrichment-induced recovery of memory deficits These results indicated a profound deficit in three forms of nonspatial memory in CA1-KO mice. Enriched experience can significantly improve performance in spatial maze tests in normal animals26 and attenuate memory deficits in animals with hippocampal lesions27. However, it is not known whether enriched experience can enhance performance of the animals in the three nonspatial memory tasks we used, nor if it is able to rescue the memory deficits observed in CA1-KO mice. To address these issues, we evaluated nonspatial memory using the same tasks with additional groups of adult animals after they were trained daily in an enriched environment for two months. In the novel-object recognition task, we observed an increase in preference for exploring novel objects in both control and CA1-KO mice (n = 14; n = 12, respectively; Fig. 5a). A self-ANOVA between training and retention tests revealed significant differences for control (F3,52 = 24.771, p < 0.001) and CA1-KO mice (F3,44 = 3.797, p < 0.05), indicating that training in the enriched-environment not only enhanced the performance of normal mice but also attenuated the deficits in CA1-KO mice. An integrated statistical analysis of animals with the same genotype confirmed the effects of training on memory in control mice (F 1,29 = 7.674, p < 0.05) and in CA1-KO mice (F 1,22 = 6.674, p < 0.05). However, a repeated measures ANOVA on retention tests still revealed a highly significant group difference between

Cued freezing (%)


Fig. 3. Social transmission of food preference. Olfaction-discrimination memory is expressed in terms of food preference. (a) The memory of either control (n = 15) or CA1-KO (n = 13) mice was measured 24 h after training. (b) Food consumption data are expressed as mean ± s.e. *p < 0.05 determined by one-way ANOVA.

CS

Fig. 4. Fear-conditioning tasks. (a) Contextual fear conditioning. Memory is expressed as percent of freezing responses. Immediate learning indicates the freezing response during the 30 s immediately after shock in the training session. Contextual fear memory in controls (n = 14) and CA1-KO (n = 12) was measured 24 h after training. (b) Cued fear conditioning. Memory is expressed as percentage of freezing responses. There was no significant difference in proportion of freezing responses either at pre-CS or retention test between control mice (n = 14) and the CA1-KO mice (n = 12). All data are expressed as mean ± s.e.; **p < 0.01, determined by one-way ANOVA. Similar results were obtained from another set of experiments. nature neuroscience • volume 3 no 3 • march 2000


articles

Naive Enriched

Contextual freezing (%)

c Exploratory preference (%)

a **

* *

Control

Naive Enriched

Control

CA1-KO

CA1-KO

24-hour retention

24-hour retention

Control

*

Naive Enriched

CA1-KO

24-hour retention

Enrichment-induced anatomical changes What are the molecular and structural mechanisms underlying the enrichment-induced effects? Enriched experience promotes various biochemical and morphological changes in several brain regions26,28–31. Synaptic structural changes are believed to be the anatomical substrate of long-term storage of learned experience, and it is assumed that NMDA receptors are required for experience-dependent anatomical changes in the adult brain. As the first step toward the genetic dissection of the molecular mechanisms of experience-induced structural plasticity in the adult hippocampal CA1 region, we used light and electron microscopy to examine the effects of enrichment on anatomical changes in the CA1 region and the role of the NMDA receptor in this effect. Nissl-stained tissue of CA1-KO mice shows no gross anatomical abnormality12. We used the Golgi-impregnation technique32 to initially assess dendritic structures. Golgi-impregnated CA1 pyramidal neurons of control and CA1-KO naive animals presented similar dendritic morphology (Fig. 6a and b). Quantitative analysis of dendritic-spine density on CA1 pyramidal cells revealed no significant difference between naive control (8.1 ± 0.8 spines per 10 µm of apical dendrite, mean ± s.e.) and naive CA1-KO mice (8.6 ± 0.6 spines; Fig. 6c). This suggests that conditional knockout of the NMDAR1 gene in the CA1 region, which occurs in the postnatal third and fourth weeks, did not result in abnormal spine density. To characterize the effects of enrichment on dendritic spine density, we then counted the number of spines on dendrites of

Fig. 6. Enrichment-induced increase in CA1 spine density in both normal control and CA1-KO mice. Photomicrographs of Golgi-impregnated apical dendritic segments of CA1 pyramidal neurons of naive control (a) and naive CA1-KO (b) mice. (c) Bars represent a scatter plot of individual average numbers of dendritic spines per 10 µm length of CA1 pyramidal dendritic segments. Diamonds represent group means ± s.e. (n = 3 for each group). The sexes (M or F) and ages in months of animals in each column (from top to bottom) are naive normal control (M, 3.5; M, 4; M, 3.5), enriched normal control (M, 3.5; M, 5; M, 4.5), naive CA1KO (M, 4.5; M, 5; M, 4) and enriched CA1-KO (M, 5; M, 5; M, 4.5). *p < 0.05, Mann-Whitney U-test. nature neuroscience • volume 3 no 3 • march 2000

*

*

Control

CA1-KO

24-hour retention

CA1 pyramidal cells of enriched animals. We observed a significantly higher density of dendritic spines in enriched contol mice compared with naive animals (10.8 ± 1.1 versus 8.1 ± 0.8; p < 0.05; Fig. 6c). This is consistent with a previous report of increased spine density on CA1 neurons in rats after enrichment16. Surprisingly, enrichment similarly increased spine density in CA1-KO mice (10.0 ± 0.3 versus 8.6 ± 0.6; p < 0.05). Statistical analysis revealed no significant difference between enriched contol and enriched CA1-KO animals (Fig. 6c).

a

b

c Dendritic spines per 10 µm

Food preference (%)

Naive Enriched

Cued freezing (%)

d

b


***

*

Fig. 5. Effects of enriched experience on nonspatial memory in both control and CA1-KO mice. (a) Enriched experience increased exploratory preference in both control (n = 14) and CA1-KO (n = 12) mice in the novelobject recognition task. (b) Enriched experience rescued memory deficits of CA1-KO (n = 12) mice in the social transmission of food preference. For control mice, n = 15. (c) Enriched experience enhanced contextual freezing response in both control (n = 13) and CA1-KO (n = 11) mice in the fear-conditioning task. There was no significant difference in immediate freezing between the two groups (data not shown). (d) Enriched experience increased cued freezing in the fear-conditioning task. There was no significant difference in pre-CS freezing or cued freezing between control mice (n = 13) and CA1-KO (n = 11) mice. All data are expressed as mean ± s.e. *p < 0.05, ** p < 0.01, ***p < 0.001 in one-way ANOVA.

Naive Enriched Control

Naive Enriched CA1-KO

241


articles

242

b

c

d

e

Naive Enriched (n = 7) (n = 5) Control

Naive Enriched (n = 4) (n = 4) CA1-KO mice

To investigate the possible differential effect of enrichment and NMDA receptor activity on changes in the density of synaptic subtypes, we examined the distribution of CA1 synapses in three major subclasses (non-perforated, perforated and shaft synapses) categorized according to the profile of the postsynaptic density (PSD) in pairs of serial sections. The PSD was classified as non-perforated if the thickening was continuous or as perforated if the PSD was interrupted by electron-lucent regions (Fig. 8a). Synapses with asymmetric PSDs on dendritic shafts were identified using Gray’s criteria36

a

b

d

c Perforated Non-perforated Shaft

Naive control

**

Enriched control

Synapses per 100 µm3

Fig. 8. Major subclasses of CA1 synapses in the CA1 region. (a, b) Electron micrographs illustrating the ultrastructure of different types of synapses in the stratum radiatum of CA1. Scale bars, 0.5 µm. (a) Non-perforated axospinous synapse (np) and perforated axospinous synapse (p). (b) Axodendritic synapses involving dendritic shafts (s). (c, d) Estimated distribution of CA1 synapses in the categories described above for control (c) and CA1-KO (d) animals, either naive or reared in an enriched environment. Synaptic densities were calculated per surface area and then expressed as the number of synapses per 100 µm3 by using the stereological coefficient obtained with the disector method. The data are expressed as mean ± s.e.; *p < 0.05, Mann-Whitney U-test.

a


These data indicate that enriched experience resulted in changes in dendritic-spine density and suggest that this increase in spine density can occur in the absence of NMDA receptor activity. These results are consistent with the finding that mature CA1 pyramidal neurons in vitro can grow spines in absence of NMDA-receptor activation19. However, Golgi staining should be interpreted cautiously. First, this method does not reveal spines hidden beneath or sitting above the dendritic segment, thus, spine density measured this way probably underestimates the actual number of spines. Second, some branched spine heads receive no innervation33; therefore, spine density should not be simply extrapolated to synapse density. To address these concerns, we used electron microscopy to quantitatively analyze synapse density in the stratum radiatum of the CA1 hippocampal region (Fig. 7a–d). We ensured unbiased sampling by using the stereological disector technique34,35. No significant difference in synaptic density was observed between naive control mice (76.9 ± 1.9 synapses per 100 µm3, mean ± s.e.) and naive CA1-KO mice (70.0 ± 4.4 synapses, Fig. 7e). However, after the enrichment training, control animals showed a significant increase of CA1 synaptic density (93.8 ± 5.3) over the naive control group (p < 0.05). Synaptic density was also strikingly higher in trained CA1-KO mice (91.2 ± 2.9) compared to the naive CA1-KO group (p < 0.01, Fig. 7e). Furthermore, no significant difference in synaptic density was found between enriched control and enriched CA1-KO mice. These data indicate that enriched experience promotes synaptic structural changes in the CA1 hippocampal region. Furthermore, these results suggest that the NMDA receptor is not required for the increase of CA1 synaptic density.



Fig. 7. Enrichment-induced increase in CA1 synapse density in both control and CA1-KO mice. Representative electron photomicrographs illustrating synapses (indicated by arrows) in the stratum radiatum of the CA1 region of naive controls (a), naive CA1-KO mice (b), enriched controls (c) and enriched CA1-KO mice (d). Scale bars, 1 µm. (e) Diamonds represent group means ± s.e. of estimated synaptic densities in the stratum radiatum of the CA1 region before and after enrichment, calculated with the stereological disector. Bars represent a scatter plot of the CA1 synaptic density in individual animals in each group. The sexes (M or F) and ages in months of animals in each column are (from top to bottom) naive control (M, 4.5; M, 5; M, 4; M, 4; M, 3; M, 3.5; M, 4), enriched control (M, 4.5; M, 5; M, 5.5; M, 4; M, 5), naive CA1-KO (M, 4; M, 4.5; F, 3.5; M, 4) and enriched CA1-KO (M, 3.5; F, 4; M, 4.5; M, 5). *p < 0.05, determined by Mann-Whitney U-test.

*

Naive CA1-KO

Enriched CA1-KO




articles

and were counted separately from those on dendritic spines (Fig. 8b). We found a significantly higher density of non-perforated synapses in control mice after enrichment (77.1 ± 5.2, mean ± s.e.) than in naive control animals (62.6 ± 2.3, p < 0.05; Fig. 8c). Remarkably, enrichment also significantly increased the density of non-perforated synapses in CA1-KO mice (78.3 ± 5.4 versus 60.7 ± 5.1; p < 0.05; Fig. 8d), whereas the densities of perforated and shaft synapses remained unchanged after enrichment in both control and CA1-KO mice. For better observation of the synaptic structural changes, a sophisticated approach using three-dimensional reconstruction of synapses viewed through serial electron micrographs would be valuable33. Possible tissue shrinkage during the embedding process can be ruled out as explanation for these findings. First, the effect is synapse specific. If there were a generalized shrinkage in the enriched animals with no other change, then all types of synapses should be affected uniformly. Second, the mitochondrial crosssectional diameters were uniform across the naive and enriched conditions, showing that no generalized shrinkage of individual process occurred (data not shown).

DISCUSSION Here we examined three kinds of nonspatial memory in mice lacking NMDAR1 in the CA1 region. Our analysis provides evidence supporting the important role of NMDA receptor activity in the CA1 region for the formation of hippocampus-dependent nonspatial memory. Taking into account the deficit in spatial learning in the CA1-KO mice12, we conclude that NMDA receptor activity in CA1 is essential for the formation of both spatial and nonspatial memory. More interestingly, we found that the memory deficits associated with the disruption of the hippocampal NMDA receptor could be rescued by daily training in an enriched environment. We show that behavioral experience can enhance learning and memory in the mutant animals. Thus, the memory deficits observed in gentic mutant animals are not necessarily irreversible, and enriched experience could promote recovery of some of the deficits. To investigate further the possible cellular mechanisms underlying these enrichment effects, we systematically evaluated the anatomical changes using light and electron microscopy. We showed that enriched experience promotes a significant increase in synapse density in the CA1 hippocampal field. Because there is no neurogenesis in the CA1 region26, the increased synapse density is likely to reflect a higher number of synapses per neuron rather than an increased number of pyramidal cells in CA1. At the ultrastructural level, we showed that mice lacking the NMDA receptor in the CA1 region also increase synaptic density following exposure to an enriched environment. This suggests that NMDA-receptor activity is not required for structural plasticity in the CA1 region of adult animals induced by behavioral experience, and thus the molecular mechanism underlying adult activitydependent structural plasticity is very different from mechanisms active in the developing brain17,18. As synaptic plasticity can also be induced in the CA1 region by an NMDA receptor-independent process37,38 such as LTP dependent on voltage-gated calcium channels, other mechanisms might serve as putative candidates for experience-induced structural plasticity in adult brain. It remains to be determined whether the enrichment-induced increase in CA1 synaptic density has any functional role in the concomitant behavioral effects. It is likely that other brain regions may undergo similar biochemical and structural changes after enrichment, and also might participate in the enrichment-induced behavioral improvement. Enhanced synaptic coincidence detection in the forebrain of transgenic mice via upregulation of subnature neuroscience • volume 3 no 3 • march 2000

unit 2B of the NMDA receptor also leads to overall enhancement of learning and memory10. Thus, changes in the composition (for instance, ratio of NR2A to NR2B subunits) of the NMDA receptor complex may be a possible mechanism for the enrichmentinduced effects. Taken together, these findings indicate that learning and memory might be enhanced in mammals by genetic factors as well as behavioral experience.

METHODS Animals. The CA1-KO and control mice were produced as described11. Throughout the behavioral and histological experiments, observers were blind to the genotype of an individual animal. In situ hybridization. Described procedures10 were used. Briefly, an antisense 42-mer oligonucleotide probe (5′-ACC ACT CTT TCT ATC CTG CAG GTT CTT CCT CCA CAC GTT-3′), which recognizes NR1 exon 20, was end labeled with [35S]-dATP. After hybridizing with the oligonucleotide probe (5 × 105 cpm per slide) at 47°C for 24 h, brain sections (20 µm) were washed in 2× SSC at room temperature (RT) followed by two washes in 0.2× SSC at 60°C and one wash in 0.1× SSC at RT. Interestingly, preliminary observations indicated that the deletion of the NR1 gene in six-month-old knockouts seemed to spread to other forebrain regions such as the cortex, probably reflecting the recombination threshold effect as the CaMKII promoter-driven Cre accumulated. Enriched environment training. Adult littermates (1.5–2 months old) were randomly distributed into two experimental groups. One group was kept in standard cages (naive group) and the other group trained in an enriched environment for three hours daily for two months (enriched group). The enrichment training was carried out in specially designed boxes (1.5 m × 0.8 m × 0.8 m), in which various toys, running wheels and small houses were changed every other day to encourage exploration. Food and water were also available in the boxes. Because electrophysiological, behavioral and anatomical experiments showed similar results for wild-type and Cre transgenic mice, these two genotypes of mice were pooled together as controls. Novel-object recognition task and fear-conditioning tasks. The experimental protocol was the same as described previously10, except that the duration of the training period in the object-recognition task was 15 min. Social transmission of food preference task. In this experiment, ‘observer’ mice were tested for memory and ‘demonstrator’ wild-type mice were used to interact with observers according to a described procedure22. After interaction, the observers were deprived of food for ∼24 h and then tested for memory. During testing, observers were offered both cinnamon- and cocoa-scented food for 2 h, and consumption of each food was subsequently measured. The preferred food had the same scent as food eaten by the demonstrator with whom the observer interacted. Preference was calculated as percentage ratio of the amount of preferred food consumed over 50% of the total amount of food consumed. Histology. For electron microscopy, contol groups were composed of males, 3.5–5.5 months old. The CA1-KO naive mice were 3 males (4.5, 4 and 4 months) and 1 female (3.5 months), and the CA1-KO enriched mice were 3 males (5, 4.5 and 3.5 months) and 1 female (4 months). Because CA1 dendritic spine density in rats fluctuates over the estrous cycle39, females were killed during proestrus, the stage at which the spine numbers of female rats approach those of male rats (although it is not known whether the same phenomenon exists in mice). Animals were anesthetized and perfused with 2% paraformaldehyde, 2% glutaraldehyde and 1.5% saturated picric acid in 0.1 M phosphate buffer (pH 7.4) and postfixed overnight in the same solution. Following a modified version of the single-section Golgi impregnation technique32, 100 µm-thick coronal sections were cut with a vibratome into a bath of 3% potassium dichromate in distilled water and incubated overnight in this solution. The slide assemblies were incubated in the dark in a solution of 1.5% silver nitrate for 2 days. For electron microscopy, we used a vibratome to cut 250 µm-thick coronal sections into a bath of 0.1 M sodium cacodylate buffer. Hippocampal 243


articles


sections corresponding to figures 47–50 of the mouse brain atlas40 were dissected, rinsed in distilled water and postfixed in 1% osmium tetroxide followed with 1% ferrocyanide-reduced osmium tetroxide. Tissues were rinsed, stained en bloc overnight in 1% aqueous uranyl acetate, rinsed and dehydrated and subsequently infiltrated and flat embedded with Embed 812 resin. Sections (0.5 µm) were cut and stained with 1% toluidine blue. Sections were examined to determine how the block should be trimmed for ultrathin sectioning to include the apex of the CA1 pyramidal cell layer, from the stratum radiatum to the stratum lacunosum moleculare. Ultrathin sections were cut with a diamond knife, placed on carbon Formvarcoated slot grids and stained with 1% uranyl acetate and lead citrate. Section thickness was measured by scanning electron microscope measurements to range from 0.087 to 0.105 µm, depending on the series. Data analysis: spine density. Golgi-impregnated dendritic segments selected for analysis were located 100–250 µm from the pyramidal cell bodies in the stratum radiatum, belonged to a thoroughly impregnated pyramidal neuron, remained approximately in the plane of focus and were >10 µm in length and 0.80–1 µm in width. Calculations of spine density were obtained by measuring dendritic segments matching the above criteria (control enriched n = 37, CA1-KO naive n = 38, CA1-KO enriched n = 41 and control naive n = 43) from age- and sex-balanced sets of 3 animals of each group. These segments represented a total dendritic length of 1001–1284 µm per group of mice. Data analysis: synapse density. Synapse density was estimated by the disector method34,35. From each brain, 15 electron micrographs of separate regions within the stratum radiatum, 100–250 µm from the pyramidal cell layer, were taken from different sections at 10,000× on a JEOL 100C electron microscope to create 15 ‘reference’ planes. The micrographs covered an area of approximately 2100 µm2 per animal. Micrographs of exactly the same regions were taken on an adjacent section to create ‘look-up’ planes. The number of synapses contained in a reference plane but absent in the corresponding look-up plane (Qi) was counted to determine the number of synapses present within the volume defined by the reference plane, the look-up plane and the distance between them (hi). Two sides of the rectangular picture were assigned randomly as inclusion or exclusion edges to create a counting frame that minimized potential edge effects across samples41. To increase the sampling reliability, 606–1043 synapses were counted over a total area of 8,400–14,700 µm2 from 600 electron micrographs for each experimental group of mice. Section thickness (hi) was determined by scanning electron microscope measurements (see Histology). Estimated synapse density (est Nv) was calculated as est Nv = ΣQi / Σ (Ai × hi) Note: Additional methods details can be found on the Nature Neuroscience web site (http://neurosci.nature.com/web_specials/).

ACKNOWLEDGEMENTS We thank E. Gould for help with stereology and statistics and reading the manuscript, and P. Tanapat for technical advice concerning the Golgi method. This work was supported in part by a postdoctoral fellowship from Fondation pour la Recherche Medicale to C.R. and by grants from Princeton University, Beckman Foundation and NIH to J.Z.T.

RECEIVED 12 OCTOBER; ACCEPTED 28 DECEMBER 1999 1. Rosene, D. L. & Van Hoesen, G. W. in Cerebral Cortex Vol. 6: Further Aspects of Cortical Function, Including Hippocampus (eds. Jones, E. G. & Petes, A.) 345–447 (Plenum, New York, 1987). 2. Scoville, W. B. & Milner, B. Loss of recent memory after bilateral hippocampal lesion. J. Neurol. Neurosurg. Psychiatry 20, 11–12 (1957). 3. Zola-Morgan, S., Squire, L. R. & Amaral, D. G. Human amnesia and the medial temporal region: enduring memory impairment following a bilateral lesion limited to field CA1 of the hippocampus. J. Neurosci. 6, 2950–2967 (1986). 4. Squire L. R. Memory and the hippocampus: a synthesis from findings with rats, monkeys, and humans. Psychol. Rev. 99, 195–231 (1992). 5. Mumby, D. G. et al. Ischemia-induced object-recognition deficits in rats are attenuated by hippocampal ablation before or soon after ischemia. Behav. Neurosci. 110, 266–281 (1996).

244

6. Wood, E. R., Dudchenko, P. A. & Eichenbaum, H. The global record of memory in hippocampal neuronal activity. Nature 397, 613–616 (1999). 7. Moriyoshi, K. et al. Molecular cloning and characterization of the rat NMDA receptor. Nature 354, 31–37 (1991). 8. Nicoll, R. A. & Malenka, R. C. Expression mechanisms underlying NMDA receptor-dependent long-term potentiation. Ann. NY Acad. Sci. 868, 515–525 (1999). 9. Bear, M. F. & Malenka, R. C. Synaptic plasticity: LTP and LTD. Curr. Opin. Neurobiol. 4, 389–399 (1994). 10. Tang, Y.-P. et al. Genetic enhancement of learning and memory in mice. Nature 401, 63–69 (1999). 11. Tsien, J. Z. et al. Subregion- and cell type-restricted gene knockout in mouse brain. Cell 87, 1317–1326 (1996). 12. Tsien, J. Z., Huerta, P. T. & Tonegawa, S. The essential role of hippocampal CA1 NMDA receptor-dependent synaptic plasticity in spatial memory. Cell 87, 1327–1338 (1996). 13. McHugh, T. J., Blum, K. I., Tsien, J. Z., Tonegawa, S. & Wilson, M. A. Impaired hippocampal representation of space in CA1-specific NMDAR1 knockout mice. Cell 87, 1339–1349 (1996). 14. Woolf, N. J. A structural basis for memory storage in mammals. Prog. Neurobiol. 55, 59–77 (1998). 15. Bailey, C. H. & Kandel, E. R. Structural changes accompanying memory storage. Annu. Rev. Physiol. 55, 397–426 (1993). 16. Moser, M. B., Trommald, M. & Andersen, P. An increase in dendritic spine density on hippocampal CA1 pyramidal cells following spatial learning in adult rats suggests the formation of new synapses. Proc. Natl. Acad. Sci. USA 91, 12673–12675 (1994). 17. Goodman, C. S. & Shatz, C. J. Developmental mechanisms that generate precise patterns of neuronal connectivity. Cell 72, Suppl. 77–98 (1993). 18. Li, Y., Erzurumlu, R. S., Chen, C., Jhaveri, S. & Tonegawa, S. Whisker-related neuronal patterns fail to develop in the trigeminal brainstem nuclei of NMDAR1 knockout mice. Cell 76, 427–437 (1994). 19. Kirov, S. A. & Harris, K. M. Dendrites are more spiny on mature hippocampal neurons when synapses are inactivated. Nat. Neurosci. 10, 878–883 (1999). 20. Sorra, K. E. & Harris, K. M. Stability in synapse number and size at 2 hr after long-term potentiation in hippocampal area CA1. J. Neurosci. 18, 658–671 (1998). 21. Mansuy, I. M., Mayford, M., Jacob, B., Kandel, E. R. & Bach, M. E. Restricted and regulated overexpression reveals calcineurin as a key component in the transition from short-term to long-term memory. Cell 92, 39–49 (1998). 22. Kogan, J. H. et al. Spaced training induces normal long-term memory in CREB mutant mice. Curr. Biol. 7, 1–11 (1996). 23. Strupp, B. J. & Levitsky, D. A. Social transmission of food preferences in adult hooded rats (Rattus norvegicus). J. Comp. Psychol. 98, 257–266 (1984). 24. Bunsey, M. & Eichenbaum, H. Selective damage to the hippocampal region blocks long-term retention of a natural and nonspatial stimulus-stimulus association. Hippocampus 5, 546–556 (1995). 25. Phillips, R. G. & LeDoux, J. E. Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav. Neurosci. 106, 274–285 (1992). 26. Kempermann, G., Kuhn, G. H. & Gage, F. H. More hippocampal neurons in adult mice living in an enriched environment. Nature 386, 493–495 (1997). 27. Dalrymple-Alford, J. C. & Benton, D. Preoperative differential housing and dorsal hippocampal lesions in rats. Behav. Neurosci. 98, 23–34 (1984). 28. Rosenzweig, M. R. Environmental complexity, cerebral change, and behavior. Am. Psychol. 21, 321–332 (1966). 29. Diamond, M. C. Enriching Heredity: The Impact of the Environment on the Anatomy of the Brain (Free Press, New York, 1988). 30. Fiala, B. A., Joyce, J. N. & Greenough, W. T. Environmental complexity modulates growth of granule cell dendrites in developing but not adult hippocampus of rats. Exp. Neurol. 59, 372–383 (1978). 31. Greenough, W. T., Withers, G. S. & Wallace, C. S. in The Biology of Memory (eds. Squire, L. R. & Lindenbaum, E.) 159–185 (Schattauer, Stuttgart, 1990). 32. Gabbott, P. L & Somogyi, J. The “single” Golgi impregnation procedure: methodological description. J. Neurosci. Methods 11, 221–230 (1984). 33. Sorra, K. E., Fiala, J. C. & Harris, K. M. Critical assessment of the involvement of perforations, spinules, and spine branching in hippocampal synapse formation. J. Comp. Neurol. 398, 225–240 (1998). 34. Sterio, D. C. The unbiased estimation of number and sizes of particles using the disector. J. Microsc. 134, 127–136 (1984). 35. DeGroot, D. M. G. & Bierman, E. P. B. A critical evaluation of methods for estimating the numerical density of synapses. J. Neurosci. Methods 18, 79–101 (1986). 36. Gray, E. G. Axosomatic and axodendritic synapses of the cerebral cortex: An electron microscopic study. J. Anat. 83, 420–433 (1959). 37. Nicoll, R. A. & Malenka, R. C. Contrasting properties of two forms of longterm potentiation in the hippocampus. Nature 377, 115–118 (1995). 38. Stevens, C. F. & Sullivan, J. Synaptic plasticity. Curr. Biol. 8, R151–153 (1998). 39. Woolley, C. S., Gould, E., Frankfurt, M. & McEwen, B. S. Naturally occurring fluctuation in dendritic spine density on adult hippocampal pyramidal neurons. J. Neurosci. 10, 4035–4039 (1990). 40. Franklin, K. B. J. & Paxinos, G. The mouse brain in stereotaxic coordinates. (Academic, San Diego, 1997). 41. Gundersen, H. J. G. Notes on the estimation of the numerical density of arbitrary profiles: the edge effect. J. Microsc. 111, 219–223 (1977).



articles

Muscles express motor patterns of non-innervating neural networks by filtering broad-band input Lee G. Morris1, Jeff B. Thuma2 and Scott L. Hooper2 1

Department of Physiology and Biophysics, Mt. Sinai Medical School, Box 1218, 1 Gustave L. Levy Place, New York, New York 10029, USA

2

Neuroscience Program, Department of Biological Sciences, Irvine Hall, Ohio University, Athens, Ohio 45701, USA The first two authors contributed equally to this work.


Correspondence should be addressed to S.L.H. ([email protected])

We describe three slow muscles that responded to low-frequency modulation of a high-frequency neuronal input and, consequently, could express the motor patterns of neural networks whose neurons did not directly innervate the muscles. Two of these muscles responded to different frequency components present in the same input, and as a result each muscle expressed the motor pattern of a different, non-innervating, neural network. In an analogous manner, the distinct dynamics of the multiple intracellular processes that most cells possess may allow each process to respond to, and hence differentiate among, specific frequency ranges present in broad-band input.

Neurons receive input across an extremely wide frequency range (from high-frequency sound at tens of kHz to circadian rhythms at ∼10 µHz). Muscles receive input across a narrower but still wide range (10 µHz to several hundred Hz, the maximum firing rate of neurons). Neurons and muscles cannot accurately follow the higher portions of these ranges. For instance, the refractory period limits neuron firing to a few hundred Hz, and relaxation kinetics prevent muscles from accurately following motor neuron bursts faster than ∼1 Hz1. Neuronal and muscle intracellular processes (for example, Ca 2+ and second messenger concentration, protein kinase activation, protein phosphorylation and gene expression) are also slow, with time courses ranging from tens of milliseconds (10–100 Hz) to minutes (∼0.01 Hz) or longer2–15, and therefore would also be expected to be unable to accurately follow high-frequency input. Restriction of responsiveness to a particular frequency range might be thought to limit function. Indeed, the inhibitory and neuromodulatory muscle innervation in several invertebrates may be present to increase the frequency range over which the muscles can accurately follow their inputs16–19. However, selectivity for low frequencies may also be advantageous because it allows receivers to demodulate, and hence respond to, low-frequency signals carried in a modulated high-frequency (carrier wave) input. The wide range of kinetics present in the responses of neurons and muscles to input suggests that these systems might have the cellular mechanisms necessary to perform such demodulation, and different patterns of temporal input can differentially alter various intracellular processes12,20–24. However, except for the exclusion of inputs with frequencies of 1 Hz or greater in vertebrate skeletal muscle1, no examples in which the signal of interest is carried in the low-frequency component of a broad-band input have been described in biological systems on the cellular level. We describe here three slow muscles driven by a bursting neural network with a cycle frequency of 1 Hz. This network’s nature neuroscience • volume 3 no 3 • march 2000

output is modulated by two other, much slower rhythmic neural networks. The muscles responded to these modulations and followed the slow network activity, even though no neuron of the slow networks innervates the muscles. These results extend the known range of functional low-frequency response some two orders of magnitude, and provide an example of a signal carried partially or exclusively in a slow modulation imposed on the target’s driver from elsewhere in the nervous system. Furthermore, two of these muscles are innervated by the same motor neurons and, hence, receive identical neuronal input. However, because of differences in their contractile properties, one muscle followed both slow networks, whereas the other followed only one. These two muscles responded to different frequency domains of one signal, and thus provide a clear example of both encoding of information by the nervous system and its decoding by neuronal followers, on the basis of frequency. Preliminary accounts of this work appeared in abstract form (L.G.M. et al., Soc. Neurosci. Abstr. 25, 1642, 1999; J.B.T. & S.L.H., Soc. Neurosci. Abstr. 25, 1641, 1999).

RESULTS We studied the pyloric neuromuscular system of the lobster (Panulirus interruptus), a part of the stomatogastric system that is driven by the pyloric neural network. Every 0.5 to 2 seconds, the motor neurons of this network produce 5–10 action potential bursts of 100–500 millisecond duration25. Pyloric muscles are non-spiking muscles that contract as a graded function of their neuronal input26. Based on the pyloric network’s cycle period and bursting nature, it was thought that each motor neuron burst induces a single muscle contraction that fully relaxes between bursts, and that the muscles thus temporally mirror the bursting activity of their input 25,27–30. Work on pyloric muscle contractions evoked by nerve stimulation in the crab, Cancer borealis, and the shrimp, Palaemon serratus, support this belief31–33. 245


articles

a PD neuron 15 mV 48 µm

Dorsal PD muscle

0.5 s

Contraction amplitude (µm)

b

Time (s)

However, in Panulirus some pyloric muscles contract and relax much too slowly to follow their motor neuron bursts faithfully34,35 (Fig. 1a). The top trace shows the characteristic rhythmic action potential bursts of a pyloric dilator (PD) neuron. The second trace shows an isotonic contraction (muscle shortening against constant load) of an extremely slow muscle innervated by the PD neurons (the dorsal PD muscle). This contraction was induced by stimulation of the motor nerve with parameters matching the first burst in the neuron trace. If the neuron were driving this muscle (in all experiments the motor nerve was cut to prevent spontaneous neuron input to the muscle), the next contraction would have occurred before the contraction induced by the previous burst had ended (arrow; angled to account for the long delay between the neuron burst and the contraction beginning). When the motor nerve was rhythmically stimulated with bursts of shocks using physiologically relevant parameters, the muscle contractions temporally summated (Fig. 1b). After the

1 PY neuron

10 mV

2 aln

3

10

Overall 7.5 spike 5 freq. (Hz) 2.5 0

4

0.15 mm

PY muscle

2s

246

c

Average amplitude (mm)


Fig. 1. In some pyloric muscles, overall spike frequency codes steadystate average contraction amplitude. (a) Top trace, intracellular PD neuron recording; bottom trace, isotonic contraction of an extremely slow muscle innervated by the PD neurons (the dorsal PD muscle) induced by motor nerve stimulation mimicking the first burst of action potentials (vertical lines below muscle trace) in the PD neuron trace. If the muscle were driven by the neuron, the next contraction (arrow) would occur before the first contraction finished, resulting in intercontraction summation. Most hyperpolarized PD neuron membrane potential, –62 mV. (b) Rhythmic stimulation (burst duration, 390 ms; spike frequency, 20 Hz; cycle period, 2 s) of the motor nerve innervating the dorsal PD muscle resulted in (after intercontraction summation had stabilized) a sustained tonic contraction on which rhythmic contractions rode. (c) After intercontraction summation had stabilized, overall spike frequency (spike number per burst divided by cycle period) coded average contraction amplitude (one half phasic plus tonic contraction) for this muscle. All data are from the cpv1b muscle39.

Overall spike freq. (Hz)

summation had stabilized, the muscle’s response therefore consisted of a sustained, tonic baseline contraction on which rode phasic contractions that matched the stimulation cycle period. Here the average contraction amplitude (one half the phasic contraction amplitude plus the tonic contraction amplitude) of the muscle is particularly important, and after muscle summation has stabilized, this average amplitude is well predicted by the overall spike frequency (spike number per burst divided by cycle period) of the motor neuron (Fig. 1c; ref. 34 and L.G.M. & S.L.H., Soc. Neurosci. Abstr. 24, 1891, 1998). These data suggest that if the overall spike frequencies of pyloric motor neurons were slowly modulated by extrinsic influences, average contraction amplitude of pyloric muscles might similarly vary. The stomatogastric nervous system also contains the gastric mill and cardiac sac networks, which have much slower cycle periods (5–10 seconds and ∼1 minute, respectively) than the pyloric network25. Gastric mill and cardiac sac activity modulates pyloric activity25,28,36–38, and the overall spike frequency of several pyloric neurons varies with gastric mill and cardiac sac activity (J.B.T. & S.L.H., Soc. Neurosci. Abstr. 25, 1641, 1999). Using stimulation parameters determined from previously measured effects of gastric mill and cardiac sac network activity on

Fig. 2. A muscle innervated by PY neurons expressed a gastric mill network contraction pattern even though no gastric mill neurons innervate the muscle. Trace 1, intracellular recording of the PY neuron whose activity was used to stimulate the muscle’s motor nerve. Most hyperpolarized PY neuron membrane potential, –55 mV. Trace 2, extracellular recording from the aln of the activity of the GM neurons of the gastric mill network. Trace 3, PY neuron overall spike frequency. Trace 4, isotonic contractions of a PY neuron-innervated muscle. Gray shading shows the tonic component of the muscle contraction; the muscle almost fully relaxed only at the end of each gastric mill cycle (dashed lines mark one cycle period). Solid horizontal line, fully relaxed muscle length. Recording is from p8 muscle39. nature neuroscience • volume 3 no 3 • march 2000


articles

1 2

Fig. 3. Although no cardiac sac neurons innervate it, the extremely slow dorsal PD muscle primarily contracted in a pattern matching the very slow cardiac sac rhythm. Trace 1, schematic representation of cardiac sac network activity; rectangles correspond to cardiac sac bursts, dashed lines mark one cardiac sac cycle period. Trace 2, intracellular recording of the PD neuron whose activity was used to stimulate the muscle’s motor nerve. Trace 3, time expansion of PD neuron activity immediately before, during and after a cardiac sac burst. Most hyperpolarized PD neuron membrane potential, –65 mV. Trace 4, PD neuron overall spike frequency. Trace 5, isotonic contractions of a dorsal PD muscle (inset is a time expansion showing small contractions in pyloric time). Scale bars, 20 s (traces 1, 2, 4 and 5) or 2 s (trace 3) and 5 mV (traces 2,3) or 60 µm (trace 5). Recording is from the cpv1b muscle39.

Cardiac sac

PD neuron

3

5 mV 2s


4

20 Overall 15 spike 10 freq. (Hz) 5 0

5 Dorsal PD muscle baseline (full relaxation)

60 µm 20 s

pyloric network output, we explored the effects of altering pyloric neuron activity on pyloric muscles by stimulating motor nerves innervating various pyloric muscles (Fig. 2). The first trace shows an intracellular recording from a pyloric (PY) motor neuron; the second trace is an extracellular recording from the anterior lateral nerve (aln), which carries the axons of the gastric mill (GM) neurons of the gastric mill network39. PY neuron firing waxed and waned as a function of gastric mill network activity (the dashed lines mark one gastric mill network cycle), and the third trace shows that this change in PY neuron activity resulted in corresponding changes in PY neuron overall spike frequency. The fourth trace shows the contractions of a muscle innervated by the PY neurons induced by motor nerve stimulation exactly matching the PY neuron spiking pattern shown in the first trace. As overall spike frequency increased at the beginning of each GM neuron burst, the phasic amplitude and summation of the pyloric-timed PY muscle contractions increased and, consequently, the muscle’s tonic contraction component (gray shading) increased. As overall spike frequency decreased during the GM neuron interburst interval, the muscle’s phasic contraction amplitude and summation and, thus, its tonic contraction, decreased, allowing it to return nearly to rest length before the next GM neuron burst. Thus, although no gastric mill neuron innervates this muscle, it nonetheless displayed both a pyloric (the rapid rhythmic contractions) and a gastric mill (the tonic contraction component) motor pattern. The changes in the PY muscle tonic contraction can be qualitatively understood by considering Fig. 1b and c. Contractions of the extremely slow dorsal PD muscle took 10–15 pyloric cycles to stabilize (Fig. 1b). The PY muscle is faster than the dorsal PD muscle, but it nonetheless took 2–3 cycles to stabilize in response to rhythmic stimulation and shows similar coding dependence (M. Rehn, L.G.M. & S.L.H., unpublished observations). When PY neuron overall spike frequency changes, the muscle’s contraction amplitude changes toward an average contraction appronature neuroscience • volume 3 no 3 • march 2000

priate for this new spike frequency. If overall spike frequency remained at the new value, the muscle would stabilize at the average contraction amplitude predicted by the equivalent of Fig. 1c for the PY muscle. However, overall spike frequency does not stabilize; consequently, the muscle is driven toward different regions of this curve each pyloric cycle, which generates the slow variations in average contraction amplitude (Fig. 2). For the relatively fast PY muscle, the contractions in pyloric time were a large percentage of total contraction amplitude. In contrast, the extremely slow dorsal PD muscle responded to PD neuron input modulated by the very slow, ∼0.01 Hz, cardiac sac network primarily with a slow, rhythmic variation in tonic contraction amplitude (Fig. 3). During cardiac sac network bursts (trace 1; each rectangle represents one cardiac sac burst, dashed lines mark one cardiac sac cycle), PD neuron activity was markedly altered, but the neuron continued to burst in pyloric time (trace 2; expanded over one cardiac sac burst in trace 3). PD neuron overall spike frequency (trace 4) showed a triphasic increase–decrease–increase pattern with each cardiac sac burst. When the motor nerve innervating the extremely slow dorsal PD muscle was stimulated with this activity pattern, the muscle’s contractions in pyloric time (shown on expanded time scale in inset) were only ∼8% of the maximum contraction amplitude, whereas its contractions in cardiac sac time were ∼50% of maximum contraction amplitude (trace 5). Thus, even though no motor neurons of the cardiac sac network innervate it39, the muscle primarily contracted in cardiac sac time. These data show that the PD and PY muscles can respond to slow modulations of PD and PY neuron activity. However, because the activity patterns of the PD and PY neurons are different, the PD and PY muscles receive different input; these data, therefore, do not show how different muscles respond to the same input. As well as innervating the extremely slow dorsal PD muscle, the PD neurons also innervate the faster ventral PD muscle, which allowed us to investigate whether the two muscles selectively respond to different frequency domains within an identical neural input. In addition to showing large (∼15 Hz) variations in overall spike frequency in time with the slow cardiac sac network (∼0.01 Hz; Fig. 3), the PD neurons often also showed small (∼2.5 Hz) variations in overall spike frequency (Fig. 4, trace 1) in time with the faster gastric mill network (0.1–0.2 Hz; trace 2; dashed lines mark one gastric mill cycle). The extremely slow dorsal PD muscles showed much smaller variations in gastric mill time than did the faster ventral PD muscles and, depending on gastric mill cycle period, sometimes completely excluded the gastric mill signal. In these cases (Fig. 4, trace 3), the 247


articles

1

Fig. 4. Different muscles can respond to different frequency domains in the same neural signal. The PD neurons innervate both the very slow dorsal PD muscle and the faster ventral PD muscle. PD neuron overall frequency (trace 1) varied as a function of gastric mill phase (trace 2, GM neuron activity recorded from the aln; dashed lines mark a single cycle period). Muscle contraction amplitude of the extremely slow dorsal PD muscle did not vary as a function of gastric mill phase (trace 3), whereas that of the faster ventral PD muscle did (trace 4). Trace 5, intracellular PD neuron recording used to stimulate the muscles’ motor nerves. Most hyperpolarized potential of PD neuron membrane,–65 mV. The dorsal PD muscle is the cpv1b muscle; the ventral is the cpv2b muscle39.

12.5 10 Overall spike freq. (Hz)

7.5 5 2.5 0

2 aln

3

Dorsal PD muscle


baseline (full relaxation)

0.2 mm

4 Ventral PD muscle

0.2 mm baseline (full relaxation)

5 PD neuron

5 mV 2s

output of the dorsal PD muscle consisted of a tonic contraction, on which were superimposed very small rhythmic contractions in pyloric time but no measurable contractions in gastric mill time. In contrast, the faster ventral PD muscle (trace 4) was rhythmically active in both pyloric and gastric mill time. Both muscles also showed large contractions during cardiac sac bursts (dorsal PD muscle, Fig. 3; ventral PD muscle, data not shown). Thus, the relatively fast ventral PD muscle was rhythmically active in pyloric (1 Hz), gastric (0.15 Hz) and cardiac sac (0.01 Hz) time, whereas the slow dorsal PD muscle responded to the rapid pyloric and very slow cardiac sac signals, but failed to respond to the intermediate gastric mill signal. Note also the sensitivity of the ventral PD muscle to input in gastric mill time; the small changes in PD neuron overall spike frequency (Fig. 4, trace 1) in gastric time were almost indistinguishable in the neuron intracellular recording (trace 5), but resulted in a pronounced (50% of maximum contraction amplitude) gastric mill component in the muscle contraction.

DISCUSSION These data demonstrate that three slow muscles of the pyloric neuromuscular system can respond to slow modulation of the rapid pyloric pattern, and hence express, either partially (the PY and ventral PD muscles) or almost exclusively (the dorsal PD muscle), the rhythmicity of networks of neurons that do not innervate the muscles. They further demonstrate that two targets (the slow ventral and faster dorsal PD muscles) can respond to different frequency domains in the same neuronal 248

input. These results have implications not only for interpreting studies of the pyloric system, but also, more generally, for understanding how and to what extent biological systems can respond to input signals carried in different frequency domains. Pyloric network function In response to bath application of various neuromodulatory substances 29,40–43 or stimulation of modulatory inputs25,28,29,37,38,44–48, the pyloric network produces different neural outputs. It is tempting to interpret these changes as inducing changes in a pyloric-timed motor pattern. The contraction of several pyloric muscles in phase with other, slower neural networks suggests, however, that changing pyloric neural activity will have complex effects on pyloric motor output. In particular, these data imply that understanding the functional effects of changes in pyloric neuron activity may require consideration not only of pyloric activity on a pyloric cycle-by-cycle basis, but also the effects of intercontraction summation over time scales 5- to 10-fold and 60- to 100-fold longer (corresponding to gastric mill and cardiac sac cycle periods, respectively). Pyloric cycle period varies between 0.5 and 2.0 seconds; depending on pyloric period, the relaxation times of some pyloric muscles may or may not allow intercontraction summation (T.A. Ellis, P.I. Harness, T.J. Koehnle, M. Rehn, L.G.M. & S.L.H., unpublished observations). For instance, the pyloric network shown in Fig. 2 had a period of ∼0.7 seconds. At a period of 2.0 seconds, this muscle’s contractions would nearly or fully relax between neuron bursts, intercontraction summation and tonic contraction amplitude would dramatically decrease, and the variations of tonic contraction amplitude in gastric mill time would be a much smaller component of total muscle contraction. Thus one effect of changing pyloric cycle period may be to alter the expression of gastric mill activity as a variation in tonic contraction amplitude; at very slow pyloric cycle periods, these variations could be completely abolished. Signal extraction from different frequency domains Considerable evidence indicates that neuronal intracellular processes are sensitive to the temporal patterning of the input the neuron receives. For instance, action potential patterning alters Ca2+ and second messenger concentration, protein expression and gene regulation12,20–23. The specificity of gene activation by Ca2+ oscillations depends on oscillation frequency24, and modeling studies suggest that multiple feedback systems sensitive to different time scales49 may be necessary to explain the response of stomatogastric neurons to temporal variations in input patterns50. Although these observations are intriguing, the functional significance of these data is unclear, as the expernature neuroscience • volume 3 no 3 • march 2000



articles

imental stimulation patterns used are not based on actual inputs. In contrast, our input exactly matched that produced by the system’s innervating neurons. More generally, previous experimental work on sensitivity to temporal patterns does not address the issue of whether particular second messengers selectively respond to different frequency domains. Our demonstration that two muscles responded to different frequencies in a single input (Fig. 4) directly addresses this point and demonstrates a differential response as a function of frequency domain. Presumably, this differential response largely stems from the different temporal properties of the muscles. Intracellular responses show a wide kinetic range2–15, and differential frequency response, therefore, may also be present in neuronal intracellular responses to broad-band input. Our data thus support the hypothesis that “these processes [intracellular messengers] may have...features...that limit their involvement to certain patterns of stimulation”12. The low-frequency response shown here does not require rhythmicity of the underlying input. Slow processes driven by irregular input would smooth out high-frequency variation and would respond primarily to the input’s average frequency10,14. However, if this average were slowly modulated, the system would be expected to follow the moving average and, hence, to respond to the modulation. Thus, one consequence of the distinct dynamics of various intracellular processes in a biological system may be that each process responds to specific frequency domains present in broad-band inputs. Furthermore, as expression of these processes can itself be regulated, cells may also be developmentally and functionally tuned to specific frequency domains contained in their input.

METHODS Stomatogastric neuromuscular systems were dissected and prepared for extracellular nerve recording and stimulation, intracellular neuron recording, and measurements of muscle contraction using standard techniques25,34,35,39. Nerve recording and stimulation were performed using suction electrodes or pin electrodes insulated with petroleum jelly, intracellular recordings were made with glass microelectrodes (filled with 0.55 M K2SO4, 20 mM KCl) and an Axoclamp 2A (Foster City, California), and contractions were measured with a Harvard Apparatus (South Natick, Massachusetts) 60-3000 isotonic muscle transducer. Muscle rest length was maintained at approximately physiological levels. Muscle loading was determined by inducing single contractions and adjusting muscle load to achieve the maximum contraction amplitude consistent with full relaxation after the stimulation. A support bar was then placed under the transducer arm to prevent muscle overstretching. In all cases, the lateral ventricular nerve, which contains both the PD and PY neuron axons, was stimulated to induce muscle contractions. There are two PD neurons and six to eight PY neurons. To ensure activation of all axons innervating a muscle, stimulation amplitude was incrementally increased using single-burst stimulations until increases in contraction amplitude ceased. No neuromodulatory axons are known to innervate the PD and PY muscles and no pyloric neurons are known to contain modulatory co-transmitters, Furthermore, comparisons of contractions before and after long stimuli revealed no modulation by the nerve. We therefore believe that our data reflect simple classical neuromuscular innervation and muscle contraction. Data were digitized with a Cambridge Electronic Design (CED, Cambridge, UK) 1401 plus and analyzed using CED and Kaleidagraph (Reading, Pennsylvania) software; figures were prepared in Canvas (Miami, Florida). In Fig. 1, the nerve was stimulated using a World Precision Instruments (WPI: Sarasota, Florida) stimulus isolation unit and a Grass (Quincy, Massachusetts) S48 stimulator; in Figs. 2–4, nerves were stimulated using a WPI Pulsemaster A300 stimulator driven by the CED. nature neuroscience • volume 3 no 3 • march 2000

ACKNOWLEDGEMENTS This work was supported by Ohio University, NSF and NIMH. We thank R.A. DiCaprio for discussion and advice and A. L. Weaver for setting up the CED to stimulator interface and writing scripts to transform CED events into waveforms used to drive the stimulator.

RECEIVED 2 JULY; ACCEPTED 20 DECEMBER 1999 1. Partridge, L. D. Signal-handling characteristics of load-moving skeletal muscle. Am. J. Physiol. 210, 1178–1191 (1966). 2. Smith, S. J., MacDermott, A. B. & Weight, F. F. Detection of intracellular Ca2+ transients in sympathetic neurones using arsenazo III. Nature 304, 350–352 (1983). 3. Gamble, E. & Koch, C. The dynamics of free calcium in dendritic spines in response to repetitive synaptic input. Science 236, 1311–1315 (1987). 4. Delaney, K. R., Zucker, R. S. & Tank, D. W. Calcium in motor nerve terminal associated with posttetanic potentiation. J. Neurosci. 9, 3558–3567 (1989). 5. Breer, H., Boekhoff, I. & Tareilus, E. Rapid kinetics of second messenger formation in olfactory transduction. Nature 345, 65–68 (1990). 6. Delaney, K. R. & Tank, D. W. A quantitative measurement of the dependence of short-term synaptic enhancement on presynaptic residual calcium. J. Neurosci. 14, 5885–5902 (1994). 7. Regehr, W. G., Delaney, K. R. & Tank, D. W. The role of presynaptic calcium in short-term enhancement at the hippocampal mossy fiber synapse. J. Neurosci. 14, 523–537 (1994). 8. Wang, S. S.-H., Alousi, A. A. & Thompson, S. H. The lifetime of inositol 1,4,5-triphosphate in single cells. J. Gen. Physiol. 105, 149–171 (1995). 9. Feller, M. B., Delaney, K. R. & Tank, D. W. Presynaptic calcium dynamics at the frog retinotectal synapse. J. Neurophysiol. 76, 381–400 (1996). 10. Helmchen, F., Imoto, K. & Sakmann, B. Ca2+ buffering and action potentialevoked Ca2+ signaling in dendrites of pyramidal neurons. Biophys. J. 70, 1069–1081 (1996). 11. Brezina, A., Orekhova, I. V. & Weiss, K. R. Control of time-dependent biological processes by temporally patterned input. Proc. Natl. Acad. Sci. USA 94, 10444–10449 (1997). 12. Fields, R. D., Eshete, F., Stevens, B. & Itoh, K. Action potential-dependent regulation of gene expression: temporal specificity in Ca2+, cAMP-responsive element binding proteins, and mitogen-activated protein kinase signaling. J. Neurosci. 17, 7252–7266 (1997). 13. Chatton, J.-Y., Cao, Y. & Stucki, J. W. Perturbation of myo-inositol-1,4,5trisphosphate levels during agonist-induced Ca2+ oscillations. Biophys. J. 74, 523–531 (1998). 14. Wang, X.-J. Calcium coding and adaptive temporal computation in cortical pyramidal neurons. J. Neurophysiol. 79, 1549–1566 (1998). 15. Pivovarova, N. B., Hangpaisan, J., Andrews, S. B. & Friel, D. D. Depolarization-induced mitochondrial Ca accumulation in sympathetic neurons: spatial and temporal characteristics. J. Neurosci. 19, 6372–6384 (1999). 16. Evans, P. D. & Siegler, M. V. S. Octopamine mediated relaxation of maintained and catch tension in locust skeletal muscle. J. Physiol. (Lond.) 324, 93–112 (1982). 17. Weiss, K. R. et al. Peptidergic cotransmission in Aplysia: functional implications for rhythmic behaviors. Experientia 48, 456–463 (1992). 18. Clarac, F. & Cattaert, D. Invertebrate presynaptic inhibition and motor control. Exp. Brain Res. 112, 163–180 (1996). 19. Stevenson, P. A. & Meuser, S. Octopaminergic innervation and modulation of a locust flight steering muscle. J. Exp. Biol. 200, 633–642 (1997). 20. Fields, D. R. & Nelson, P. G. Resonant activation of calcium signal transduction in neurons. J. Neurobiol. 25, 281–293 (1993). 21. Sheng, H. Z., Fields, R. D. & Nelson, P.G. Specific regulation of immediate early genes by patterned neuronal activity. J. Neurosci. Res. 35, 459–467 (1993). 22. Itoh, K., Stevens, B., Schachner, M. & Fields, R. D. Regulated expression of the neural cell adhesion molecule L1 by specific patterns of neural impulses. Science 270, 1369–1372 (1995). 23. Li, W., Llopis, J., Whitney, M., Zlokarnik, G. & Tsien, R. Y. Cell-permeant caged InsP3 ester shows that Ca2+ spike frequency can optimize gene expression. Nature 392, 936–941 (1998). 24. Dolmetsch, R. E., Xu, K. & Lewis, R. S. Calcium oscillations increase the efficiency and specificity of gene expression. Nature 392, 933–936 (1998). 25. Selverston, A. I., Russell, D. F., Miller, J. P. & King, D. G. The stomatogastric nervous system: Structure and function of a small neural network. Prog. Neurobiol. 7, 215–290 (1976). 26. Govind, C. K., Atwood, H. L. & Maynard, D. M. Innervation and neuromuscular physiology of intrinsic foregut muscles in the blue crab and spiny lobster. J. Comp. Physiol. A 96, 185–204 (1975). 27. Maynard, D. M. & Selverston, A. I. Organization of the stomatogastric ganglion of the spiny lobster. IV. The pyloric system. J. Comp. Physiol. A 100, 161–182 (1975).

249



articles

28. Sigvardt, K. A. & Mulloney, B. Sensory alterations of motor patterns in the stomatogastric nervous system of the spiny lobster Panulirus interruptus. J. Exp. Biol. 97, 137–152 (1982). 29. Kennedy, M. B. & Marder, E. in An Introduction to Molecular Neurobiology (ed. Hall, Z. H.) 463–495 (Sinauer, Sunderland, Massachusetts, 1992). 30. Turrigiano, G. G. & Heinzel, H.-G. in Dynamic Biological Networks. The Stomatogastric Nervous System (eds. Harris-Warrick, R. M., Marder, E., Selverston, A. I. & Moulins, M.) 197–220 (MIT Press, Cambridge, Massachusetts, 1992). 31. Meyrand, P. & Moulins, M. Myogenic oscillatory activity in the pyloric rhythmic motor system of Crustacea. J. Comp. Physiol. A 158, 489–503 (1986). 32. Meyrand, P. & Marder, E. Matching neural and muscle oscillators: control by FMRFamide-like peptides. J. Neurosci. 11, 1150–1161 (1991). 33. Jorge-Rivera, J. C. & Marder, E. TNRNFLRFamide and SDRNFLRFamide modulate muscles of the stomatogastric system of the crab Cancer borealis. J. Comp. Physiol. A 179, 741–751 (1996). 34. Morris, L. G. & Hooper, S. L. Muscle response to changing neuronal input in the lobster (Panulirus interruptus) stomatogastric system: Spike number vs. spike frequency dependent domains. J. Neurosci. 17, 5956–5971 (1997). 35. Morris, L. G. & Hooper, S. L. Muscle response to changing neuronal input in the lobster (Panulirus interruptus) stomatogastric system: Slow muscle properties can transform rhythmic input into tonic output. J. Neurosci. 18, 3433–3442 (1998). 36. Mulloney, B. Organization of the stomatogastric ganglion of the spiny lobster. V. Coordination of the gastric and pyloric systems. J. Comp. Physiol. A 122, 227–240 (1977). 37. Russell, D. F. & Hartline, D. K. A multiaction synapse evoking both EPSPs and enhancement of endogenous bursting. Brain Res. 223, 19–38 (1981). 38. Claiborne, B. J. & Selverston, A. I. Localization of stomatogastric IV neuron cell bodies in lobster brain. J. Comp. Physiol. A 154, 27–32 (1984). 39. Maynard, D. M. & Dando, M. R. The structure of the stomatogastric neuromuscular system in Callinectes sapidus, Homarus americanus, and Panulirus argus (decapoda crustacea). Philos. Trans. R. Soc. Lond. B Biol. Sci. 268, 161–220 (1974).

250

40. Beltz, B. et al. Serotonergic innervation and modulation of the stomatogastric ganglion of three decapod crustaceans (Panulirus interruptus, Homarus americanus, and Cancer irroratus). J. Exp. Biol. 109, 35–54 (1984). 41. Flamm, R. E. & Harris-Warrick, R. M. Aminergic modulation in the lobster stomatogastric ganglion. I. Effects on the motor pattern and activity of neurons within the pyloric circuit. J. Neurophysiol. 55, 847–865 (1986). 42. Hooper, S. L. & Marder, E. Modulation of the lobster pyloric rhythm by the peptide proctolin. J. Neurosci. 7, 2097–2112 (1987). 43. Nusbaum, M. P. & Marder, E. A neuronal role for a crustacean red pigment concentrating hormone-like peptide: neuromodulation of the pyloric rhythm in the crab, Cancer borealis. J. Exp. Biol. 135, 165–181 (1988). 44. Dickinson, P. S. & Nagy, F. Control of a central pattern generator by an identified modulatory interneurone in Crustacea. II. Induction and modification of plateau properties in pyloric neurones. J. Exp. Biol. 105, 59–82 (1983). 45. Nusbaum, M. P. & Marder, E. A modulatory proctolin-containing neuron (MPN). II. State-dependent modulation of rhythmic motor activity. J. Neurosci. 9, 1600–1607 (1989). 46. Katz, P. S. & Harris-Warrick, R. M. Neuromodulation of the crab pyloric central pattern generator by serotonergic/cholinergic proprioceptive afferents. J. Neurosci. 10, 1495–1512 (1990). 47. Norris, B. J., Coleman, M. J. & Nusbaum, M. P. Pyloric motor pattern modification by a newly identified projection neuron in the crab stomatogastric nervous system. J. Neurophysiol. 75, 97–108 (1996). 48. Blitz, D. M. et al. Different proctolin neurons elicit distinct motor patterns from a multifunctional neuronal network. J. Neurosci. 19, 5449–5463 (1999). 49. Liu, Z., Golowasch, J., Marder, E. & Abbott, L. F. A model neuron with activity-dependent conductances regulated by multiple calcium sensors. J. Neurosci. 18, 2309–2320 (1998). 50. Turrigiano, G., Abbott, L. F. & Marder, E. Activity-dependent changes in the intrinsic electrical properties of cultured neurons. Science 264, 974–977 (1994).



articles

Microsaccadic eye movements and firing of single cells in the striate cortex of macaque monkeys Susana Martinez-Conde, Stephen L. Macknik and David H. Hubel Dept. of Neurobiology, Harvard Medical School, 220 Longwood Avenue, Boston, Massachusetts 02115, USA


Correspondence should be addressed to S.M.-C. ([email protected])

When viewing a stationary object, we unconsciously make small, involuntary eye movements or ‘microsaccades’. If displacements of the retinal image are prevented, the image quickly fades from perception. To understand how microsaccades sustain perception, we studied their relationship to the firing of cells in primary visual cortex (V1). We tracked eye movements and recorded from V1 cells as macaque monkeys fixated. When an optimally oriented line was centered over a cell’s receptive field, activity increased after microsaccades. Moreover, microsaccades were better correlated with bursts of spikes than with either single spikes or instantaneous firing rate. These findings may help explain maintenance of perception during normal visual fixation.

Our interest in microsaccades and burst firing grew out of an attempt to study the activity of single cortical cells in an awake monkey during free viewing of a visual scene. We made continuous recordings of eye position while recording spikes from a cortical cell. We marked on a video screen the position of the eyes a suitable length of time before each spike (eye-positioncorrected reverse correlation). If the scene viewed by the monkey consisted, for example, of a large bright circle on a dark background, we expected the cell to fire whenever its receptive field was crossed by an appropriately oriented contour 1. To sample the scene evenly, we had the animal fixate on a spot whose position changed at random every few seconds. For some cells, we could see a clear correlation between the stimulus and the eye positions that were spike related, but for other cells, the constellation of spots marking spike-associated eye positions showed little or no relationship to the scene (Fig. 1a and b). During our recordings, we noticed that for certain gaze locations, cells tended to fire in bursts rather than trains of random spikes. When we filtered our records to examine only burst firing, we saw a much clearer correlation between the cells’ activity and the stimulus (Fig. 1c and d). Evidently, any tendency for cells to fire in high-frequency bursts was enhanced by the contours of the stimulus. This project stemmed in part from our suspicion that the bursts were produced largely in response to microsaccades that occurred during the intermittent periods of visual fixation. Here we examine the relationship between microsaccades and various firing patterns of V1 cells, specifically, single spikes, instantaneous firing rates and spike bursts of various lengths.

RESULTS We stimulated each V1 cell with an optimally oriented bright, stationary bar centered over the cell’s receptive field while the monkey fixated on a spot of light on another part of the screen. From simply listening to the cell’s firing while watching the eye position on the computer screen, it immediately nature neuroscience • volume 3 no 3 • march 2000

became clear that the cell’s firing and the monkey’s microsaccades were correlated, and that the eye movements tended to be associated with bursts of spikes. This paper is devoted to analyzing this relationship between microsaccades and spikes. The results are based on recordings from 258 cells in 3 monkeys. We asked how often, with or without a stimulus, a microsaccade was followed by spike activity and whether any suppression of activity accompanied or followed a microsaccade. We also asked the converse question: how often were spikes preceded by microsaccades, and how would the correlation be influenced by whether one looks at single spikes, instantaneous firing rates or bursts of spikes. To ask these questions meaningfully, we began by defining the terms ‘microsaccade’ and ‘burst’. Analysis of microsaccades In most studies of longer-range eye movements (movements outside the scope of this study), determining length, direction and velocity of the movements is relatively simple. Large saccades reach velocities in the hundreds of degrees per second, and so a straightforward velocity detector can be used to determine when and where each saccade begins and ends. Moreover, the starting and end positions of the eye movements are generally determined in advance. With microsaccades, starting and end positions and timing are relatively random, and their velocities are low. Previous studies identify microsaccades with instantaneous velocity thresholds2,3 set to fairly high levels (10° per second) to protect the analysis from noise. By using the fastest portion of the microsaccade to categorize them in terms of size and speed, these studies neglected portions of the microsaccade at speeds of less than 10° per second, a considerable portion of a microsaccade because of its slow speed and small size. In addition, these studies measured either the horizontal or vertical directional component of microsaccades, but not both. Thus the size of obliquely directional microsaccades was underestimated by as much as √2 (or 41% of the actual size). 251



articles

The problem is made more difficult by confusion as to what, exactly, constitutes a microsaccade. Our new algorithm depends on both freedom from bias as to microsaccade direction and accurate assessment of total size and speed of eye movements to help separate out drifts and blinks. Our algorithm therefore expresses eye position in polar coordinates, uses low velocity thresholds and is free of directional bias (see Methods). We defined microsaccades to be those eye movements that had a total angular subtense between three minutes and two degrees of arc, had an instantaneous velocity (measured each millisecond) that never fell below 3° per second and never varied from a straight trajectory by more than 15° in any direction. Microsaccades were recorded from two monkeys during several two-second intervals (Fig. 2).

Fig. 1. Responses of a cell located within the operculum of striate cortex to a white 8° circle. The upper left inset shows the fovea (small circle) in relationship to the cell’s receptive field. The fixation spot (fovea) is surrounded by a fixation window, 2° × 2°. (a, b) Each pixel represents the eye position associated with a single spike. (c, d) The same data, filtered so that each pixel represents a burst of 4 or more spikes occurring within an interval of 20 ms at a delay of 35 ms. (b, d) Same data as (a, c), except that the circle is removed to reveal the pixels. Duration of the recording was about 40 min.

Probability of discharges following microsaccades We set out first to determine whether microsaccades were correlated with modulations in neural activity, and if the modulations were stimulus driven or caused by the microsaccades themselves and unrelated to the visual stimulus. In the main portion of our analysis, we examined 246 cells from 2 monkeys. After the beginning of a microsaccade when an optimal stimulus was centered over the cell’s receptive field, average spike probability showed a clear but small increase of about 0.5%, peaking at about 70 ms (Fig. 3a, upper curve). The average did not show suppression of activity associated with the onset of microsaccades. We examined the individual

records to see if the effect of microsaccadic suppression had been drowned out by the averaging process, and found that only 6 of the 246 cells showed clear suppression of firing associated with microsaccade onset. In all six cells, the suppression preceded a period of excitation. Average probability of spikes recorded from 45 cells after microsaccades without any visual stimulus (the monitor was black except for the fixation point) revealed no correlation between the microsaccades and modulation of neural activity (Fig. 3a, lower curve), indicating that microsaccade-induced activity in V1 was caused by microsaccades driving V1 neurons by moving their receptive fields over stationary stimuli, and not by direct excitation from the motor system. We found similar increases in neural firing correlated with the end (as opposed to the beginning)

Fig. 2. Spikes and eye positions recorded during several two-second trials from two monkeys. In each panel, the upper (green) and middle (blue) traces represent x and y eye position (raw data); the lowermost record (red) represents amplitude of each microsaccade after processing to the same scale as the x and y plots. Each peak in the lower record thus represents a microsaccade. (Its height indicates the size of the microsaccade.) On the whole, there is a good agreement between the deflections in the eye position traces (green and blue) and the deflections in the microsaccade plot (red). Where possible discrepancies occur, we placed a red dot under the panel. Notice that frequency of eye movements can vary between monkeys. Vertical black lines represent spikes. 252



articles

a With stimulus

Spike probability

Without stimulus

Time after start of microsaccade (ms)

Microsaccade probability

of microsaccades. The result held true even when we correlated neural activity with each individual millisecond of each microsaccade in flight (data not shown). We ruled out the possibility that the observed excitation after microsaccades was caused by flicker on the stimulus monitor by recording from 12 additional neurons from a third monkey using a bar projected from a tungsten incandescent bulb as a visual stimulus. We again found excitation after microsaccades, ruling out the possibility that microsaccade-related activity was due to flicker in our video display. Probability that a spike was preceded by a microsaccade We next computed the reverse correlation between spikes and preceding microsaccades, that is, the probability of a microsaccade preceding the discharge for the 246 cells from the main portion of our study (Fig. 3b). The peak probability of spikes correlated with previous microsaccades was enhanced by 9%, many times the peak probability seen with the forward correlation (compare with Fig. 3a). We presume that the difference arose because the reverse correleation (Fig. 3b) considered only those microsaccades that were effective, whereas the forward correlation (Fig. 3a) included all microsaccades, regardless of their effectiveness. We considered whether a microsac- a cade would be rendered ineffective when its direction was parallel to the stimulus orientation. Microsaccades

Fig. 4. Instantaneous firingrate analysis. (a) Hypothetical train of spikes. The interspike intervals (ISIs) were calculated for each successive pair of spikes and expressed in milliseconds (below each pair). Spikes are represented as vertical lines. (b) Three-dimensional plot of the probability of a microsaccade before a pair of spikes, as a function of latency and ISI, for a single V1 cell. (c) Average probability of a microsaccade before a pair of spikes for all cells, as a function of ISI, at the average peak latency for the population of cells (65 ms). (d) Normalized contour plot of the data in (b).

Spike

Time before spike (ms)

perpendicular to the orientation of the cell’s receptive field might then preferentially activate the cell. We found, however, no clear relationship between microsaccade direction and response magnitude. This finding may have been due to the aperture problem: a V1 cell cannot distinguish between a slow microsaccade that moves the stimulus perpendicularly across its receptive field and a fast microsaccade that moves the stimulus in an oblique direction to the receptive field. Thus two unlikely events would need to occur simultaneously for microsaccade direction to be involved: the stimulus would need to be perfectly aligned with the angle of the receptive field, and the microsaccade direction would need to be perfectly parallel to the angle of the receptive field. In addition, it may be that only a fraction of microsaccades activated area V1 cells effectively because of imperfectly main-

Instantaneous firing-rate analysis (ISIs in ms)

b

Interval between spike pair (ms)

Time before spike pair (ms)

d

c Probability of microsaccade before spike pair


b

Probability of microsaccade before spike pair

Fig. 3. Correlation between microsaccades and spikes. (a) Probability of a spike after the start of a microsaccade. Starts of all microsaccades were aligned (vertical line). The probability of spikes after microsaccades increased in the presence of the stimulus, whereas it did not change if the stimulus was absent. (b) Probability that a microsaccade preceded any given spike. (All spikes were aligned at the vertical line.)

Int

Interval between spike pair (ms)


er v

al b pa etwe ir ( en ms sp ) ik

e

Tim

eb

r efo

es

pik

a ep

ir (

ms

)

Probability of microsaccade before spike pair

253


articles

a

Burst of 2

b

Burst of 1 (lone spike)

Lone spikes

Burst of 5

c

Bursts with 5 spikes

Probability of microsaccade before burst

Probability of microsaccade before burst


Fig. 5. Burst analysis. (a) The same train of spikes as in Fig. 4a, grouped into bursts. An ISI of 30 ms has been chosen arbitrarily for this example. Because the first spike in the train was separated from the second one by 20 ms, and this interval was smaller than 30 ms, both spikes were assigned to the same burst. The third spike occured 60 ms after the second spike, and so was part of another burst, and so on. (b) Three-dimensional plot of burst correlations using bursts of one spike (lone spikes) for the same V1 cell as in Figs. 4b and d. (c) Three-dimensional plot of burst correlations using bursts of five spikes for the same cell. The probability of a microsaccade before a burst was plotted as a function of latency (time before the first spike in each burst) and ISI. (d, e) Normalized contour plots of the data in (b, c). For this cell, bursts of five spikes were best correlated with previous microsaccades.

Burst analysis (with an ISI of 30 ms)

Bursts of spikes Because previous experiments (Fig. 1) suggested that bursts may convey more reliable e d information about the visual scene than single spikes1, we were especially interested in studying the relationship between bursts ) s s) m (m t( (of one or more spikes) and preceding t s r rs bu bu microsaccades. We assigned each spike e e r r fo fo IS IS be uniquely to a burst by determining the I be I e e m m Ti interval between it and the spikes precedTi ing and following: two spikes with a given ISI (or less) were thus considered members Probability of microsaccade before burst Probability of microsaccade before burst of a single burst, and two spikes with an interval greater than the given ISI were considered elements of separate bursts (S. M. Smirnakis, D. K. Wartained fixation: some microsaccades may have begun and ended land, M.J. Berry and M. Meister, Neural Information Coding with the stimulus outside the receptive field, without ever crossConference, Snowbird, Utah, 1997; Fig. 5a). The ISI chosen is cruing it. Such occurrences would diminish the average enhancecial in determining the overall distribution of burst lengths (numment of firing by microsaccades but not affect the incidence of ber of spikes per burst). Because we could not know in advance microsaccades preceding cell firing. the most appropriate ISI for any given cell, we compared, for all the records and for all the cells, all ISIs between 1 and 100 ms and Instantaneous firing rate measured the probability of a microsaccade occurring some time We examined the relationship between the interspike intervals (latency; ≤ 200 ms) before the first spike in each burst. We then (ISIs) that separated pairs of successive spikes (within the range sorted all bursts according to length (numbers of spikes per burst), of 1–100 ms) and the microsaccades that preceded these spike which ranged from 1 (lone spike) to 20. We then constructed 20 pairs by up to 200 ms (Fig. 4a). We define instantaneous fir(for each burst length) three-dimensional arrays; from each of ing rate as the inverse of interspike interval (1/ISI)4. Plots of these we plotted the probability of a microsaccade preceding a burst the probability of a microsaccade preceding the first of two against latency (≤ 200 ms) and ISIs (≤ 100 ms). These plots allowed successive spikes with a given ISI (Fig. 4b and d) demonstrate us to assess the probability of a microsaccade as a function of burst that microsaccades were more closely linked to high instantasize, latency and ISI. neous firing rates (small ISIs) than to slow rates. The probaWe show three-dimensional plots and normalized contour bility of a microsaccade was maximal about 44 ms before a plots for bursts of one (Fig. 5b) and five (Fig. 5c) spikes for the spike pair. At higher ISIs, the probability of a microsaccade same cell as in Figs. 4b and d. Burst of one, or lone spikes, were before a pair of spikes did not drop to zero, presumably because relatively poor indicators of previous microsaccades. Bursts of 5 of spontaneous firing. were more closely correlated with previous microsaccades in this We plotted average microsaccade probability against ISI cell, and had a peak correlation of almost one. Peak probability for the cell population by averaging all three-dimensional varied with latency in all cells, whereas peak latencies remained graphs and slicing the result parallel to the probability–ISI stable as burst size varied and were consistent with V1 responses. plane at the average peak latency, 65 ms (Fig. 4c). Pairs of For this cell, bursts of 5 spikes were the best indicators that a spikes were most closely related to preceding microsaccades microsaccade had occured about 44 ms before. when the interspike interval was short. Averaged across all We next compared the strengths of the associations for all the cells, the peak probability of a microsaccade preceding a pair different burst lengths examined, regardless of ISI or latency of spikes was 0.42. ISI

254

Time before burst (ms)

ISI

Time before burst (ms)



articles

Peak probability from firing-rate analysis


Peak probability of a microsaccade

a

Burst size (number of spikes)

b Probability of a microsaccade

(Fig. 6a; same cell). The optimal burst size was clearly five, with a fall-off for shorter and longer bursts. But why did the correlation decrease for bursts longer than five? The occurrence of longer bursts during the experiment that were not well correlated with microsaccades led us to wonder if larger eye movements omitted from our analysis were associated with longer bursts. To test this, we examined all eye movements longer than three minutes of arc and found a clear correlation with all bursts measured (data not shown). For the same cell, we then plotted probability against ISI for bursts of 1 and 5 spikes at a latency of 44 ms (the cell’s optimal latency; Fig. 6b). This amounted to cutting Fig. 5b and c by planes normal to the latency axis at 44 ms. For single spikes, it showed an almost constant probability of about 0.4. For bursts of five spikes, the probability variations were presumably explained by the tendency of spikes to occur in clumps separated by periods of relative quiescence, often lasting hundreds of milliseconds: when ISIs that defined bursts were small, they tended to cut these clumps into fragments that were poorly correlated with microsaccades. ISIs greater than the optimum value did not affect the probabilities unless sufficiently large that successive clumps coalesced. To sum up, burst size and latency were highly correlated with previous microsaccades, whereas the ISI used to define bursts was less crucial, at least for values greater than 20–30 ms. The correlation with preceding microsaccades was always higher for bursts, defined in terms of ISIs, than for pairs of spikes that determined instantaneous firing rates. This is demonstrated by plotting for all ISIs and latencies the peak probability of a microsaccade before the optimum burst size against the peak probability of a microsaccade before the optimum instantaneous firing rate (Fig. 7). The difference in the probabilities obtained by the two methods was significant (p < 0.001; t-test). Thus for every cell tested, burst configuration indicated a previous microsaccade with more reliability than the best possible instantaneous spike rate. From Fig. 6a, it is clear that a burst size of five supplied by far the best indicator of a previous microsaccade for the cell under consideration. We were curious to learn the variability of such a preference across our sample population.We therefore plotted

Peak probability from burst analysis


Fig. 6. (a) Peak probability that a microsaccade occurred before bursts, as a function of burst size, for all latencies and ISIs. Same cell as in Figs. 4 and 5. (b) Probability that a burst was preceded by a microsaccade, as a function of ISI, at the peak latency (44 ms; same cell as in a). The twodimensional plot was obtained from the three-dimensional graphs in Figs. 5b and c.

Bursts of 5 spikes

Single spikes

Tighter bursts

ISI used when assigning spikes to bursts

for all ISIs and latencies the distribution of burst sizes associated with peak probabilities (Fig. 8). Peak burst sizes showed a wide spread, with an average of 11 (± 5, s.d.) spikes per burst and a substantial fall-off for short bursts. The average latency associated with microsaccade activity was 65 ± 46 ms, and the average peak ISI was 56 ± 29 ms.

DISCUSSION The phenomenon of fading of a stabilized image comes as a great surprise to someone who learns of it for the first time; when first discovered5,6 in the early 1950s, it was little short of astonishing. Determined attempts to reproduce the fading by careful fixation often fail, especially for objects in or near the fovea, and observers are quite unaware of the microsaccades whose occurrence largely accounts for failure to perceive fading of stationary images. It was perhaps equally surprising to learn, in the late '50s, that cells in the mammalian primary visual cortex respond far more vigorously to moving stimuli than to stationary ones7,8. These first studies of visual cortex pointed out the possible relationship between this and the fading of stabilized images. Our sensory

Fig. 7. Peak probability that a spike was preceded by a microsaccade for each cell. Peak probabilities determined by instantaneous firing rate were plotted against determinations using optimum burst sizes, across all latencies and ISIs. For every cell, analysis from the burst pattern predicted preceding microsaccades better than determinations based on instantaneous firing rate. 255


Number of cells

articles

Fig. 8. Distribution of burst sizes yielding peak probabilities across the population of cells. Optimal burst sizes tended to be greater than three spikes.

systems seem to be organized on the principle that, whether we are predators or prey, changing stimuli are essential for survival and that sensory adaptation—the decline with time of a response to a maintained stimulus—guarantees that a novel stimulus will stand out, especially if it moves. The normal inability to see one’s own retinal blood vessels or optic disc is presumably also related to this process9. Ironically, the process works so well that some further mechanism seems to be required to overcome the fading, so that we can see a stationary scene despite adaptation. If involuntary eye movements during fixation are crucial role in maintaining the visibility of stationary scenes, we should be able to observe the relationship in awake alert animals by recording eye movements while observing the firing of single cells in the presence of an adequate visual stimulus. This was the purpose of the present study. Quantitatively correlating microsaccades and bursts of spikes has been a challenging problem. Under the term ‘bursts,’ we included both ‘clustered activity’ (short, high-frequency trains of two to Single session 239 sessions eight or more spikes) and ‘bursty’ firing, consisting of highly irregular longer-lasting periods of rapid activity separated by irregular silent periods. Both types of firing were described in 1958 using these terms 7 . Here, although recognizing that their origins may differ, we lumped the two together, simply referring to both as ‘bursts’. We found an excitatory effect of microsaccades on V1 cells. Such effects were seen in all 258 cells we examined. Correlating a cell’s activity with the beginning of the microsaccades (or its end, or any point between) showed that the probability of spike production increased when a stimulus, a stationary Microsaccade magnitude bar, was centered over the receptive (degrees of visual angle) field. No such correlation was seen in the absence of the stimulus (Fig. 3a), Fig. 9. The main sequence analysis. The left panel represents 523 microsaccades measured while studyand we conclude that microsaccade- ing a single neuron; it plots the linear regression of these microsaccades and 95% confidence intervals. related activity is a direct product of The right panel shows the linear regression of the main sequence from 691,899 microsaccades recorded visual-sensory mechanisms rather than from 239 cells from a single monkey (99% confidence intervals were obscured by the regression line). Microsaccade velocity (degrees per s)


Burst size yielding peak probability

a byproduct of oculomotor mechanisms. Further, its overall effect is excitatory. Our results are consistent with previous findings that much of the variability of responses in striate cortex of awake behaving monkeys during visual stimulation is due to the effects of microsaccades10. This is not to say that visual responses driven by microsaccades are likely to be restricted to V1 cells: one would expect to find similar results both earlier and later in the visual pathway. We were surprised to find so little evidence for microsaccaderelated suppression, as there is ample evidence of suppression in the much faster long-range saccades11–16. Visual responses in primate area V1 are suppressed when the receptive field of a single cell crosses a stationary bar during a saccade12–14. This suppression occurs only in cells that do not respond during fixation to stimuli moving at saccadic velocities. The difference between these results and ours is probably related to the ability of most neurons to respond to the stimulus velocities evoked by microsaccades (∼30° per s) while failing to respond to stimuli moving at large saccadic velocities (typically, over 300° per s). Hypothetically, suppression associated with microsaccades could block awareness of fixational eye movements, which produce displacements of images across the retina sufficient to be easily seen if produced by a moving object. We saw microsaccade-related suppression in only 7 of the 258 cells we studied, and the suppression was always followed by excitation. It is possible that our experimental protocol made excitation easier to detect than inhibition. In any case, we found no conclusive mechanism to explain unawareness of microsaccades, at least at the level of the striate cortex. The enhancement of firing in V1 cells by microsaccades that we observed is at odds with similar recordings of cells in monkey V1 during fixation with a small grating covering the receptive field 3. This study revealed enhancement of firing after microsaccades in only 6 of 35 cells, suppression of firing in 13 cells and no changes in the remaining 16 (but enhancement of firing downstream, in areas V2, V4 and IT). The failure to see microsaccade-related activation in V1 surprised us, given the marked elevation of firing rate we consistently saw whenever a stationary bar was placed in the receptive field and the clear asso-

256




articles

ciation, to simple inspection, between eye movements and firing. The discrepancy in results may be due to differences in methods. We stimulated our cells with stationary bars of light, whereas they used circular patches of oriented gratings that sometimes were binocularly rivalrous, with orthogonal orientations in the two eyes. Their cell population in V1 was much smaller, and they sampled eye position at 200 Hz, interpolating 80% of their eye movement data to simulate a sampling rate of 1 KHz. In addition, our monkeys fixated continuously during the entire recording session (nothing in our experimental protocol signaled the beginning or end of a trial to the monkey), so the visual stimulus remained on the screen throughout. Their monkeys, on the other hand, performed a visual discrimination task with discrete trials that were separated by blank screens. In the present analysis, we did not look for effects of slow drifts in eye position on the firing of cells, and cannot rule out a possible role for them in the prevention of image fading. The relative role of microsaccades and drifts has been a subject of considerable discussion in the literature. Several studies17,18 address both types of movement but do not isolate microsaccades and drifts from each other. The strongest claim against a role of microsaccades in preserving visual perception19 is mainly based on two ideas. First, visual thresholds may be elevated during saccades. This is true of long-range saccades, but the evidence for suppression during the much slower microsaccades is controversial, and we found little physiological evidence for it in the present study. Second, subjects can be trained to suppress their microsaccades without experiencing fading of images. The possibility that drifts alone can prevent fading does not prove that microsaccades play no part. Moreover, if one fixates steadily on a point while paying attention to a small (and, preferably, low contrast) object in the periphery of the visual field, the object will disappear almost immediately20. If such precise fixation eliminates microsaccades, this would seem to support their role in preventing fading. A careful analysis of the physiological effects of slow drifts in eye position would be useful, and we are currently attempting such an analysis. Having determined the probability that a microsaccade is followed by spike activity, our next step was to ask whether, conversely, neural discharges were associated with an increased probability of previous microsaccades. By doing so, we could concentrate on those microsaccades that had been effective in driving the cells but ignore ones that were ineffective. (The stimulus can at times drift away from the receptive field, among other possibilities.) We found that the probability of microsaccades before a spike (Fig. 3b) was much higher than the probability of a spike after microsaccades (Fig. 3a). The instantaneous firing rate of each cell, defined as the inverse of the interval between the two members of a pair of successive spikes, was still better than single spikes in indicating a preceding microsaccade. Furthermore, bursts of spikes were better correlated with preceding microsaccades than single spikes or instantaneous firing rates. For most cells, bursts of fewer than four spikes were less effective than longer bursts, and on average there was no sharp optimum size. Why did V1 cells show such a marked tendency to fire in bursts when microsaccades made their receptive fields move across a visual stimulus? Here we will not discuss the intrinsic properties of neurons that cause them to fire in bursts, but will instead consider the teleological significance of such firing. Are bursts useful in terms of enhancing the information conveyed by one cell to a postsynaptic cell? Specifically, are spikes arriving at a synaptic terminal more or less likely to evoke spikes in a postsynaptic cell if they occur in bursts? One can begin by asking whether the probnature neuroscience • volume 3 no 3 • march 2000

ability that a microsaccade precedes a single burst of arbitrary size (say of three spikes) is greater than the sum of the probabilities associated with three spikes considered separately. In our results, this seemed, on average, not to be so. For example, if a given spike was related to a previous microsaccade with a probability of over 0.3 (Fig. 3b), simple addition should predict that a rapid succession of three spikes should give a probability of three times 0.3, and on average this is roughly what we found. There may thus be nothing magic about bursts: by definition, several spikes in quick succession necessarily constitute a burst. We cannot, on the other hand, rule out the existence of presynaptic facilitation in some synapses. At a presynaptic level, both synaptic facilitation and depression are described in vitro21,22. Regardless of these effects, however, rapidly sequential EPSPs should sum postsynaptically, so that even when synaptic efficacy is reduced in size, as in presynaptic depression, multiple EPSPs should improve the chances of postsynaptic signaling through classical temporal summation. For a pair of spikes, a clear example of paired synaptic facilitation has been shown in vivo for retinogeniculate synapses23. Thus, although we have no direct evidence pertaining to cortical synapses, given the simplest assumptions of temporal summation, it seems reasonable to expect a burst of spikes to be more reliable than single spikes in influencing a postsynaptic cell. Equally interesting is the expectation that each microsaccade will produce synchronous bursts in many cells. Receptive fields of neighboring cells in striate cortex overlap extensively and share the same orientation preference. Consequently, as a stimulus is swept across their receptive fields, many cells should fire more or less synchronously, in bursts that can be expected to overlap in time. The result should be strong spatial summation of messages carried by axons converging at the next stage. Microsaccades would thus seem to represent an ingenious mechanism for enforcing and refreshing information coming from stationary visual stimuli. It is furthermore possible that microsaccades may be important for allowing us to use latency in our visual discriminations. Latency differences in neural responses are proposed as an encoding mechanism for changes in the magnitude of contrast24–26. But it has remained unclear how the brain could use latency information, as the brain cannot know the latency a priori. Microsaccades could theoretically be used by the visual system to measure latency, because the brain knows both when the microsaccade motor signal starts and when the visual signal arrives. Relative latencies of responses to stimuli could then be used to indicate contrast.

METHODS We used three rhesus monkeys, Macaca mulatta. A midline stainless-steel head post was mounted by screws over the skull, and a steel recording chamber was mounted over the occipital operculum. We recorded from single neurons in area V1 with lacquer-coated electropolished tungsten electrodes. A visual search coil attached to the sclera of one eye recorded the animals’ eye movements. Standard sterile surgical techniques and animal care methods are described1,27. The Harvard Medical Area Standing Committee on Animals (protocol #02935) approved all electrophysiological experimentation. Stimuli and receptive-field mapping. Monkeys were trained to fixate on a small spot displayed on a monitor at a distance of 58 cm. A monkey was required to keep fixation within a 2° window to receive a drop of juice; eye movements exceeding the window’s limits were also recorded. The stimuli displayed on the monitor had a luminance of 24.3 cd per m2, and the monitor background was 3.84 cd per m2. We recorded V1 over the occipital operculum (eccentricity, 6°–8°) or from the folds of calcarine cortex immediately beneath. Receptive fields were about 257


articles


0.33° × 0.5°, and thus the same size or larger than many eye movements during fixation. We first mapped the receptive field roughly by hand, using computer-generated moving slits whose orientation, size and rate of movement were controlled with a computer mouse. We then used a bar of optimum length, width and orientation that flashed in quick succession in various parts of the visual field to more precisely determine the position and size by reverse correlation. A stationary bar with these optimal parameters of width, length and orientation was then positioned over the center of the receptive field, and collected spikes were subsequently correlated with the monkey’s microsaccades. The bar turned on before data collection and remained on throughout the recording session for each cell. Within total recording times of 100–800 s (mean, 334 s per cell), we recorded eye movements and spikes in consecutive 2-s trials. The beginning and end of each trial were unknown to the animal, and the fixation point persisted between trials. Analysis of microsaccades. We developed an algorithm for determining when the eye was stopped and when it was moving. We regarded an eye movement as ended when either its speed dropped below some threshold or its direction changed by more than some predetermined angle. First, we digitally sampled the horizontal (x) and vertical (y) eye tracker voltages (with 11-bit resolution at a sampling rate of 1 kHz, each A/D unit corresponding to 0.982 arcmin of visual angle); we then created a table of change in x and y (dx and dy) computed as the difference between values at successive intervals. After differentiating the data, so that each element then represented the instantaneous velocity of the eye in horizontal and vertical space, we smoothed the processed data with a 31 ms-wide unweighted boxcar filter to reduce noise. Varying the width of this filter, we found that averaging each value of dx and dy over a period up to ±15 ms surrounding each sampling interval gave results consistent with categorization of microsaccades by inspection of raw data from the eye-movement monitor, the method used in most previous studies. We prepared an array of vectors representing the instantaneous direction (θ) and size (r) of motion of the eye at each 1-ms interval of the recording. Thus r may be thought of as the instantaneous velocity of the eye in degrees of visual space per ms. In addition to θ and r, we applied a velocity threshold of r < 3° per s as a measure of when the eye had stopped, giving a third datum for each interval: a value of 1 (stopped) or 0 (not stopped) was recorded to generate the ‘eye-stopped array’. At each interval, a rate-of-turn indicator checked whether θ had changed by >15°; if so, we considered the eye to have stopped between intervals and amended the eye-stopped array appropriately. This indicator was important for accuracy of the eye-stopped array, as the eye seldom had zero velocity even after the data had been smoothed—probably because of noise and drifts in eye position. During a true saccade, successive measurements of θ were similar (aligned in the direction of the overall eye movement). If θ continually changed, we considered saccades to be absent. To determine the overall distance and direction over an entire microsaccade (average, 29 ms), we used simple vector addition to generate cumulative distance and directions of the movement (integrated r and θ) at each interval between eye stops as registered in the eye-stopped array. At the final 1-ms interval of each microsaccade, values of integrated r and integrated θ were therefore equal to the final distance and direction of the entire eye movement. To distinguish microsaccades from small artifacts or large voluntary saccades or eye blinks, we applied lower and upper limits of 3 arcmin and 2°, to the final value of integrated r. Plots of the microsaccadic main sequence (size versus velocity) showed a linear regression (Fig. 9). Data processing. We calculated the reverse correlation for microsaccades preceding either pairs of spikes or bursts of spikes and plotted our results in three dimensions (Figs. 4 and 5). For particularly rare combinations of burst length and ISI, small sample size resulted in an artifactually high probability of the burst being preceded by a microsaccade; for instance,

258

if a particular burst size and ISI occurred only once and followed a microsaccade, the probability of such a unique event was thus 1. We therefore smoothed every three-dimensional array with a boxcar filter whose width was 15 elements in the ISI-latency plane; thus every probability was averaged with those corresponding to ±7 ms latency and ±7 ms ISI. (At the array edges, this boxcar was reduced to fewer elements.) This smoothing removed virtually all the random peaks of probability associated with rare bursts.

ACKNOWLEDGEMENTS We thank Gail Robertson, David Freeman, Michael Lafratta and Frederic Russo for technical assistance and Margaret Livingstone, Clay Reid, Richard Born and Max Snodderly for reading the manuscript and making comments. We also thank Murray Sherman, Jonathan Victor, John Maunsell and Guy Orban for their advice on the analysis and design of the project. This work was supported by research (to D.H.H.) and training grants (to S.L.M.) from the National Eye Institute. S.M.-C. is a fellow from the MEC-FPI program (Spain).

RECEIVED 25 MAY 1999; ACCEPTED 24 JANUARY 2000 1. Livingstone, M. S., Freeman, D. C. & Hubel, D. H. Visual responses in V1 of freely viewing monkeys. Cold Spring Harb. Symp. Quant. Biol. 61, 27–37 (1996). 2. Bair, W. & O’Keefe, L. P. The influence of fixational eye movements on the response of neurons in area MT of the macaque. Vis. Neurosci. 15, 779–786 (1998). 3. Leopold, D. A. & Logothetis, N. K. Microsaccades differentially modulate neural activity in the striate and extrastriate visual cortex. Exp. Brain Res. 123, 341–345 (1998). 4. Bair, W., Koch, C., Newsome, W. & Britten, K. Power spectrum analysis of bursting cells in area MT in the behaving monkey. J. Neurosci. 14, 2870–2892 (1994). 5. Riggs, L. A. & Ratliff, F. The effects of counteracting the normal movements of the eye. J. Opt. Soc. Am. 42, 872–873 (1952). 6. Ditchburn, R. W. & Ginsborg, B. L. Vision with a stabilized retinal image. Nature 170, 36–37 (1952). 7. Hubel, D. Cortical unit responses to visual stimuli in non anesthetized cats. Am. J. Opthalmol. 46, 110–122 (1958). 8. Hubel, D. H. & Wiesel, T. N. Receptive fields of single neurones in the cat’s striate cortex. J. Physiol. (Lond.) 148, 574–591 (1959). 9. Coppola, D. & Purves, D. The extraordinarily rapid disappearance of entoptic images. Proc. Natl. Acad. Sci. USA 93, 8001–8004 (1996). 10. Gur, M., Beylin, A. & Snodderly, D. M. Response variability of neurons in primary visual cortex (V1) of Alert Monkeys. J. Neurosci. 17, 2914–2920 (1997). 11. Dodge, R. Visual perception during eye-movements. Psychol. Rev. 7, 454–465 (1900). 12. Wurtz, R. H. Visual cortex neurons: response to stimuli during rapid eye movements. Science 162, 1148–1150 (1968). 13. Wurtz, R. H. Comparison of effects of eye movements and stimulus movements on striate cortex neurons of the monkey. J. Neurophysiol. 32, 987–994 (1969). 14. Wurtz, R. H. Response of striate cortex neurons to stimuli during rapid eye movements in the monkey. J. Neurophysiol. 32, 975–986 (1969). 15. Bridgeman, B., Hendry, D. & Stark, L. Failure to detect displacement of the visual world during saccadic eye movements. Vision Res. 15, 719–722 (1975). 16. Macknik, S. L., Fisher, B. D. & Bridgeman, B. Flicker distorts visual space constancy. Vision Res. 31, 2057–2064 (1991). 17. Gerrits, H. J. & Vendrik, A. J. The influence of stimulus movements on perception in parafoveal stabilized vision. Vision Res. 14, 175–180 (1974). 18. Sansbury, R. V., Skavenski, A. A., Haddad, G. M. & Steinman, R. M. Normal fixation of eccentric targets. J. Opt. Soc. Am. 63, 612–614 (1973). 19. Steinman, R. M., Haddad, G. M., Skavenski, A. A. & Wyman, D. Miniature eye movement. Science 181, 810–819 (1973). 20. Day, E. C. Photoelectric currents in the eye of the fish. Am. J. Physiol. 38, 369–398 (1915). 21. Magleby, K. L. in Synaptic Function (eds. Edelman, G. M., Gall, W. E. & Cowan, W. M.) 21–56 (Wiley, New York, 1987). 22. Zucker, R. S. Short-term synaptic plasticity Annu. Rev. Neurosci. 12, 13–31 (1989). 23. Usrey, W. M., Reppas, J. B. & Reid, R. C. Paired-spike interactions and synaptic efficacy of retinal inputs to the thalamus. Nature 395, 384–387 (1998).



articles

Lack of cortical contrast gain control in human photosensitive epilepsy


Vittorio Porciatti1, Paolo Bonanni2, Adriana Fiorentini1 and Renzo Guerrini2,3 1

Institute of Neurophysiology, Area Ricerca CNR, 1 via Alfieri, 56100 Pisa, Italy

2

Institute of Developmental Neurology and Psychiatry, University of Pisa and Institute for Medical Research IRCCS Stella Maris, 2 via Giacinti, 56018 Calambrone, Pisa, Italy

3

Division of Neurosciences, King’s College Hospital, University of London, Denmark Hill, London SE5 9RS, UK Correspondence should be addressed to V.P. ([email protected])

Television and video games may be powerful triggers for visually induced epileptic seizures. To better understand the triggering elements of visual stimuli and cortical mechanisms of hyperexcitability, we examined eleven patients with idiopathic photosensitive epilepsy by recording visually evoked potentials (VEPs) in response to temporally modulated patterns of different contrast. For stimuli of low–medium, but not high, temporal frequency, the contrast dependence of VEP amplitude and latency is remarkably abnormal for luminance contrast (black–white), but not so for chromatic contrast (equiluminant red–green) stimuli. We conclude that cortical mechanisms of contrast gain control for pattern stimuli of relatively low temporal frequency and high luminance contrast are lacking or severely impaired in photosensitive subjects.

Photosensitive epilepsy (PSE) is the most common form of stimulus-induced epilepsy1. Its prevalence in children 4–14 years old is substantial (0.5%–0.8%), and its incidence is increasing as a result of the proliferation of television display units and video games, which may act as triggers2. During a recent television showing of the ‘Pocket Monsters’ cartoon in Japan, 685 children experienced epileptic seizures3,4. The degree of danger inherent in such seizures ranges from nil to potentially life threatening, in exceptional cases5. Epileptic seizures induced by photic stimulation may be primarily generalized (tonic-clonic, myoclonic, absence seizures) or focal (occipital, with or without spreading to other cortical areas). PSE is defined as ‘pure’ when seizures are exclusively photic induced, and is usually idiopathic (with no other etiology than a genetic background)6. Because of the complexity and heterogeneity of video-generated patterns, reports of pattern-induced epilepsies are mainly anecdotal. Oriented lines are considered more powerful than checkerboards in inducing epileptiform EEG changes7, and oscillating patterns more epileptogenic than static patterns8. However, the characteristics of the visual stimulus and the cortical dynamics leading to the hypersynchronous neuronal response underlying PSE are poorly understood. A better knowledge of the triggering elements of visual stimuli might help in devising safer video-generated patterns. The temporal frequency of an intermittent photic stimulus is crucial for arousing epileptic activity. This suggests that, even for patterned stimuli, the temporal frequency may be critical. However, there is no systematic study on the visually evoked potentials to temporally modulated patterned stimuli in PSE patients. Another crucial parameter of pattern stimuli is spatial contrast (related to the spatial variations in brightness). It is conceivable that contrast is also a critical parameter for cortical excitability. nature neuroscience • volume 3 no 3 • march 2000

We studied both control subjects and patients with pure idiopathic PSE whose seizures originated from the occipital lobe9,10. VEPs were recorded in response to simple visual patterns (black/white and red/green sinusoidal gratings of different contrast), sinusoidally contrast-reversed at various temporal frequencies. The EEG activity remained in a physiological state throughout the experiment. In patients, the dependence of VEP amplitude on luminance-contrast for stimuli of relatively low temporal frequency (4–10 Hz) was dramatically altered. Specifically, at increasing contrast, the VEPs saturated in amplitude and shortened in phase in controls, whereas in patients there was little saturation or phase shift. These results suggest that cortical mechanisms of contrast gain control were lacking or severely impaired in PSE.

RESULTS Luminance gratings: effect of temporal frequency In both controls and patients, steady-state VEPs were recorded as a function of temporal frequency of luminance-contrast gratings. Stimulus contrast (90%) and spatial frequency (2 cycles per degree) were chosen to maximize response amplitude11,12. In agreement with previous studies11,12, the form of the temporal function consistently showed two amplitude peaks (at 3–8 Hz and 16–20 Hz) and a local minimum at 10–12 Hz in normal subjects (Fig. 1a and c). In PSE patients (Fig. 1b and c), the high temporal-frequency cutoff was comparable to that of controls. However, the shape of the temporal function was more variable. In particular, low- and high-frequency amplitude peaks were less defined, and in some subjects, substantial activity was present at intermediate frequencies (10–12 Hz), at which controls responded poorly. Overall, VEP amplitude was significantly larger in patients than controls (Fig. 1c; two-way ANOVA, F1,271 = 6.2, p = 0.012). 259


articles


a

b

c

contrast, and saturated at about 20% contrast (Fig. 2a). In patients, the average contrast function was virtually identical to that of controls in the low contrast range (3–20%; Fig. 2a), indicating comparable contrast threshold and contrast dependence at low contrast. Unlike controls, however, the response did not saturate with increasing contrast above 20%, and reached abnormally higher-amplitude values at maximum contrast (two-way ANOVA, F1,236 = 17.6, d e f p < 0.001). The difference in shape of the contrast function between the two groups was evaluated from individual curves (see below). At higher temporal frequencies (Fig. 2b), the VEP contrast function was comparable in controls and patients, and did not show amplitude saturation. Average VEP phase showed contrast dependence in the low temporal-freFig. 1. Luminance-contrast gratings: effect of temporal frequency. VEP amplitude and phase plots of quency range (Fig. 2c). Because VEPs individual control subjects (a, d) and patients (b, e). (c, f) Average (± s.e.) VEP amplitude and phase of were recorded at different temporal frecontrol subjects and patients are shown superimposed. quencies in individual subjects, data were normalized by transforming original phase values into latency values. Phase data were first divided by 4 (modulo = 2 π rad * 2nd harmonic) The VEP phase progressively lagged with increasing temporal and then multiplied by the stimulus period in milliseconds. frequency. Both raw data (Fig. 1d and e) and averaged data Latency values of responses to maximum contrast were normal(Fig. 1f) indicate that the phase plots were similar in controls and ized to zero-latency lag, and latencies for all other contrasts were patients. The response latency (in seconds) can be obtained from expressed as relative changes. Normalization of latency for the slope of the phase plot (π rad / Hz) divided by 4 (modulo = 2 responses to maximum contrast was possible, as VEPs of conπ rad * 2nd harmonic; see ref. 12). The average latencies, evaluattrols and patients showed comparable latencies at high contrast ed from individual phase plots, did not significantly differ between (see above). The VEP latency of controls progressively lagged controls and patients (106.7 ± 2.5 ms, s. e., versus 107.7 ± 2.2 ms). with decreasing contrast (Fig. 2c). In patients, however, the response latency did not slow down with decreasing contrast in Luminance gratings: effect of contrast the medium- to high-contrast range (10–90%). The different For each subject, contrast functions were evaluated at the lowlatency dependence between controls and patients was evaluated and high-frequency peaks of the individual temporal functions from individual curves (see below). (Fig. 1). In agreement with previous reports13, the average VEP In the higher temporal-frequency range (Fig. 2d), the conamplitude of controls progressively increased with increasing trast dependence of average VEP phases was virtually identical b a in controls and patients. The abnormal dependence of VEP amplitudes and phases on contrast in the low temporal-frequency range was very consistent among patients. We plotted individual response curves for both controls and patients (Fig. 3). The individual curves of most controls showed amplitude saturation at high contrasts (Fig. 3a), whereas this was not observed in the majority of patients (Fig. 3b). To provide a quantitative evaluation of saturation, a saturation index14 was derived for the individual amplitude curves. The saturation index is defined as (1 – c1/2)/c1/2 c d where c1/2 is the contrast that elicits responses with half the amplitude of the response at maximum contrast. A value of one corre-

Fig. 2. Luminance-contrast gratings: effect of contrast. Average (± s.e.) amplitude (a, b) and latency (c, d) for two different ranges of temporal frequencies. In the 4–10 Hz range, the contrast dependence differed remarkably between controls and patients at medium–high contrast. In the 16–22 Hz range, the contrast dependence of amplitude (b) and phase (d) was virtually identical in control subjects and patients. 260



articles

p < 0.001; PSE > controls). The average amplitude curves actually showed a slight difference in bandwidth (half-height width), but there was no significant temporal-free quency × group interaction (F 1,197 = 1.55, p = 0.14), possibly because of the larger variability of patient data. The VEP phase (Fig. 4c) lagged as a function of temporal frequency, with a comparable slope in d c controls and patients. Average latencies, evaluated from individual phase plots, tended to be longer in controls (142.2 ± 7 ms) than in patients (129.3 ± 6 ms). However, the difference was not significant. Contrast functions were evaluated at individual peak temporal frequencies (range, 3–6 Hz). In both controls and patients (Fig. 4b), the Fig. 3. Luminance-contrast gratings of 4–10 Hz temporal frequency: effect of contrast in individual sub- VEP amplitude progressively jects. Amplitude and latency plots of control subjects (a and c) and patients (b and d). (e) Scatter plot of increased with increasing contrast, saturation and latency indices (see text) for control subjects and patients. and tended to saturate at the highest (70%–90%) contrasts. The average saturation index of individual response curves was not significantly different between controls (2.1 ± 0.32) and patients sponds to lack of saturation. The average saturation index was (1.81 ± 0.34; t21 = 0.61, p = 0.54). The variation in response latenmuch higher in controls than in patients (13.45 ± 2.39 versus 2.8 ± 0.63; t21 = 3.91, p = 0.0008). To provide an index of latency cy (after normalization of phase data; see above) was contrast dependent (Fig. 4d). As for luminance-contrast stimuli, the VEP decrement with increasing contrast, the individual curves latency progressively lagged with decreasing contrast, with a (Fig. 3c and d) were linearly interpolated between 20% and maxsomewhat steeper slope for controls than for patients. Latency imum contrast. The average slope of the regression line (ms per decrement indices were obtained by linearly interpolating unit contrast) was found to be significantly steeper in controls than between 20% and maximum contrast individual curves. The in patients (19.52 ± 6.18 versus 1.64 ± 5.15; t21 = 2.16; p = 0.04). average slope of the regression line (ms per unit contrast) did In a scatter plot of saturation index and latency decrement not significantly differ between controls and patients (27.3 ± 11.4 (Fig. 3e), there was little overlap between the two groups, with versus 8.0 ± 8.9; t21 = 2.3; p = 0.2). patients showing less saturation and lower latency decrement than most controls. Interestingly, the two indices were correlated in controls (R = 0.72, p = 0.008) but not in patients (R = 0.12, p = 0.73). DISCUSSION The correlation between saturation and latency decrement is conIntermittent photic stimulation using a stroboscope is exploitsistent with proposed models of contrast gain control14–16. ed in the EEG laboratory to trigger photoparoxysmal responses in PSE patients, with 11–20 flashes per second representing the most Chromatic gratings: temporal frequency and contrast b a We explored whether steady-state VEPs to chromatic gratings were abnormal in PSE patients. Previous studies report failure of equiluminant red/green stimuli to induce paroxysmal EEG activity7. Temporal and contrast functions were evaluated for equiluminant red/green gratings (see Methods), and the results were summarized (Fig. 4). The temporal function for high contrast chromatic gratings (Fig. 4a) had a simple profile with an amplitude peak at about 5 Hz and rapid attenuation at higher frequencies11,17. Both the shape of the temporal function and the d c high-frequency cutoff were similar in PSE and controls, differing mainly in amplitude (two-way ANOVA, F 1,197 = 20.9,


a

b

Fig. 4. Chromatic-contrast gratings: effect of temporal frequency and contrast. Average (± s.e.) VEP amplitude and phase as a function of temporal frequency (a, c) and contrast (b, d). The shape of the temporal and contrast functions (a, b) was comparable in control subjects and patients. The slopes of the phase and latency plots (c, d) tended to be steeper in controls than in patients. nature neuroscience • volume 3 no 3 • march 2000

261



articles

effective frequencies2,18. Photoparoxysmal activity may also be provoked by viewing a striped black/white pattern19. Contrast values above 40%, a spatial frequency of 2–4 cycles per degree and 10–20 Hz reversal frequency are most epileptogenic19. Human photosensitivity is a form of reflex epilepsy arising in the visual cortex, with high tendency toward generalization19. The efficacy of the triggering stimulus depends on its ability to elicit action potentials in a hyperexcitable visual cortex, on synchronizing effects and on the size of the neural population activated8. The present findings, obtained with patterned stimuli of high contrast sinusoidally modulated in time, show that a range of relatively low temporal frequencies may be critical for cortical hyperexcitability. Indeed, the VEP responses of the patients in this temporal frequency range tend to be higher in amplitude in comparison with controls, in spite of most patients being treated with antiepileptic drugs, which may reduce the size of the VEPs2,21. To investigate the origin of the different VEP responses of patients and controls, we measured the amplitude and phase dependence of VEPs on stimulus contrast at two temporal frequency ranges, 4–10 Hz and 16–22 Hz. Whereas the VEP dependence on contrast is comparable for patients and controls at higher temporal frequency, there is a clear difference at the lower temporal frequency. The amplitude saturation at high contrasts and the phase advance with increasing contrast typically found in normal subjects22 were absent or much reduced in patients. Interestingly, the contrast threshold and the contrast dependence in the low contrast range were not affected in PSE. The critical range of reversal of square-wave patterns reported to elicit epileptic EEG activity (10–20 reversals per second)19corresponds to a fundamental temporal frequency of 5–10 Hz, in agreement with our findings of sine-wave temporal modulation. Single cells of the primary visual cortex of the cat and monkey show response saturation and phase advance in response to drifting sinusoidal gratings of increasing contrast14,23. These nonlinear response properties could result from a contrast gain control stage that scales the input contrast taking into account the average contrast in the surround16 or the average activity of a large population of surrounding cells14,24. In the cat visual cortex, removing inhibitory effects by local application of bicuculline abolishes VEP response saturation and reduces phase shift, sug-

gesting that a mechanism normally responsible for contrast gain control has been suppressed25. Our VEP data do not allow us to advance suggestions about the site and nature of the gain-control mechanism14–16,26. Note, however, that the pattern ERG in response to both luminance and chromatic gratings does not show significant saturation and phase shift with increasing contrast27. The average saturation index of VEP amplitudes and the average latency decrement with increasing stimulus contrast in controls are in agreement with data obtained from simple cortical cells under similar stimulus conditions14. This encourages us to interpret the control data as indicating a mechanism of contrast gain control and to speculate that such a mechanism is defective or absent in the visual system of PSE patients. VEP responses at the higher temporal frequencies, at which normal subjects did not show amplitude saturation, were unaffected in PSE. VEPs to high-contrast red/green equiluminant gratings tended to be larger than normal and showed a smaller phase advance in PSE, although saturation indices were not different from those for controls. The present findings do not contradict the previous report that stationary equiluminant patterns are ineffective in inducing photoparoxysmal EEG activity7. In conclusion, our results indicate that pattern stimuli of relatively low temporal frequency and high contrast may be particularly effective in uncovering cortical hyperexcitability in PSE, possibly because of an impairment of contrast gain-control mechanisms normally present at these temporal frequencies. Highcontrast stimuli endowed with such temporal characteristics are common in TV images and in video games, and may be important in triggering the abnormal cortical response underlying visually induced epileptic seizures.

METHODS

Subjects. Subjects were 12 normal volunteers (mean age, 15.2 ± 3.2 years; n = 6 males) and 11 adolescents and young adults (mean age, 18 ± 3 years; n = 3 males) with PSE9 (Table 1). All patients showed high-amplitude VEPs (with normal waveform) to both bright flashes and checkerboard patterns21. Nine patients were being treated with antiepileptic drugs during the study. Antiepileptic drugs (especially VPA) attenuate the photoparoxysmal response and may reduce the VEP amplitude2,21. These effects are interpreted as resulting from reduced spread of cortical activity rather than alteration of mechanisms responsible for its generation28. EEG monitoring throughout the VEP sessions did not show epileptiform

Table 1. Details of the 11 photosensitive patients. Gender/age M/19 F/20 F/22 F/21 F/13 F/21 M/19 F/21 F/19 F/14 M/15

Seizure types CPS CPS SPS + II gen SPS + II gen SPS CPS + II gen

AEDs at VEPs examination VPA no CBZ CBZ VPA VPA, PB

SPS + II gen CPS CPS SPS + II gen CPS

no VPA, CBZ VPA VPA VPA

Reported enviromental triggers bright sunlight TV TV, computers TV TV, bright sunlight TV, bright sunlight, video games video games TV, bright sunlight TV TV TV, highly contrasted patterns

Photosensitivity range (frequency) IPS Pattern (reversal rate) 10–30 Hz 2–20 Hz 18–25 Hz NP 15–25 Hz 10–18 Hz 15–18 Hz NP 12–30 Hz NP 15–40 Hz 2–18 Hz 12–21 Hz 5–15 Hz 13–18 Hz 15–21 Hz 4–50 Hz

10–20 Hz 5–20 Hz 10–18 Hz 10–20 Hz 4–18 Hz

AEDs, antiepileptic drugs; VEPs, visual evoked potentials; Y, years; IPS, intermittent photic stimulation; M, male; CPS, complex partial seizure; VPA, valproic acid; Hz, hertz; F, female; TV, television; NP, not performed; SPS, simple partial seizure; II gen, secondary generalization; CBZ, carbamazepine; PB, phenobarbital.

262



articles


activity. Experiments followed the tenets of the Declaration of Helsinki. Informed consent was obtained after the nature of the technique and the aim of the research were explained. Visual stimuli. Methods for generating luminance contrast (black–white) and chromatic contrast (red-green) stimuli are reported in detail elsewhere17,29. Briefly, stimuli were horizontal sinusoidal gratings of 2 cycles per degree and different Michaelson contrast (3–90%, in half-octave steps), sinusoidally reversed in contrast at 3–24 Hz. Both luminance and chromatic gratings were generated by a VSG/2 graphic card (Cambridge Research System, Cambridge, UK) and displayed on a gamma-corrected color monitor (Barco CCID 7751, Kortrijk, Belgium). To establish equiluminance, subjects viewed a red/green grating of 50% contrast alternating at 15 Hz, and adjusted the relative luminance of red and green30 to minimize perception of flicker. Chromatic patterns were viewed through yellow filters (Kodak Wratten 16, Rochester, New York) to attenuate wavelengths lower than 500 nm. CIE coordinates (evaluated by a Minolta Chromameter CS 100, Osaka, Japan) were x = 0.637, y = 0.362 (red); x = 0.416, y = 0.582 (green). Visual stimuli had a mean luminance of 14 cd per m2 and subtended an area of 14° × 14° at the viewing distance of 100 cm. Subjects monocularly fixated a black spot at the center of the screen. Stimuli in a comparable spatiotemporal range provoked abnormal EEG activity under diagnostic investigation (Table 1). To prevent triggering abnormal EEG activity during the experiment, we limited stimulation to one eye and to a relatively small area of the visual field. Moreover, abrupt changes in retinal stimulation were avoided using sinewave contrast reversal and horizontal gratings that reduced the effects of saccadic eye movements. VEP recording. Steady-state VEPs were recorded from scalp electrodes (resistance < 5 kΩ) placed over the occipital (Oz) and frontal (Fz) cortices, and referenced to the vertex (Cz). The left mastoid was grounded. Signals were filtered (0.3–100 Hz, –6 dB per octave) amplified (50,000fold), digitized (2 kHz, 12-bit resolution) and averaged (at least 160 sums), with rejection of signals exceeding a threshold voltage. This assured that single sweeps occasionally contaminated by epileptiform spikes did not contribute to the averaged records. To reduce patients’ exposure time to the patterns, the stimuli were presented for a short time (corresponding to 20 sweeps) and intermingled with brief interruptions. Partial averages (20 sums) of the total averages were used to evaluate response consistency11; these were of the same order in patients as in controls. Averaged VEPs were submitted to DFT analysis to evaluate the second-harmonic (major response component) amplitude and phase11. Responses to patterns of zero contrast were frequently recorded to obtain a measure of residual noise (average, 0.3 µV). Responses recorded simultaneously from Fz were close to the noise level and are not shown.

ACKNOWLEDGEMENTS The authors would like to thank M. C. Morrone for suggestions and discussion and C. Orsini for technical help.

RECEIVED 1 DECEMBER 1999; ACCEPTED 19 JANUARY 2000 1. Kasteleijn-Nolst Trenité, D. G. A. in Reflex Epilepsies and Reflex Seizures. Advances in Neurology (eds. Zifkin, B. Andermann, F. Beaumanoir, A. & Rowan, J.) 99–121 (Lippincott-Raven, Philadelphia, 1998). 2. Harding, G. F. A. & Jeavons, P. M. Photosensitive Epilepsy (Mac Keith, London, 1994).


3. Sullivan, K. Perplexity and outrage over TV show. 200 remain hospitalized in Japan after seizures from viewing cartoon. The Herald International Tribune December 19, 5 (1997). 4. Ishida, S. et al. Photosensitive seizures provoked while viewing “Pocket Monsters” a made-for-television animation program in Japan. Epilepsia 39, 1340–1344 (1998). 5. Christie, T. Death after an attack precipitated by playing Nintendo game. The Sun January 21, 5 (1993). 6. Commission on Classification and Terminology of the International League Against Epilepsy. Epilepsia 30, 389–399 (1989). 7. Wilkins, A. J., Darby, C. E. & Binnie, C. D. Neurophysiological aspects of pattern-sensitive epilepsy. Brain 102, 1–25 (1979). 8. Binnie, C. D., Findlay, J. & Wilkins, A. J. Mechanisms of epileptogenesis in photosensitive epilepsy implied by the effects of moving patterns. Electroenceph. Clin. Neurophysiol. 61, 1–6 (1985). 9. Guerrini, R. et al. Idiopathic photosensitive occipital lobe epilepsy [published erratum appears in Epilepsia 37, 310, 1996] Epilepsia 36, 9, 883–91 (1995). 10. Guerrini, R., Bonanni, P., Parmeggiani, L. & Belmonte, A. Adolescent onset of idiopathic photosensitive occipital epilepsy after remission of benign rolandic epilepsy. Epilepsia 38, 777–781 (1997). 11. Porciatti, V., Burr, D. C., Morrone, C. & Fiorentini, A. The effects of ageing on the pattern electroretinogram and visual evoked potential in humans. Vision Res. 32, 1199–1209 (1992). 12. Simon, F. The phase of PVEP in Maxwellian view: influence of contrast, spatial and temporal frequency. Vision Res. 32, 591–599 (1992). 13. Campbell, F. W. & Maffei, L. Electrophysiological evidence for the existence of orientation and size detectors in the human visual system. J. Physiol. (Lond.) 207, 635–652 (1970). 14. Carandini, M., Heeger, D. J. & Movshon, J. A. Linearity and normalization in simple cells of the macaque primary visual cortex. J. Neurosci. 17, 8621–8644 (1997). 15. Shapley, R. M. & Victor, J. D. How the contrast gain control modifies the frequency response of cat retinal ganglion cells. J. Physiol. (Lond.) 318, 161–179 (1981). 16. Geisler, W. S. & Albrecht, D. G. Cortical neurons: isolation of contrast gain control. Vision Res. 8, 149–1410 (1992). 17. Fiorentini, A., Porciatti, V., Morrone, M. C. & Burr, D. C. Visual ageing: unspecific decline of the responses to luminance and colour. Vision Res. 36, 3557–3566 (1996). 18. Newmark, M. E. & Penry, J. K. Photosensitivity and Epilepsy (Raven, New York, 1979). 19. Binnie, C. D. & Wilkins, A. J. in Reflex Epilepsies and Reflex Seizures. Advances in Neurology (eds. Zifkin, B., Andermann, F., Beaumanoir, A. & Rowan, J.) 123–138 (Lippicott-Raven, Philadelphia, 1998). 20. Binnie, C. D. in Epileptic Seizures and Syndrome (ed. Wolf, P.) 49–54 (John Libbey, London, 1994). 21. Guerrini, R. et al. in Reflex Epilepsies and Reflex Seizures. Advances in Neurology (eds. Zifkin, B., Andermann, F., Beaumanoir, A. & Rowan, J.) 159–178 (Lippincott-Raven, Philadelphia, 1998). 22. Spekreijse, H., Van Der Tweel, L. H. & Zuidema, T. Contrast evoked responses in man. Vision Res. 13, 1577–1601 (1973). 23. Albrecht, D. G. & Hamilton, D. B. Striate cortex of monkey and cat: contrast response function. J. Neurophysiol. 48, 217–237 (1982). 24. Carandini, M. & Heeger, D. J. Summation and division by neurons in primate visual cortex. Science 264, 1333–1336 (1994). 25. Morrone, M. C., Burr, D. C. & Speed, H. D. Cross-orientation inhibition in cat is GABA mediated. Exp. Brain Res. 67, 635–644 (1987). 26. Bernardete, E. A. & Kaplan, E. The dynamics of primate M retinal ganglion cells. Vis. Neurosci. 16, 355–368 (1999). 27. Morrone, M. C., Porciatti, V., Fiorentini, A. & Burr, D. C. Pattern electroretinogram in response to chromatic stimuli: I Human. Vis. Neurosci. 11, 861–871 (1994). 28. Wilkins, A. J., Binnie, C. D., Darby, C. E. & Kasteleijn-Nolst Trenité, D. G. A. in Reflex Seizures and Reflex Epilepsies (eds. Beaumanoir, A., Gastaut, H. & Naquet, R.) 153–162 (Editions Médicine & Hygiène, Geneva, 1989). 29. Porciatti, V., Di Bartolo, E., Nardi, M. & Fiorentini, A. Responses to chromatic and luminance contrast in glaucoma: a psychophysical and electrophysiological study. Vision Res. 37, 1975–1987 (1997). 30. Mullen, K. T. The contrast sensitivity of human colour vision to red-green and blue-yellow chromatic gratings. J. Physiol.(Lond.) 359, 381–400 (1985).

263


articles

Learning to find a shape M. Sigman and C. D. Gilbert The Rockefeller University, 1230 York Avenue, New York, New York 10021-6399, USA Correspondence should be addressed to C.D.G. ([email protected])


We studied the transition of stimuli from novel to familiar in visual search and in the guidance of attention to a particular object. Ability to identify an object improved dramatically over several days of training. The learning was specific for the object’s position in the visual field, orientation and configuration. Improvement was initially localized to one or two positions near the fixation spot and then expanded radially to include the full area of the stimulus array. Characteristics of this learning process may reflect a shift in the cortical representation of complex features toward earlier stages in the visual pathway.

In a visual search task, a target must be detected within a field of distractors. The target can be defined by various attributes, such as color, orientation or form. Depending on the combination of target and distractors, search efficiency may be influenced by the number of distractors1–4. According to feature integration theory, the difficulty of visual search is determined by a target’s uniqueness in the map of some elementary feature1. Visual input is processed in two stages. The first stage uses a set of retinotopically organized maps coding for an elementary attribute such as color or orientation. This stage operates in parallel across visual space, but it produces no information about conjunctions of elementary features. Detection of conjunctions represents a second stage that operates serially to produce the percept of a whole object. This theory considers a shape to be a conjunction of elementary features or strokes3. Identifying a shape would therefore require serial search, and the ability to identify it should diminish with the number of distractors. However, this degradation of performance can be strongly counteracted by familiarity, even when the lowlevel features of targets and distractors are held constant. For instance, recognizing the digit 2 among an array of the digit 5 becomes much harder when rotating the entire image by 90° renders the characters less familiar5. Under some circumstances, performance in a visual search task can also be improved by priming. The effect of priming is thought to be limited to simple visual attributes and is passive and automatic6,7. Here we show that perceptual learning extends the range of priming effects and is important in the ability to guide attention to a particular object. We chose a visual search task that involved searching for a triangle (a target defined by form) among an array of distractors, in this case triangles of other orientations (Fig. 1a). We show that perceptual learning dramatically increases the ability to find a shape. Moreover, we show specificity of this learning for visuotopic position, object orientation and object configuration.

RESULTS Performance before training We used a visual search task in which the observer was required to find a target embedded in an array of distractors. The target consisted of a triangle of one of four possible orientations (up, left, right or down), surrounded by triangles of the other three orientations. The triangles were presented in a 5 × 5 stimulus array with a cen264

trally positioned fixation spot. A screen, in which the target was either present in a randomized position or absent, was presented every 3 seconds for a duration of 300 milliseconds (Fig. 1a). The subject’s task was to report whether the target was present. We measured the percent of correct responses for a fixed presentation time. To compensate for guessing, 20% of the trials were a null condition in which no target was present. A separate false-positive rate was calculated for each experiment. This false-positive rate, fp, was used to adjust the percentage of positive responses, p, according to the formula p′ = (p – fp)/(1 – fp) to yield p′, the true-positive rate, which we averaged for all subjects performing each experiment. The falsepositive rate was below 3% for all subjects. The tests of significance were carried out using a two-tailed t-test over the data collected from all subjects; error bars correspond to standard deviations. In certain types of search tasks, the target attracts attention even when the observer has no knowledge about its characteristics. For example, even without a previous cue, a red object embedded in an array of blue distractors will draw the viewer’s attention6. Our search task, however, required that the viewer have explicit knowledge of the object sought. Subjects could not perform the task unless they were instructed to find a triangle of a particular orientation. Experiments were run in blocks of 150 trials with a target of a single orientation. Before each block, subjects were informed of the orientation of the target. Each session consisted of eight different blocks. In two sessions before training, performance levels on detecting triangles of the four different orientations were tested. Naive subjects showed an average performance below 20% for all different orientations. Effects of training After having measured performance levels before training, we chose a single orientation, and the subject was trained by repeating blocks in this particular orientation. Different orientations were used as targets for different subjects. All subjects substantially improved performance over the training period. Training stopped when subjects reached threshold, which we arbitrarily set in the range of 70–80% correct responses (Fig. 1b). For different subjects, the time to reach threshold varied between 4 and 6 days, corresponding to 5000–7000 trials. To measure the change in performance for the trained and untrained orientations, we then repeated the two test sessions in which subjects were tested on triangles of the four orientations. nature neuroscience • volume 3 no 3 • march 2000


articles

a

Percent correct

b

Days trained

Before training

d

Target (present)

Before training

After training

Trained orientation Untrained orientation

Before training

Subjects showed an average 5-fold increase of performance in detecting triangles of the trained orientation (p′ = 15.4 ± 5.3% before training; p′ = 74.0 ± 2.9% after training; significance, p < 10–6) but no significant increase in detection of triangles of the untrained orientations (p′ = 19.5 ± 4.7% before training, p′ = 21.3 ± 5.0% after training; significance, p > 0.3, average over 4 subjects; Fig. 1c). Once training for one orientation was completed, we waited one month and repeated the test session to examine retention of the improvement. There was no change in the training effect after the 1-month hiatus (p ′ = 74 ± 2.5% after learning and p′ = 77 ± 1.9% after hiatus, average over 2 subjects, p > 0.2), indicating that the improvement showed no extinction over time (Fig. 1d). After training on triangles of a second orientation, however, performance on the initially trained orientation declined, dropping from p′ = 74.0 ± 2.9% after the first training to p′ = 57.0 ± 3.5% after subjects were trained for 7 days in a second orientation, averaged over 2 subjects (significance, p < 0.05; Fig. 1d). Spatial dependence of the learning The results above were averaged over all spatial locations at which the stimuli appeared. One can examine the visuotopic specificinature neuroscience • volume 3 no 3 • march 2000

After training

Percent correct


Target (not present)

Trained orientation Untrained orientation

Percent correct

c

Fig. 1. Training on triangles of a particular orientation resulted in improvements in detection specific to the object at the trained orientation. (a) A stimulus consisting of a 5 × 5 array composed of triangles of 4 possible orientations (right, left, up or down) was presented for 300 ms. The target, a triangle of particular orientation, was present in 80% of the cases. The distractors were triangles in the three remaining orientations. (b) One subject’s progress through the course of learning. (c) Averaged responses (four subjects) for the trained orientation. Performance improved fivefold after training. No change was seen for the untrained orientations. (d) Improvement (averaged over two subjects) lasted for at least one month without practice with no degradation in performance. Two subjects were trained on a second orientation; subsequently, performance for the first trained orientation degraded, reflecting a negative-transfer function in the orientation domain.

After 1 month after After training on a training training second orientation

ty of the training effect by comparing the change in performance at specific locations within the array. In particular, we wanted to determine if the learning occurred sequentially in different locations of the visual field or if the improvement resulted from a globally and uniformly increased ability in all the locations of the array. Learning tended to occur sequentially in different locations of the visual field, expanding from the fovea to the periphery, and the spatial pattern of performance levels was very similar for consecutive or nearly consecutive blocks (Fig. 2a). Furthermore, the expansion in the spatial coverage of the learning tended to occur between adjacent sites in the array. This spatial correlation was not exclusively a function of eccentricity. That is, if a subject was more likely to detect a target in one particular position in the array than in another, he would more easily detect a target in the same and neighboring locations in the subsequent trials. To quantify this, we measured the Euclidean distance between the blocks. Put simply, distance is a quantitative measure of the spatial differences in performance level between consecutive blocks, D1, or blocks separated by greater intervals (D2, D3,..., Dn). We plotted Euclidean distances between positions as a function of block separation under three different conditions, either in the real data, in shuffled data (in which the responses 265


were scrambled in the a different positions for each block of the original data) or in angularly scrambled data, in which positions were shuffled but all points retained their initial radial distance from the fixation point (Fig. 2b). The increase in Euclidean distance as a function of block separation in the original data demonstrated a strong correlation between successive blocks. The results for the scrambled data show that this correlation was not fully accounted for by the increase in the Fig. 2. Learning showed visuotopic specimean rate of correct ficity. It progressed serially from fovea to Scrambled data responses, because the periphery; positions showing improvescrambling the responsment were correlated from trial to trial. Angularly scrambled data b es through different (a) Percent of correct responses, for one Real data subject, as a function of position for differpositions (keeping the ent blocks during the learning period. Each total rate of responses square corresponds to one block of 150 constant through blocks) trials; within each square, the gray-scale increased the distance value of each circle represents the perforbetween neighboring trimance level for a particular location within als considerably. The the 5 × 5 array. The fixation spot in the results for the angularly center of the array is indicated with a small scrambled data showed black circle. (b) The learning showed spathat the correlation was tial specificity. Average distance between not simply radial correblocks for 4 subjects (see Methods) were significantly smaller for measured data than lation due to the profor scrambled data in 2 different condigression of learning from Block separation (number of blocks) tions, either with all 24 positions scramfovea to periphery; bled or with only those positions rather, they demonstratequidistant from the fovea scrambled. ed correlation of precise locations throughout the course of learning. It is important to remember that the target was presented Form specificity of the learning with equal probability at each location within the trained array. The last series of experiments were designed to test whether our Thus, the observed visuotopic specificity was not due to trainsearch task involved solely what are considered low-level mechaing on particular locations within the array. To exclude the posnisms, such as orientation discrimination or texture segmentation. sibility that the improvement might result merely from an We tested subjects who had been trained on our search task with increase in speed in deciding whether a particular shape two novel stimulus configurations. In the first configuration, trianmatched the target, regardless of its position, we tested the gles were replaced by arrowheads, which were still clearly recogniztrained subjects in a condition in which both target and a variable as pointing left, right, up or down, but which did not have the able number of distractors were presented outside the area of property of closure (Fig. 4b). The learning effect was measured as the training array. We then compared performance levels for the ratio between performance levels using the trained and untrained various numbers of distractors outside the training array with orientations. The orientation specificity in the levels of performance that for the same number of distractors presented along with for closed triangles did not transfer to the arrowheads. The means for targets within the training array. Within the area of the trainarrowheads were p′ = 36.2 ± 8.0% in the trained orientation and ing array, performance levels did not change with the number of p′ = 35.0 ± 5.5% in the untrained orientation (averaged over 3 subdistractors when the target was at the trained orientation, but jects; significance, p > 0.5; Fig. 4d). This shows that subjects learned did change when it was at an untrained orientation. Outside of not to discriminate orientation but actually to find an object at a the training array, levels of performance for both trained and particular orientation. In the second configuration, the target was untrained orientations decreased with increasing number of not changed, but the field of distractors was completely novel. This distractors (Fig. 3). This shows that the improvement was spewas done to determine whether the learning was specific for the tarcific for a particular shape and for a particular region of the get itself or for a more generalized textural difference between forevisual field on which subjects were trained. ground and background. The new figures used as distractors did Distance


articles

266



articles

b

c Trained orientation

the failure of training effects to transfer to arrowheads suggests that this task was not a simple orientationUntrained region discrimination task. Interestingly, after training, all subjects claimed that there was no conscious perceptual distinction between different triangles, even though they performed considerably better for target triangles of one orientation than for other orientations. The effect of learning on this Number of distractors task represents an extension of findings on the involuntary nature of priming, which is thought to be limited to simple visual (elenot include the triangles in the other three orientations (Fig. 4c) and mentary) attributes as opposed to form6. The degradation of perwere presented at different contrasts to increase distractor variabil3 ity, making the task more difficult . We observed that the specificity formance on figures of new orientations after learning one orientation further extends the analogies with priming effects for of the training extended to this new background: for the trained oriposition, which shows distractor inhibition7, and might result entation, p′ = 68.1 ± 11.0%, compared with p′ = 37.9 ± 11% for untrained orientation (averaged over 3 subjects; significance, from a difficulty in ignoring targets whose processing has been p < 10–5; Fig. 4d). automated as a consequence of perceptual learning11–13. Learning in this task was not just a consequence of perceptual exposure, but must have involved top-down influences13–21. DISCUSSION This is clear because learning occurred only for the target orienWe studied the effects of training on a search task in which the tation, even though the subject was exposed seven times more target was defined by form. In this task, search efficiency was sigoften to triangles in each of the untrained orientations during nificantly increased as a consequence of learning. This learning the whole course of learning. Identification of form demands was object specific and resulted from a progressive acquisition of attention22–26, as is suggested by the decrease in performance with the ability to identify the given object in different locations in the visual field. The results suggest that learning in visual search can increases in number of distractors. This implies not only that be targeted to a specific object. Although it is suggested that learnattention is required to obtain learning, but that, conversely, ing in visual search involves a general improvement in performlearning is required to rapidly direct the attentional mechanism ing searches8, other studies show orientation dependence of toward a particular object. Visual search tasks are usually classified as parallel or serial learning pop-out detection9. based on whether the performance depends on the number of The task used in our experiments did not involve texture segdistractors1, though it is suggested that this classification reprementation or orientation discrimination, but identification of an oriented object. This is supported by three observations. First, the sents not a real dichotomy, but two extreme cases of a continusubjects could not perform the task if they did not have previous um3,27,28. Here we show that increasing the number of distractors knowledge of the target characteristics. Triangles of a particular oriwithin the training region did not change search efficiency for entation embedded in triangles of other orientations could not be the trained orientation. However, search efficiency diminished detected as unique objects. Second, we showed that learning in this when distractors were added outside of the training region. This task was specific for the target and transferred to different backshowed that the dependence of search efficiency on number of grounds. In contrast, learning effects in texture discrimination are distractors may be a function of distractor position as a consespecific for the field of distractors but not for the target10. Third, quence of perpetual learning. Another characteristic of learning Untrained orientation

Percent correct


a

Fig. 3. Performance as a function of number of distractors within and outside the training region. After training in the 5 × 5 array (gray square), performance was tested outside the training region. The target and either 7 (a) or 23 (b) distractors were presented at eccentricities ranging from 3.5° to 4.7°. (c) Performance within the training region was tested for a 3 × 3 array (7 distractors) and a 5 × 5 array (25 distractors), and was averaged over 3 subjects. Within the training region, performance for the untrained orientation, but not for the trained orientation, declined with number of distractors. Outside the training region, performance for both trained and untrained orientations declined with number of distractors.


Trained region

267


articles

a

Trained

b

Target (present)

Trained

d

New background

Arrowheads

Target (present)

New background

Percent correct


Target (present)

c

Arrowheads

was that the inherent dependence of performance on position within the visual field was greatly reduced with training, as for the detection of oriented gratings9. In our experiments, the target could appear at any position within the array. Therefore, the spatial dependence of the learning observed within the training region was not due to the location of the stimulus, as is the case when localized improvement results from practice in a fixed position of the visual field10,19,29,30. This specificity results, therefore, from intrinsic mechanisms that may reflect the sequence of sites targeted by the search strategy. It could be argued that the improvement resulted from an increase in the speed with which the subjects, independent of visuotopic position, could determine whether a given item was the target. This, combined with a search strategy in which subjects started scanning close to the fovea and proceeded to the periphery, could account for the spatial specificity we found within the training region. If this were the case, a subject trained for ‘left’ triangles should perform better for ‘left’ than for ‘right’ triangles outside of the area used for training. However, we showed that when target and distractors were presented outside of the training region, detectability was no better for the trained orientation. Poorer performance for targets of the trained orientation presented outside the training region, therefore, showed that the improvement was localized to a particular region of the visual field. Based on its spatial specificity, one may speculate that early cortical processing might be involved in this process. The progression of learning across the visual field suggests that repre268

Fig. 4. Learning was specific for object configuration and transferred to a new background. (a) The training array. (b) Test using arrowheads as target and distractors. (c) Target used in training (triangle) in a new field of distractors. For the new background, we used open figures, circles, squares, diamonds and semicircles as distractors (c). The target was presented at the same luminance (60 cd per m2) used in the other experiments. Distractors were randomly assigned a contrast of 33–91 cd per m2. (d) Performance was averaged over three subjects for targets and for arrowheads in trained and untrained orientations, which lacked the feature of closure but were still oriented figures. Performance did not differ for arrowheads in trained and untrained orientations. Performance in the new background (average over three subjects) was better for the trained orientation, demonstrating object specificity of the learning.

sentations of the trained object may be built repeatedly for different positions across the cortical area. Even within V1, cells are selective for much more complex stimulus configurations than originally believed31–34, suggesting a role for V1 in the identification of complex forms. Connections within V1 are plastic35,36, and modification of these connections may contribute to the plasticity of elementary-feature maps. Representation of more complex features at earlier levels may enhance efficiency and rapidity in recognizing these features in a complex background at the expense of requiring multiple shape representations in areas showing smaller receptive fields and greater visuotopic order.

METHODS Psychophysical experiments on human observers (male and female, 23–27 years of age) were designed to study the effects of learning in a visual search task. All subjects gave written informed consent in accordance with procedures and protocols approved by the Rockefeller University Institutional Review Board. Stimuli were presented on a NEC monitor 5FGp refreshed at a rate of 60 Hz, and were observed a distance of 150 cm with both eyes, with normal pupil apertures and without head restraint. Each trial consisted of a 3000-ms cycle. A 5 × 5 array consisting of a central fixation spot and 24 shapes in the remaining locations was presented for 300 ms; a response was recorded during the subsequent 270ms interstimulus interval. As an auditory cue to alert the observer, a short beep was sounded at the onset of the visual stimulus. The psychophysical experiments investigated the observer’s ability to identify a target triangle among an array of distractors. Both target and distractors were presented at high contrast (60 cd per m2) against a uniform background (2 cd per m2). The target randomly appeared in any nature neuroscience • volume 3 no 3 • march 2000


articles

location within the array. After each presentation, the subject indicated whether or not a target shape was present by pressing the appropriate button of a computer mouse. The array subtended 4.2° × 4.2°, and a small fixation spot of one arcmin radius (1′) was positioned in its center. Figures used as target or distractors in the different experiments were equilateral triangles, squares, diamonds or arrowheads. The sides of all shapes were 27′ in length and their centers were separated by 54′. Average distance as a function of block separation Dn, was calculated as follows. From each block we calculated a 5 × 5 matrix (Ma, is the corresponding matrix to block a) where each position of the matrix (mai,j) is defined as the level of performance in the corresponding location of the visual field in this block (Fig. 2a). We then have an ordered array of matrixes, and we can consider the distances between any two of those matrixes.

d(Ma, Mb) =

√

5

5

Σ Σ(m

a

i=1 j=1

i,j —

mbi,j)2


We then define the average distance as block separation to be

Dn = 〈d(Mi, Mi+n)〉i With the exception of one subject (author M.S.), all subjects were naive and were told only what the target was for each block. A session for one day consisted of eight blocks, each comprising 150 trials. All results correspond to the average values for all subjects performing each experiment, and two-tailed t-tests over the data collected from all subjects were used as tests of significance. All errors plotted correspond to standard deviations. Individual tests of significance gave comparable results.

ACKNOWLEDGEMENTS We thank R. Crist for discussions and comments on the manuscript. This work was supported by NIH grant EY07968 and a Burroughs Wellcome fellowship to M.S.

RECEIVED 7 JULY 1999; ACCEPTED 4 JANUARY 2000 1. Treisman, A & Gelade, G. A feature integration theory of attention. Cognit. Psychol. 12, 97–136 (1980). 2. Sagi, D. & Julesz, B. ‘Where’ and ‘what’ in vision. Science 228, 1217–1219 (1985). 3. Duncan, J. & Humphrey, G. W. Visual search and stimulus similarity. Psychol. Rev. 96, 433–458 (1989). 4. Rubinstein, B. S. & Sagi, D. Spatial variability as a limiting factor in texturediscrimination tasks: implications for performance asymmetries. J. Opt. Soc. Am. A 7, 1632–1643 (1990). 5. Wang, Q., Cavanagh, P. & Green, M. Familiarity and pop-out in visual search. Percept. Psychophys. 56, 495–500 (1994). 6. Maljovic, V. & Nakayama, K. Priming of pop-out detection: role of features. Mem. Cognit. 22, 657–672 (1994). 7. Maljovic, V. & Nakayama, K. Priming of pop-out: II. role of position. Percept. Psychophys. 58, 977–991 (1996). 8. Sireteanu, R. & Rettenbach, R. Perceptual learning in visual search: fast, enduring but non-specific. Vision Res. 35, 2037–2043 (1995). 9. Efron, R. & Yund, E. W. Guided search: the effects of learning. Brain Cogn. 31, 369–386 (1996).


10. Karni, A. & Sagi, D. Where practice makes perfect in texture discrimination: evidence for primary visual cortex plasticity. Proc. Natl. Acad. Sci. USA 88, 4966–4970 (1991). 11. Schneider, W. & Shiffrin, R. M. Controlled and automatic human information processing: I. detection, search and attention. Psychol. Rev. 84, 1–66 (1977). 12. Shiffrin, R. M. & Schneider, W. Controlled and automatic human information processing: II. perceptual learning, automatic attending and a general theory. Psychol. Rev. 84, 127–191 (1977). 13. Treisman, A., Verira, A. & Hayes, A. Automaticity and preattentive processing. Annu. Rev. Neurosci. 105, 341–362 (1992). 14. Ahissar, M. & Hochstein, S. Learning pop-out detection: specificities to stimulus characteristics. Vision Res. 36, 3487–3500 (1996). 15. Braun, J. Vision and attention: the role of training. Nature 393, 424–425 (1998). 16. Ahissar, M. & Hochstein, S. Attentional control of early perceptual learning. Proc. Natl. Acad. Sci. USA 90, 5718–5722 (1993). 17. Ito, M., Westheimer, G. & Gilbert, C. D. Attention and perceptual learning modulate contextual influences on visual perception. Neuron 20, 1191–1197 (1998). 18. Fahle, M. & Morgan, M. No transfer of perceptual learning between similar stimuli in the same retinal position. Curr. Biol. 6, 292–297 (1996). 19. Crist, R. E, Kapadia, M., Westheimer, G. & Gilbert, C. D. Perceptual learning of spatial localization: specificity for orientation, position and context. J. Neurophysiol. 78, 2889–2894 (1997). 20. Shiu, L. P. & Pashler, H. Improvement in line orientation discrimination is retinally local but dependent on cognitive set. Percept. Psychophys. 52, 582–588 (1992). 21. Ahissar, M. & Hochstein, S. Task difficulty and the specificity of perceptual learning. Nature 387, 401–406 (1997). 22. Bravo, M. J. & Nakayama, K. The role of attention in different visual search tasks. Percept. Psychophys. 51, 465–472 (1992). 23. Wolfe, J. M. in Current Directions in Psychological Sciences 124–128 (Cambridge Univ. Press, Cambridge, 1992). 24. Wolfe, J. M., Cave, K. R. & Franzels, S. R. Guided Search: an alternative to the feature integration model of visual search. J. Exp. Psychol. Hum. Percept. Perform. 15, 419–433 (1989). 25. Joseph, J. S., Chun, M. M. & Nakayama, K. Attentional requirements in a preattentive feature search task. Nature 387, 805–807 (1997). 26. Chun, M. M. & Jiang, Y. Contextual cueing: implicit learning and memory of visual context guides spatial attention. Cognit. Psychol. 36, 28–71 (1998). 27. Braun, J. & Sagi, D. Vision outside the focus of attention. Percept. Psychophys. 48, 45–58 (1990). 28. Nakayama, K. & Joseph, J. S. in The Attentive Brain (ed. Parasuraman, R.) 279–298 (MIT Press, Cambridge, Massachusetts 1997). 29. Fiorentini, A. & Berardi, N. Learning in grating waveform discrimination: Specificity for orientation and spatial frequency. Vision Res. 21, 1149–1158 (1981). 30. Nazir, T. A. & O’Regan, J. K. Some results on translation invariances in the human visual system. Spat. Vis. 5, 81–100 (1990). 31. Kapadia, M. K., Ito, M., Gilbert, C. D. & Westheimer, G. Improvements in visual sensitivity by changes in local context: Parallel studies in human observers and in V1 of alert monkeys. Neuron 15, 843–856 (1995). 32. Posner, M. I. & Gilbert, C. D. Attention and primary visual cortex. Proc. Natl. Acad. Sci. USA 96, 2585–2587 (1999). 33. Sillito, A. M., Grieve, K. L., Jones, H. E., Cudeiro, J. & Davis, J. Visual cortical mechanisms detecting focal orientation discontinuities. Nature 378, 492–496 (1995). 34. Das, A. & Gilbert, C. D. Topography of contextual modulations mediated by short-range interactions in primary visual cortex. Nature 399, 655–661 (1999). 35. Darian-Smith, C. & Gilbert, C. D. Axonal sprouting accompanies functional reorganization in adult cat striate cortex. Nature 368, 737–740 (1994). 36. Gilbert, C. D., Das, A., Ito, M., Kapadia, M. & Westheimer, G. Spatial integration and cortical dynamics. Proc. Natl. Acad. Sci. USA 93, 615–622 (1996).

269


articles

Seeing multiple directions of motion—physiology and psychophysics Stefan Treue, Karel Hol and Hans-Jürgen Rauber Cognitive Neuroscience Laboratory, Dept. of Neurology, University of Tübingen, Auf der Morgenstelle 15, 72076 Tübingen, Germany


Correspondence should be addressed to S.T. ([email protected])

Dot patterns sliding transparently across one another are normally perceived as independently moving surfaces. Recordings from direction-selective neurons in area MT of the macaque suggested that this perceptual segregation did not depend on the presence of two peaks in the population activity. Rather, the visual system seemed to use overall shape of the population response to determine the number and directions of motion components. This approach explained a number of perceptual phenomena, including susceptibility of the motion system to direction metamers, motion patterns combining three or five directions incorrectly perceived by subjects as comprising only two directions. Our findings offer insights into the coding of multi-valued sensory signals and provide constraints for biologically based computational models.

Analysis of visual motion is an important aspect of processing visual information. It is therefore not surprising that primates have cells and cortical areas specialized for visual motion processing. The specialized neurons are direction selective: each responds only to a particular subset of directions and speeds of motion within its receptive field. Cortical areas along the dorsal pathway of the primate visual system are specialized for the analysis of visual motion; most notably, these include area MT, which is dominated by direction-selective cells, and area MST, which also contains cells encoding more complex motions like rotation or expansion1. MT and MST are arguably the most intensively studied extrastriate cortical areas. The curve facing out and left in Fig. 1a is an idealized tuning curve of an MT cell to a random dot pattern moving in a single direction. The response distribution is bell shaped and is well fit by a Gaussian curve. As is characteristic of direction-selective neurons throughout the visual system, responses are broadly tuned, with a tuning width of approximately 90°, on average 2–4 . Although neurons differ in their preferred directions, their tuning curves are otherwise very similar. If typical tuning curves are combined and plotted as a function of direction preference, we obtain a three-dimensional graph of population responses as a function of both stimulus direction and preferred directions of the neurons (Fig. 1a). Experimentally, this threedimensional plot of activity can be derived by determining responses of individual cells to movement in different directions (contour lines in Fig. 1a). The brain, however, must calculate this function from information derived from the opposite scenario: a single stimulus simultaneously activates a population of neurons, each preferring a different direction. We can predict this population response to a single stimulus from the three-dimensional plot by taking neural responses along a line parallel to the axis denoting direction preference (the curve facing right in Fig. 1a). This population response has the same shape as the tuning curve for an individual cell. We will there270

fore use the terms tuning curve (which describes a single neuron’s response to many directions of motion) and population response (which describes the response of a population of neurons tuned to different directions of motion to one stimulus) interchangeably. The bell-shaped distribution of population activity suggests that the visual system might recover the direction of a stimulus by determining the most active neurons. Such an approach of taking the preferred direction of the most active of a population of direction-tuned neurons as the perceived direction may explain the ability of the visual system to extract the overall direction of motion even for displays combining more than one direction5,6. Even for these displays, activity is greatest for neurons whose preferred directions match the center of the distribution of stimulus directions. Not all visual motion displays create a perception of unidirectional motion, though. Combination of two patterns sliding past each other often creates the impression of two separate surfaces, each moving in a distinct direction. If the percept of a distinct direction is based on the existence of a peak in the population activity, then the perception of transparent motion implies the presence of two peaks. This is either assumed in or a feature of several models of motion perception7–9 and cortical motion processing10,11. If we hypothesize that the response to a transparent stimulus is simply the scaled sum of the activities evoked by the individual components12,13, we can generate a hypothetical curve for population activity in response to movement in two directions at a large angle (Fig. 1c). (Several studies suggest that the response to two motions is the average of the individual responses, giving a scaling factor of 0.5. Note, however, that the exact value is not critical, as multiplication by any factor gives the same shape for the population activity as the unscaled sum of the two individual responses.) However, if the two directions of motion form such a small angle that the two peaks merge (Fig. 1d), the percept nature neuroscience • volume 3 no 3 • march 2000


articles

b

a Hypothetical population response to a single direction

Response

c


Hypothetical population response to two widely spaced stimulus directions

d Stimulus direction

Cell’s preferred direction

Fig. 1. Direction-tuning curves and hypothetical population responses to transparent patterns. (a) A series of hypothetical tuning curves plotted three-dimensionally as a function of stimulus direction (x axis) and the preferred direction of a cell (y axis). The z axis plots responses normalized to the maximal response (response to the preferred direction). The curve facing out and left is the response curve of a neuron tuned to 0°, and the curve facing to the right shows the population response to a stimulus moving at 180°. (b) A hypothetical population response to a single direction of motion. (c) A hypothetical response to movement of two transparent stimuli at an angle large enough to produce a double-peaked population response. The dashed lines indicate the population responses to each of the two directions. (d) Response as in (c) but with an angle small enough to merge the two curves into a single-peaked population response.

Hypothetical population response to two closely spaced stimulus directions

should change from transparent motion to motion in a single direction intermediate to the two actual component directions. Perceptually, the transition from perceiving two separate directions to only one direction occurs when the two superimposed transparent random dot patterns move at a relative angle of about 10° or less14. Although several physiological studies examine responses of direction-selective neurons to transparent patterns, little is known about the population activities evoked by patterns combining similar directions, as almost all of these experiments involve only a small subset of directions or directions at large relative angles. The hypothesis that the profile of responses to motion in multiple directions is the scaled sum of the responses to the individual components makes a clear prediction: the sum of two Gaussians shows distinct peaks if the distance between the centers of each distribution exceeds their width (which for directiontuned neurons averages about 90°). Conversely, transparent motion at any acute angle should yield an activity profile with a single peak. This means that either the response profile to two directions is a highly non-linear combination of the individual directions, resulting in two peaks even for small directional separations between the components, or that a percept of transparent motion does not depend on the presence of two peaks in the population activity. Here we show that the latter seems to be the case. Perception of multidirectional motion does not require distinct peaks in the population activity. Rather, the visual system seems to be able to interpret the overall shape of the activity profile. This allows the visual system to segregate motions differing even by only a small angle. However, we also show that summing the responses to the individual motion components makes the motion system susceptible to direction metamers, stimuli that are perceptually indistinguishable even though they contain different direction components. nature neuroscience • volume 3 no 3 • march 2000

RESULTS We recorded the responses of 152 direction-selective cells in area MT of 3 macaques to moving random dot patterns while the monkeys performed a fixation task. The stimuli moved coherently within a stationary virtual aperture either in one direction or in two directions separated by 30°, 60°, 90° or 120°; the 5 resulting activity profiles are shown for a narrowly tuned example cell (Fig. 2a). The top panel plots the responses for single motion. The line represents the best fitting Gaussian (tuning width, 58°). Given the tuning width of this neuron, the scaled sum of the responses to the individual components of the transparent motion predicts a single-peaked activity profile for the 30° pattern and two peaks for the 90° and 120° stimuli (Fig. 2a, lower panels). As predicted, two peaks were generated only for stimuli separated by 90° or 120°. To test our prediction for our entire sample of neurons, we aligned every activity profile to the cells’ preferred direction, normalized its height to the response to that direction and then averaged all profiles (Fig. 2b). The tuning width of the averaged responses to the single-direction stimulus was 96°; therefore, as predicted, the activity profiles did not show two peaks for stimuli at acute angles, remaining essentially flat for the 90° stimulus. Note, though, that because tuning width varied somewhat between neurons, we were concerned that deviations from our prediction might be occluded by averaging across neurons irrespective of their tuning width. Therefore, we recoded the data to test our prediction more directly. For every cell, responses were expressed as a function of the neuron’s tuning width to single directions. Similarly, the angles between transparent stimuli were also expressed as fractions of the width of a cell’s tuning curve to a single direction. Averaging across all cells generated a contour plot (Fig. 2c). The x axis plots the average stimulus direction as a proportion of the tuning width, the y axis gives the normalized angle formed by the two directions in the stimulus, and the contour surface represents the 271


articles

Aligned, normalized and pooled responses Singlestimulus

Single-direction stimulus

30° transparent stimulus


Normalized response

Response (spikes per s)








Stimulus direction (degrees)

Angle between component directions, divided by tuning widths

single-direction

Single cell

Stimulus direction (degrees away from preferred direction)

Direction of stimulus from preferred direction, divided by tuning width

Fig. 2. Population activity profiles for transparent motion of various angles. (a) Response profiles for an example cell to the single-direction stimulus (top curve) and the bidirectional stimuli at 30°, 60°, 90° and 120° (bottom 4 curves). The x axis plots the stimulus direction relative to the cell’s preferred direction. For the transparent motion patterns, the stimulus direction plotted is the average of the two direction components. (b) Response profiles averaged across all cells after normalizing the firing rates to the response to the preferred single direction (determined by fitting with a Gaussian curve). (c) Contour plot of the responses across all cells. The data from each neuron are normalized to the highest response in each response profile. The resulting data points form a horizontal line in the graph at the y position defined by the angle between the two directions present in the stimulus normalized by the width of the neuron’s tuning curve for single directions. The x axis similarly expresses the angle of the stimulus direction relative to the neuron’s single-direction tuning width. The four horizontal dashed lines indicate data for the example cell from the left column. From top to bottom, they show data points for the single direction, 30°, 60° and 90° stimulus conditions on the contour plot. Because of the narrow tuning of the example cell, the data from the 120° stimulus condition fell below the scale. The data are in agreement with our prediction that the population activity splits into two peaks when the stimulus angle exceeds the bandwidth (when y > 1).

normalized firing rate. The graph forms an inverted ‘Y’, clearly showing that for ratios less than one (when the angle between stimuli is smaller than the neuron’s tuning width), the response profile is single-peaked, whereas there are two peaks for larger angles. This demonstrates that the simple scaled sum of the responses to individual components accounts for responses to multidirectional stimuli very well. Thus, the response profile was not multi-peaked when the angle between component directions was 272

at the tuning width of ∼90° or smaller. Nevertheless, the visual system can recover the directions underlying transparent motion with much smaller directional separation. How is this remarkable ability achieved? Note that, although the population activity was not multi-peaked, it nevertheless contained information about the underlying directions. Single-lobed activity profiles generated by bidirectional stimuli are notably broader than tuning curves for a single direction (Figs. 1d and 2). This change in overall shape of the activity profile might be used by the visual nature neuroscience • volume 3 no 3 • march 2000



articles

age direction-tuning curve) that must be summed to produce the observed activity profile. For motion in a single direction, this would be a single Gaussian positioned at the actual direction. When movement in two directions is presented, the resulting population activity is too broad to be accounted for by a single Gaussian, but can be described with two Gaussians. This approach would yield the correct percept even when the component directions form an acute angle and, therefore, result in a singlepeaked activity profile (Fig. 1d). Such an approach can be modeled formally (see Discussion). Although this approach is norFig. 3. Psychophysical stimuli. We created two stimuli with three (b1) and five (b2) directions that should be directional metamers to the ±40° bidirectional stimulus (b). The three and five directions mally a robust method of recovering and the corresponding dot densities were chosen such that the sum of the individual curves (shown the underlying direction(s), it should along the right) was virtually identical to the activity evoked by the ±40° bidirectional stimulus. If the fail to distinguish between different population activity was indeed the one proposed here and the only information available to the stimuli when they evoke the same motion system, these three stimuli (b, b1, b2) should appear to contain the same direction compo- activity profile. Such metamers, stimnents. In addition to the potentially metameric bidirectional stimulus (b), we created two other bidi- uli that are perceptually indistinrectional stimuli (a and c). These cause population activities different from the three potential guishable despite actual differences, metamers (as shown in the left graph) and should therefore be perceptually distinguishable from the can be used to psychophysically estiother stimuli. Note that the directions in the ±50° bidirectional stimulus (c) are present as the mate direction-tuning bandwidth16. steepest directional components in stimuli b1 and b2. Because the subjects were asked to judge the most upward component in all stimuli, stimuli b1 and b2 should result in the same responses as the If the visual system indeed recovers ±50° stimulus if the motion system correctly encoded the individual motion components. Tuning direction components of transparent width for the response profiles used here was assumed to be 90°, although note that the width of motion by assessing the width or the summed activity for the three- and five-direction stimuli (b1 and b2) remains close to the activ- overall shape of the population ity width evoked by the two-directional stimulus (b) for a large range of tuning-curve widths. response, it should perceive only two Further material and demonstrations of our stimuli can be viewed at directions in patterns containing three http://neurosci.nature.com/web_specials/. or five directions, as long as the resulting profile of the population activity matches that expected for two directions of motion (Fig. 3). In a series of psychophysical experiments, human subjects were system to identify direction(s) evoking the activity. Component asked to report the number of directions perceived in these patdirections of a bidirectional stimulus can even be approximated terns under eccentric fixation. In more than 94% of the trials with (to within ±10°) by simply subtracting a cell’s tuning width for stimuli containing 3 directions, subjects reported perceiving two a single-direction stimulus from the width of the response profile directions. Even the stimulus with 5 directions was reported to to the given bidirectional stimulus. contain just 2 directions in more than 73% of the trials. We proNote that such an approach is different from vector averagceeded to perform a more direct test of our hypothesis. We preing, which recovers a direction by calculating the vector sum of sented the 3 control stimuli (bidirectional stimuli with the preferred directions of all activated neurons, weighted by the components at ±30°, ±40° or ±50°) together with a stimulus comrespective activities of those neurons. Such an approach would prising 3 directions that contained components moving at 0° and yield an intermediate direction whenever two directions are com±50° (b1 in Fig. 3). If the visual system indeed finds the best-fitting bined. In contrast, a winner-take-all approach predicts perceived direction as the preferred direction of the most active neurons15. sum of two Gaussians for this stimulus, subjects should perceive two directions of motion moving in the same directions as in the Note that, again, transparently moving patterns would be per±40° stimulus rather than the ±50° components actually present. ceived as having a single direction of motion, and, as our data Subjects could reliably discriminate between stimuli with 2 direcshow, this direction would be equivalent to that predicted by the tions at ±30°, ±40° and ±50°, but the 2 direction components pervector-averaging model if the two actual directions of motion ceived in the stimulus containing 3 directions were indeed the formed an acute angle. At larger angles, the population activity same as for the ±40° stimulus with only 2 directions (Fig. 4, left). shows two peaks, and the winner-take-all model would pick one We wondered if this inability of the visual system to exclude of the two peaks. Clearly, this cannot account for the perception the influence of movement in the horizontal direction could be of two directions of motion when they form either a large angle overcome if we supplied additional segregation cues. We colored or a perceptually separable acute angle (larger than ∼10°). upward- and downward-moving dots red and the horizontally We suggest that the visual system uses another approach to moving ones green. Subjects were asked to judge the direction of recover the direction(s) underlying a given population activity the red, upward-moving component. Subjects still judged the across direction-selective neurons. The perceived direction(s) perceived upward motion in the stimulus comprising 3 direcrepresent the position(s) of the smallest number of Gaussiantions to be the same as the one in the ±40° stimulus (Fig. 4, censhaped activity profiles (with widths equal to that for the avernature neuroscience • volume 3 no 3 • march 2000

273


articles


DISCUSSION

Metameric

Metameric

Non-metameric

Perceived directions

ter). Interestingly, this was not the case when the horizontal motion was put on a different stereoscopic disparity plane (Fig. 4, right). With this stimulus manipulation, subjects veridically perceived the upward component, and the responses for the stimulus with 3 directions did not statistically differ from those for the ±50° stimulus.

In summary, we show that the response of direction-selective neurons in MT to transColor Disparity parent motion was well approximated by the scaled sum of their responses to the Fig. 4. Psychophysical results. The filled bars show the upward directional components reported individual motion components. Most for the ±30° (bar a), ±40° (bar b) and ±50° (bar c) stimuli. Error bars indicate standard errors. The white bar indicates the reported upward direction in the three-direction stimulus (bar b ). Left, in importantly in the context of this study, the standard condition, the difference between the two middle bars was not significant:1 subjects population activity profiles were single- reported the same upward component in the ±40° stimulus as in the 3-direction stimulus (which peaked for transparent-motion compo- contains a 50° upward component). Center, when oblique pattern components consisting of red nents moving at acute angles. Our ability dots and the horizontally moving components consisting of green dots were used, the difference to perceive multidirectional motion is, between the two middle bars was not significant. Right, here the oblique pattern components were therefore, not dependent on multiple presented with crossed and the horizontal component with uncrossed disparity. In contrast with peaks in activity profiles across MT neu- the other 2 conditions, subjects reported the same upward component in the 3-direction stimulus rons; rather, it seems to be based on an as in the ±50° stimulus. analysis of the overall shape of the population response, effectively recovering the Gaussian-shaped profiles of responses to the individual components underlying the population activity. parent motion patterns, which is strongest for angles of ∼30˚–40˚. This is supported by our demonstration of direction metamers, The population responses to such patterns are somewhat wider showing an inability of the visual system to recover motion comthan expected from our simple model, but more data and a more ponents if identical activity profiles could be created by alternaformal analysis are required to determine whether this could tive stimuli with fewer directional components. Our finding that account for the perceptual errors observed in motion repulsion. stereoscopic disparity information but not color information The demonstration of direction metamers also suggests that helps the visual system to disambiguate these stimuli supports independent access to the direction components in transparent the proposal that perception of direction in our stimuli involves motion is lost (in the absence of additional cues such as stereosignals from area MT, which contains both disparity-tuned and scopic disparity or speed30), even though the front end of the 17–19 direction-tuned neurons motion system seems to require local imbalance of motion sig. Such neurons would allow segrenals for perception of transparency31,32. Stereoscopic disparity, gation of direction components if they lay at different depths, effectively creating independent population activities for the difbut not color differences, also help to create perceptual transferent planes. Some color information is processed in the conparency in paired random-dot displays33. text of motion processing20,21, but it seems to be insufficient for The inability to use color as a segregation aid does not seem to hold under all circumstances, though. The influence of noise dots the segregation of motion components in our displays, presumon perception by human subjects and monkeys of coherently ably because MT neurons are not color-selective enough to supmoving random dot patterns seems to be reduced if they are of a port two distinct surface representations based on color different color20,21. A number of differences between previous information22,23. Our data show that, contrary to previous proposals, the vecstudies and ours might account for these discrepancies; notably, tors in multidirectional motion do not seem to be encoded indithese include the requirement of subjects to make precise direcvidually by separate populations of neurons. Rather, the broad tion judgments in our experiment and the use of long stimulus tuning of direction-selective neurons in areas such as MT leads durations and larger dot sizes in the previous experiments, posto a population code in which the overall shape of the populasibly allowing eccentric tracking of individual points. tion activity, and not (at least for acute angles) separable peaks, It is indeed possible to devise a computational model that can encodes directions of motion. Using the complete population correctly predict the perceived directions34,35, including the two activity probably also increases signal-to-noise ratios by using all directions perceived when viewing our three-direction stimuli, available information4,24. from population activities such as those we report here for MT neurons. Similarly, in some models of motion processing in Such an encoding scheme accounts for a number of motion MT36,37, acute-angled transparent motion results in single-peaked phenomena, especially single-direction9 and bidirectional10 aftereffects produced by adaptation of subjects with narrow- and activity profiles. wide-angled bidirectional patterns, respectively. Our data also In summary, using a combination of single-neuron recordmight help account for some results of psychophysical studies ings in monkeys and psychophysics in humans, we demonstratusing stimuli that combine many directions or speeds5,25–27. Simed a neural code that supports perception of multiple values without requiring separate neuronal populations to encode the ilarly, our data might contribute to an explanation of the perdifferent stimulus values. Given the straightforwardness of this ceptual phenomenon of ‘motion repulsion’14,28,29, a tendency of coding scheme, it is probably not restricted to the encoding of subjects to overestimate the angle between directions in trans274



articles

motion directions. Rather, it might represent a more general approach, suggesting that metamers similar to the ones demonstrated here for the domain of visual motion direction might exist for other sensory domains.


METHODS Physiology. All animal procedures were approved by the local animal research committee and complied with relevant laws and institutional guidelines. Standard preparation techniques were used for electrophysiological recordings from single neurons in area MT38 of three male macaque monkeys. Cells were characterized as MT neurons based on their receptive field locations, directional tuning and position in the cortex. MT cells were analyzed if they showed directional tuning (responses as a function of stimulus direction were better fit by a Gaussian than by a horizontal line, and responses to the preferred direction were at least fivefold larger than to the null direction). The monkey initiated each trial by depressing a lever, and had been trained to respond to a change in luminance of a small spot at the top of the fixation cross by releasing the lever. Throughout a trial, the animal’s gaze had to stay within 1°–1.3° of visual angle from the fixation cross. If the animal broke fixation or did not respond to the spot’s luminance change within the reaction-time window (200–500 ms after the change), the trial was terminated without reward. Only data from correctly completed trials were analyzed. During a trial, a moving random dot pattern was presented inside the classical receptive field. Its speed and size was matched to the preferred properties of the cell under study. Dots were bright white squares of 0.1° width presented at a density of 5 dots per square degree on an otherwise dark computer monitor positioned 57 cm from the animal. For singledirectional patterns, all dots moved coherently in the same direction; for bidirectional stimuli, half of the dots were assigned to each direction. During stimulus presentation, direction(s) continuously changed at a rate of 100° per s (ref. 39). For our analysis, the value of the spike density profile of the cell’s responses (kernel sigma, 30 ms) was taken every 50 ms. Control measurements were taken to ensure that our results, especially the absence of two peaks in the population response to acute angled transparent motion, could be replicated using non-changing patterns. Psychophysics: stimuli. The experiments were performed using an Apple Macintosh computer and a monitor with 74.5 Hz frame rate. The spatial resolution of the display was 33.3 pixels per degree of visual angle. Our stimuli consisted of random dot patterns (RDPs) presented behind a stationary virtual aperture 6° in diameter. Each RDP was made up of 200 dots (density, 3 dots per square degree; width, 3.6 arcmin) All dots moved uniformly at 4° per s for a stimulus duration of 500 ms, with dots that disappeared behind the aperture reappearing at the opposite side. The subjects viewed the display binocularly in a dimly lit room from a constant distance of 57 cm, maintained by a chin rest. To minimize tracking eye movements, subjects were instructed to maintain fixation on a small cross, 6° to the left of the stimulus center. Tracking eye movements would change the relative directions of the motion components in the retinal image, destroying the metameric qualities of the stimulus. The directions in all stimuli were arranged symmetrically around the horizontal direction, moving to either the right or left. (Fig. 3 shows only the rightward moving patterns.) Stimuli with potentially indistinguishable direction components (Fig. 3, b1 and b2) were constructed such that a summation of the response profiles created by the two, three or five individual components yielded almost identical single-peaked activity profiles (assuming a bandwidth of the underlying response profiles of 90°). Additionally, bidirectional stimuli were used with ±30° and ±50° components. The strength of each individual motion component was scaled multiplicatively by the number of dots moving in that particular direction. The assumption that the neural strength of activation by a particular direction component is a linear function of dot density is a simplification, at least for single-direction patterns3. Given that the overall dot density of the stimulus was kept constant and that our simple model predicted approximately identical population activities over a wide range of component dot densities, this simplification probably allowed the model to predict perception successfully nevertheless. nature neuroscience • volume 3 no 3 • march 2000

In experiment 1 and 3, all dots were black; in experiment 2, the horizontally moving dots were either green or an approximately isoluminant red. In experiment 3, RDPs were viewed through a mirror stereoscope. The patterns moving obliquely were presented with 0.25° crossed disparity, whereas the patterns moving horizontally were presented with 0.25° uncrossed disparity. Subjects. The same twelve subjects served in all experiments. Nine were paid volunteers naive to the research aim, and three were from within the laboratory. All had normal or corrected-to-normal visual acuity. Task. Psychophysical testing protocols were approved by the ethics commission of the University of Tübingen Medical School. The different patterns were presented in random order in blocks of 160 trials. Subjects had to report the upward direction component in each pattern. Immediately after the moving dot patterns, a line was presented on the screen until the subject made a decision. In a two-alternative forced choice protocol, subjects were required to indicate whether the perceived steepness of the upward moving component in the RDP was smaller or larger than the slope of the line. Data analysis. After each trial, in a 50%-staircase procedure, the line was brought toward the perceived direction of motion. For the value of perceived direction, the results of leftward and rightward motion of all subjects for each of the four stimuli were averaged. Note: further material and demonstrations of our stimuli can be viewed on the Nature Neuroscience web (http://neurosci.nature.com/web_specials/).

ACKNOWLEDGEMENTS This work was supported by a grant from the MWF Baden-Württemberg. We are grateful to O. Braddick, N. Qian and R.S. Zemel for comments on previous versions of the manuscript.

RECEIVED 29 DECEMBER 1999; ACCEPTED 25 JANUARY 2000 1. Graziano, M. S. A., Andersen, R. A. & Snowden, R. J. Tuning of MST neurons to spiral motions. J. Neurosci. 14, 54–67 (1994). 2. Albright, T. D. Direction and orientation selectivity of neurons in visual area MT of the macaque. J. Neurophysiol. 52, 1106–1130 (1984). 3. Snowden, R. J., Treue, S. & Andersen, R. A. The response of neurons in areas V1 and MT of the alert rhesus monkey to moving random dot patterns. Exp. Brain Res. 88, 389–400 (1992). 4. Britten, K. H. & Newsome, W. T. Tuning bandwidth for near-threshold stimuli in area MT. J. Neurophysiol. 80, 762–770 (1998). 5. Williams, D. & Sekuler, R. Coherent global motion percepts from stochastic local motions. Vis. Res. 24, 55–62 (1984). 6. Pasternak, T., Albano, J. E. & Harvitt, D. M. The role of directionally selective neurons in the perception of global motion. J. Neurosci. 10, 3079–3086 (1990). 7. Jasinschi, R., Rosenfeld, A. & Sumi, K. Perceptual motion transparency: the role of geometrical information. J. Opt. Soc. Am. A Opt. Image Sci. Vis. 9, 1865–1879 (1992). 8. Mulligan, J. B. Motion transparency is restricted to two planes. Invest. Ophthalmol. Vis. Sci. Suppl. (1992). 9. Verstraten, F. A., Fredericksen, R. E. & Van de Grind, W. A. Movement aftereffect of bi-vectorial transparent motion. Vis. Res. 34, 349–358 (1994). 10. Grunewald, A. & Lankheet, M. J. M. Orthogonal motion after-effect illusion predicted by a model of cortical motion processing. Nature 384, 358–360 (1996). 11. Wilson, H. R. & Kim, J. A model for motion coherence and transparency. Vis. Neurosci. 11, 1205–1220 (1994). 12. van Wezel, R. J. A. et al. Responses of complex cells in area 17 of the cat to bivectorial transparent motion. Vis. Res. 36, 2805–2813 (1996). 13. Recanzone, G. H., Wurtz, R. H. & Schwarz, U. Responses of MT and MST neurons to one and two moving objects in the receptive field. J. Neurophysiol. 78, 2904–2915 (1997). 14. Mather, G. & Moulden, B. A simultaneous shift in apparent directions: Further evidence for a ‘distribution-shift’ model of direction coding. Q. J. Exp. Psychol. 32, 325–333 (1980). 15. Salzman, C. D. & Newsome, W. T. Neural mechanisms for forming a perceptual decision. Science 264 231–237(1994). 16. Williams, D., Tweten, S. & Sekuler, R. Using metamers to explore motion perception. Vis. Res. 31, 275–286 (1991).

275



articles

17. Maunsell, J. H. R. & Van Essen, D. C. Functional properties of neurons in middle temporal visual area of the macaque monkey. II. Binocular interactions and sensitivity to binocular disparity. J. Neurophysiol. 49, 1148–1167 (1983). 18. Bradley, D. C., Qian, N. & Andersen, R. A. Integration of motion and stereopsis in middle temporal cortical area of macaques. Nature 373, 609–611 (1995). 19. DeAngelis, G. C., Cumming, B. G. & Newsome, W. T. Cortical area MT and the perception of stereoscopic depth. Nature 394, 677–680 (1998). 20. Croner, L. J. & Albright, T. D. Image segmentation enhances discrimination of motion in visual noise. Vis. Res. 37, 1415–1427 (1997). 21. Croner, L. J. & Albright, T. D. Segmentation by color influences responses of motion-sensitive neurons in the cortical middle temporal visual area. J. Neurosci. 19, 3935–3951 (1999). 22. Zeki, S. M. The distribution of wavelength and orientation selective cells in different areas of monkey visual cortex. Proc. R. Soc. Lond. B Biol. Sci. 217, 449–470 (1983). 23. Saito, H., Tanaka, K., Isono, H., Yasuda, M. & Mikami, A. Directionally selective responses of cells in the middle temporal temporal area (MT) of the macaque monkey to the movement of equiluminous opponent color stimuli. Exp. Brain Res. 75, 1–14 (1989). 24. Deneve, S., Latham, P. E. & Pouget, A. Reading population codes: A neural implementation of ideal observers. Nat. Neurosci. 2, 740–745 (1999). 25. Watamaniuk, S. N. J. & Duchon, A. The human visual system averages speed information. Vis. Res. 32, 931–941 (1992). 26. Watamaniuk, S. N. J. Ideal observer for discrimination of the global direction of dynamic random-dot stimuli. J. Opt. Soc. Am. A Opt. Image Sci. Vis. 10, 16–28 (1993).

276

27. Zohary, E., Scase, M. O. & Braddick, O. J. Integration across directions in dynamic random dot displays: Vector summation or winner take all? Vis. Res. 36, 2321–2331 (1996). 28. Marshak, W. M. & Sekuler, R. Mutual repulsion between moving visual targets. Science 205, 1399–1401 (1979). 29. Rauber, H. J. & Treue, S. Revisiting motion repulsion: Evidence for a general phenomenon. Vis. Res. 39, 3187–3196 (1999). 30. Van Der Smagt, M. J., Verstraten, F. A. J. & Van De Grind, W. A. A new transparent motion aftereffect. Nat. Neurosci. 2, 595–596 (1999). 31. Qian, N., Andersen, R. A. & Adelson, E. H. Transparent motion perception as detection of unbalanced motion signals III: Modeling. J. Neurosci. 14, 7381–7392 (1994). 32. Qian, N., Andersen, R. & Adelson, E. H. Transparent motion perception as detection of unbalanced motion signals I: Psychophysics. J. Neurosci. 14, 7357–7366 (1994). 33. Braddick, O. Local and global representations of velocity: transparency, opponency, and global direction perception. Perception 26, 995–1010 (1997). 34. Zemel, R. S., Dayan, P. & Pouget, A. Probabilistic interpretation of population codes. Neural Comput. 10, 403–430 (1998). 35. Zemel, R. S. & Dayan, P. Advances in Neural Information Processing Systems Vol. 11 (MIT Press, Cambridge, Massachusetts, in press). 36. Simoncelli, E. P. & Heeger, D. J. A model of neural responses in visual area MT. Vis. Res. 38, 743–761 (1998). 37. Wang, R. Y. A network model of motion processing in area MT of primates. J. Comput. Neurosci. 4, 287–308 (1997). 38. Treue, S. & Maunsell, J. H. R. Effects of attention on the processing of motion in macaque visual cortical areas MT and MST. J. Neurosci. 19, 7603–7616 (1999). 39. Schoppmann, A. & Hoffmann, K. P. Continuous mapping of direction selectivity in the cat’s visual cortex. Neurosci. Lett. 2, 177–181 (1976).



articles

A multimodal cortical network for the detection of changes in the sensory environment


Jonathan Downar1, Adrian P. Crawley2, David J. Mikulis2 and Karen D. Davis1,3 1

Institute of Medical Science, University of Toronto, and Toronto Western Research Institute, MP14-322, 399 Bathurst Street, Toronto, Ontario, M5T 2S8, Canada

2

Department of Medical Imaging, University of Toronto, and Toronto Western Research Institute, MP3-404, 399 Bathurst Street, Toronto, Ontario, M5T 2S8, Canada,

3

Department of Surgery, University of Toronto, Division of Neurosurgery, Toronto Western Hospital, and Toronto Western Research Institute, MP14-322, 399 Bathurst Street, Toronto, Ontario, M5T 2S8, Canada Correspondence should be addressed to K.D.D. ([email protected])

Sensory stimuli undergoing sudden changes draw attention and preferentially enter our awareness. We used event-related functional magnetic-resonance imaging (fMRI) to identify brain regions responsive to changes in visual, auditory and tactile stimuli. Unimodally responsive areas included visual, auditory and somatosensory association cortex. Multimodally responsive areas comprised a right-lateralized network including the temporoparietal junction, inferior frontal gyrus, insula and left cingulate and supplementary motor areas. These results reveal a distributed, multimodal network for involuntary attention to events in the sensory environment. This network contains areas thought to underlie the P300 event-related potential and closely corresponds to the set of cortical regions damaged in patients with hemineglect syndromes.

The ability to detect changes in the sensory environment is crucial to survival. It is necessary to attend to such changes to evaluate and modify behavior in the face of developing obstructions, opportunities or threats. For this reason, changes in the sensory environment, especially abrupt changes, tend to draw attention involuntarily. A sensory element undergoing a change also inserts itself preferentially into awareness. For example, a hiker might not notice the constant sound of birds chirping unless they were to stop abruptly, at which point the hiker would become aware of both the birds and the sudden absence of noise. When the ability to attend to stimuli in the sensory world is lost, as in patients suffering from neglect syndromes, awareness of the stimuli is lost as well1,2. Understanding the mechanisms by which the brain detects changes in the sensory environment will provide us with a better understanding of the mechanisms of both involuntary attention and awareness. We used event-related fMRI to identify the network of neuroanatomical structures underlying the detection of changes in the sensory environment. Visual, auditory and tactile stimuli were used to identify areas responding to changes in multiple sensory modalities. These multimodal areas are of particular relevance to the understanding of higher-order cognitive processes such as the construction of an integrated, multisensory perceptual environment, the directing of attention to salient features of that environment and the selection of those features for entry into awareness3. Subjects underwent fMRI while being presented with visual, auditory and tactile stimuli. To avoid activation due to response selection, planning or working memory, subjects were not required to make any sort of response during the experiment. Instead, they were merely instructed to observe the stimuli passively. Studies concerning the detection of stimulus events frequently involve nature neuroscience • volume 3 no 3 • march 2000

‘oddball’ protocols, in which subjects are presented with a train of repeating, standard stimuli occasionally punctuated by a different ‘oddball’ stimulus. Our study used a modified version of this approach. In our protocol, the stimuli in each modality were presented continuously, but alternated independently between two different states, A and B. Each alternation was termed a ‘transition’ regardless of whether the direction of the change was from A to B or B to A (Fig. 1). We used these counterbalanced transitions between stimulus states, rather than a stereotypical oddball stimulus, to ensure that activations were due to a general change in the quality of the stimulus and were not merely due to a difference in some specific feature of the oddball stimulus as compared with the standard. A transition occurred in one of the 3 sensory modalities every 14 seconds in a randomized sequence to minimize the effects of expectancy and habituation. To identify activations, we treated transitions as the stimulus events. This approach enabled us to identify cortical areas responsive to transitions within a single sensory modality, as well as a cortical network responsive to transitions in multiple sensory modalities.

RESULTS Unimodally responsive areas To identify unimodally responsive areas, we used a voxelwise ttest to compare the blood oxygenation level-dependent (BOLD) signal two to eight seconds following a transition within a given modality with the signal two to eight seconds after a transition in the other modalities. Based on previous studies, these time periods encompassed the peaks of the hemodynamic responses to brief stimuli4–6. Brain regions satisfying the criteria for unimodal activation (see Methods) are therefore maximally responsive to transitions in one modality as compared with the others (Table 1; Figs. 2 and 3). 277



articles

Fig. 1. Excerpt of stimulus-presentation protocol. Visual, auditory and tactile stimuli were presented to subjects simultaneously during imaging. The visual stimulus was either the blue or the red abstract figure shown; the auditory stimulus was either the sound of running water or the sound of croaking frogs (actual waveforms shown); the tactile stimulus was either circular brushing or rapid tapping of the lower right leg, using a 6 × 10 cm shower brush. At 14-s intervals, in pseudo-random order, 1 of the 3 stimulus modalities underwent a transition from one stimulus type to the other. A total of 44 transitions (15 visual, 15 auditory, 14 tactile) occurred during the course of the experiment. V, visual; A, auditory; T, tactile.

Visual-specific areas comprised bilateral visual association cortices of both the dorsal and ventral visual-processing streams, including the fusiform gyrus and middle occipital gyrus. The former area includes inferior temporal regions known to be involved in the identification of complex figures such as objects and faces7, whereas the latter corresponds to area V3 of visual association cortex. In addition, there was a locus of activation in the right superior parietal lobule, on the superior bank of the intraparietal sulcus. This area activates for shifts of attention in the visual modality8. Auditory-specific areas comprised Brodmann areas 41, 42 and 22 of the right and left superior temporal gyri. These areas correspond to primary and secondary auditory cortex. No other auditory-specific loci of significant activation were noted. Tactile-specific areas comprised the secondary somatosensory cortex (S2). Although stimulation was applied only to the right leg (Fig. 1), activation was observed bilaterally, consistent with the bilateral receptive fields of many S2 neurons9–11. No activation was observed in the primary somatosensory cortex (S1) for the right leg. Inhibitory mechanisms for attention, both within and between modalities, are topics of considerable interest. An fMRI study of visual cortical activity during binocular rivalry found that extrastriate areas responsive to a particular stimulus show deactivations when a competing stimulus becomes perceptually dominant12. Intermodal inhibition of visual cortical activity during a tactile discrimination task is also reported13. To assess the possibility of intermodal inhibition in our study, we examined the averaged responses of unimodal-specific areas to transitions in their own versus the other modalities (Fig. 3). Although all unimodal areas showed consistent activation for their particular modalities, only the visual-specific areas showed significant deactivation in response to transitions in other modalities (auditory, t88 = 3.69, p < 0.001; tactile, paired t82 = 6.04, p < 10–7), as measured by comparing the BOLD signal of 278

these areas during the 6 seconds before and after each transition, with a 2-second delay to accommodate the hemodynamic response. Transition-related intermodal inhibition thus appeared to affect only the visual modality. Multimodally responsive areas To identify areas responsive to transitions in all three modalities, we used a voxelwise correlation to the predicted hemodynamic response to brief stimuli spaced 14 seconds apart, based on previous studies4–6. Brain regions satisfying the criteria for multimodal activation (see Methods) are presented in Table 1 and Figs. 2, 3 and 4. Areas activated by transitions in all three sensory modalities comprised a distributed network of frontal, medial, temporoparietal and insular cortical areas (Fig. 4). The network showed a greater overall volume of activated voxels in the right hemisphere (5596 mm3) as compared with the left (1244 mm3; Table 1). The most prominent area of activation was located at the right temporoparietal junction (TPJ), immediately posterior to the secondary auditory and somatosensory cortices (Figs. 2 and 4). A much smaller activation was identified in left TPJ. Ventral and posterior to the activation in the right TPJ was a small activation in the right middle temporal gyrus (Figs. 2 and 4). Frontal areas responsive to transitions in all modalities were located bilaterally in the inferior frontal gyrus and frontal operculum. As in the TPJ, frontal activation was more extensive in the right hemisphere than in the left. There were also foci of activation in an anterior and a posterior region of the right insula. Medial multimodal activations were located in the left anterior cingulate and the supplementary motor areas (CMA and SMA).

DISCUSSION The results of our study reveal a distributed cortical network for the detection of changes in the sensory environment, with both unimodal and multimodal components. nature neuroscience • volume 3 no 3 • march 2000



articles

The unimodal components of this network corresponded to visual association cortex, primary and secondary auditory cortex, and S2. It is noteworthy that V1, V2 and S1 did not show a significant response to transitions. Because all stimuli were applied continuously throughout the experiment, it is unlikely that these areas were simply inactive during scanning. It is more likely that their overall level of activation did not change significantly during or following transitions. The activity of these areas may thus reflect a more literal representation of sensory input, less influenced by habituation or attentional modulation than that of secondary and association cortices. The unimodal auditory activations observed in primary and secondary auditory cortex are consistent with recent findings concerning the detection of sensory events in the auditory modality. Deviant or ‘oddball’ auditory stimuli embedded in a train of standard, repeating stimuli normally evoke a mismatch negativity (MMN) observable in electroencephalographic (EEG) recordings. An EEG study of patients with focal cortical lesions suggests auditory association cortex as a major generator of the MMN for oddballs presented to the contralateral ear14. Our data supports this hypothesis by demonstrating the activation of auditory cortex in response to changes in continuous auditory input.

Visual unimodal areas showed a significant deactivation in response to transitions in other modalities. This finding is consistent with a study showing that attention to tactile stimuli deactivates visual association areas 13. Our study indicates that attention-drawing transitions in the auditory modality can also deactivate visual association cortex. Such intermodal inhibition did not seem to affect auditory or somatosensory association areas, however. The reasons for the unique vulnerability of visual association cortex to transition-related intermodal inhibition remain to be investigated. The largest of the multimodal nodes of activation was found in the right TPJ. The next highest volumes of activation were in the right inferior frontal cortex, the left SMA and CMA and right anterior insula, with smaller activations in the left TPJ and inferior frontal cortex, the right posterior insula and the right middle temporal gyrus. Nonetheless, the total volume of activated voxels in the two hemispheres, the specific areas activated and their order of prominence show remarkable agreement with neuroanatomical correlates of sensory neglect. In this syndrome, the ability to attend to salient stimuli on one side of the body (usually the left), whether voluntarily or involuntarily, is lost as a result of neurological damage. Patients suffering from neglect

Table 1. Brain regions showing significant activation in response to transitions in visual, auditory, tactile and all three sensory modalities. Brain region

Brodmann area

Coordinates x y z

Volume (mm3)

t

Unimodal activations Visual R fusiform gyrus L fusiform gyrus R middle occipital gyrus L middle occipital gyrus R superior parietal lobule

19/37 19/37 18/19 18/19 18/19

29 –31 31 –29 25

–60 –63 –81 –81 –62

–14 –15 4 7 47

5835 5689 2371 2912 801

16.97 15.75 16.97 17.53 12.56

Auditory R superior temporal gyrus L superior temporal gyrus L superior temporal gyrus

22/41/42 22/42 41/42

53 –51 –53

–21 –37 –15

5 10 3

2185 1858 403

9.66 10.32 8.97

40 40

50 –59

–28 –28

25 27

2177 1259

9.15 8.34

22/39/40 22/39 21 44/45 6/44 6/32

54 –54 57 43 –51 –8 27 36

–42 –48 –57 6 –8 4 18 –16

13 10 2 5 5 41 10 6

3657 449 338 822 226 569 564 215

12.16 8.25 7.03 8.73 7.41 7.55 8.17 5.97

Tactile R postcentral gyrus (S2) L postcentral gyrus (S2) Multimodal activations R temporoparietal junction L temporoparietal junction R middle temporal gyrus R inferior frontal gyrus L inferior frontal gyrus L SMA/CMA R anterior insula R posterior insula

Activations shown are based on a voxelwise p < 0.00032 (unimodal transitions) or p < 0.00036 (multimodal transitions) and a minimum cluster size of 200 mm3. Average time course of all voxels in each cluster were used to calculate t-values. For multimodal areas, t-values are converted from r-coefficients. S2, secondary somatosensory cortex; SMA, supplementary motor area; CMA, cingulate motor area.


279


articles

also show a profound, multimodal loss of awareness of stimuli in the neglected field, concurrent with their attentional deficits2,15. Hemineglect is far more commonly observed following damage to the right hemisphere than the left. Lesions of the right TPJ are the most common neuroanatomical correlate of hemineglect1,16,17, followed by lesions of the CMA, SMA or prefrontal cortex, especially in the vicinity of Brodmann area 44 (ref. 1). It should be remembered that fMRI signal provides only an indirect measure of the neural activity of a given area. Nonetheless, the remarkable degree of overlap between the known anatomical correlates of hemineglect and the network identified in our study emphasizes the close relationship between the detection of salient events in the sensory environment and the selection of these salient stimuli for entry into awareness. Furthermore, it underscores the importance of the identified cortical regions to the performance of both functions across multiple sensory modalities. The observation of SMA and CMA activation in this study is of particular interest. These areas are generally considered to be

Visual transitions Auditory transitions

All transitions

Tactile transitions

involved in the planning of motor responses, not in the shifting of attention or the detection of sensory events. Previously reported CMA and SMA activation during spatial shifts of attention might be attributed to the planning and execution of motor responses during task performance8. However, we observed SMA and CMA activation even though our subjects made no such responses during the course of the experiment. The activations we observed in CMA and SMA could be attributed to the planning of unexecuted, involuntary motor orienting responses to the changing sensory input. A more interesting possibility, however, is that these regions may have an attentional role in senso-

Fig. 3. Responses to visual, auditory and tactile transitions. Averaged event-related hemodynamic responses to transitions in each stimulus modality are shown for representative unimodally and multimodally responsive areas. (a) Bilateral fusiform gyrus. (b) Bilateral superior temporal gyrus. (c) Bilateral secondary somatosensory cortex (S2). (d) Bilateral TPJ. Unimodal areas (a–c) respond only to transition events within visual, auditory and tactile modalities, respectively, whereas the TPJ (d) responds bilaterally to transitions in all three modalities. A significant (p = 0.01) deactivation of the fusiform gyrus in response to auditory and tactile transitions is visible in (a). Multimodal activation in the right anterior insula, not shown in Fig. 2, is visible in (d). The Talairach z coordinate of each slice is indicated in the upper left-hand corner. Error bars indicate s.e.

c Change in signal (%)

Change in signal (%)

a

Post-transition time (s)

b


280


d Change in signal (%)



Fig. 2. Surface rendering of brain regions activated by transitions of the visual, auditory and tactile stimuli. Note that unimodal activations are bilateral and correspond predominantly to association cortices, whereas multimodal activations are strongly lateralized to the right hemisphere and correspond to the temporoparietal junction, inferior frontal gyrus and insula. On the left medial wall, the supplementary and cingulate motor areas are also prominently activated, even though subjects were not required to make responses or movements during task performance. Activations represent averaged data from 10 subjects, superimposed on the standardized brain of one subject. L, left; R, right.




articles

Temporoparietal junction

SMA/CMA

Middle temporal gyrus

Fig. 4. Multimodally responsive brain regions. The plane coordinate of each slice is indicated in the upper left-hand corner. SMA, supplementary motor area; CMA, cingulate motor area.

ry processing as well as their more generally accepted role in the planning and execution of movements. This view is supported by the observation that lesions of the SMA and CMA can give rise to attentional deficits such as hemineglect1. In addition, recordings of epicortical field potentials in humans demonstrate the involvement of the SMA, CMA and other premotor areas in the anticipation of and attention to forthcoming stimuli18. Our study cannot conclusively demonstrate the involvement of premotor areas in attentional processes independent of response planning; however, the possibility bears further investigation. The involvement of the TPJ and inferior frontal gyrus in detecting changes in continuous streams of sensory input is consistent with EEG studies of the effects of focal cortical lesions on P300 event-related potentials (ERPs), which are widely used as markers of attention to salient sensory stimuli19. Lesions of the TPJ produce a marked reduction in the amplitude of both the P3a ERP, which is normally elicited by unexpected, task-irrelevant stimuli, and the P3b ERP, which is normally elicited by expected, task-relevant stimuli20,21. Such reductions are observed for visual, auditory and tactile stimuli following injury of the TPJ21–23. Lesions of the prefrontal cortex also reduce the P3a ERP for all three modalities 21,24. However, neither P3a nor P3b is

Fig. 5. Responses of the right TPJ to each direction of transition in each modality. Averaged event-related hemodynamic responses to transitions from stimulus A to stimulus B as well as stimulus B to stimulus A are shown for each sensory modality. Activations for both transition types exclude the possibility that the event-related activations in Fig. 3d reflect an increase in stimulus intensity or contrast associated with just one of the two types of transitions. Similar activation for both types of transitions was verified for all unimodal and multimodal areas reported. nature neuroscience • volume 3 no 3 • march 2000

affected by lesions of the parietal cortex immediately dorsal to the TPJ21,23,24. These EEG findings support the results of our study in suggesting the importance of TPJ and prefrontal cortex in the detection of salient stimuli across multiple modalities. There remains the possibility that additional multimodal areas were not detected in our study because their responses habituated rapidly over the first few transitions. For example, although activation in the dorsolateral prefrontal cortex was not observed in our study, evidence from EEG studies of patients with focal cortical lesions suggests that this area may be involved in detecting novel sensory events in visual, auditory and tactile modalities14,21,24. Similarly, a study using fMRI indicates that novel auditory events activate the right prefrontal cortex25. In addition,



Anterior/posterior insula

Inferior frontal gyrus


281



articles

a number of studies implicate a set of rapidly habituating areas, including prefrontal, posterior parietal and hippocampal areas, in the detection of novel events and their subsequent encoding into memory26,27. Furthermore, evidence from EEG studies using taskrelevant ‘target’ and task-irrelevant ‘nontarget’ oddballs suggests that the context in which task-irrelevant stimuli are presented, more generally than their novelty, may affect the physiological response28. Our study was not designed to identify areas responsive only to novel events. This issue will be the focus of future investigations. A widespread, multimodal cortical and subcortical network for the shifting of attention to salient features of the sensory environment was first proposed on the basis of neuroanatomical and lesion studies29. The results of our study provide an illustration of the unimodal and multimodal elements of such a network in action. The agreement of these results with previous EEG and lesion studies provides converging evidence for a close relationship between the detection of salient stimuli in the multisensory environment and the entry of those stimuli into awareness, a process mediated by a widespread network of cortical areas comprising the unimodal sensory association cortices and the multimodal TPJ, CMA, SMA, insula and inferior frontal gyrus.

METHODS Subjects. Subjects were 6 male and 4 female right-handed individuals aged 18–41 (mean age ± s.d., 27.3 ± 7.2), with no history of neurological injury. All subjects gave written informed consent for the experimental procedures, approved by the University of Toronto Human Subjects Review Committee. Stimuli. Visual, auditory and tactile stimuli were presented simultaneously during functional imaging. Visual stimuli were back-projected onto a screen viewed by the subject through a mirror incorporated into the head coil. The visual stimuli were two simple, abstract shapes, one red and one blue (Fig. 1). Auditory stimuli were generated by two pneumatic transducers and conducted through a pair of eight-mm plastic tubes into a set of headphones placed over the subject’s ears. The auditory stimuli were the sound of running water and the sound of croaking frogs. Tactile stimuli delivered with a 6 × 10 cm shower brush to the subject’s lower right leg consisted of circular brushing with the bristles (∼1 Hz) and steady tapping with the smooth back of the brush (∼3 Hz). Visual and auditory stimuli were produced using a VCR connected to the projector and audio transducer. A visual signal, projected outside the subject’s field of view, prompted the experimenter to change the tactile stimulation type. During the experiment, a total of 15 visual, 15 auditory and 14 tactile transitions occurred in a pseudo-random sequence, spaced at 14-s intervals to allow the BOLD signal to return to baseline between transitions. Subjects were instructed simply to observe the stimuli passively. After imaging, all subjects reported that the transitions had drawn their attention and that they had not counted the transitions. Imaging. A 1.5 T MRI system (Echospeed, GE Medical Systems, Milwaukee, Wisconsin) and a standard quadrature head coil were used to obtain all images. For anatomical images, a 3-dimensional SPGR sequence (flip angle = 45°, TE = 5 ms, TR = 25 ms) was used to generate 124 sagittal slices (1.5 mm thick) with 256 × 256 matrix size and 30 × 30 cm field of view resulting in an anatomical resolution of 1.5 × 1.17 × 1.17 mm. For analysis, the anatomical images were resampled to 1 × 1 × 1 mm using sinc interpolation. For functional imaging, 25 contiguous axial slices (4 mm thick) were chosen to provide whole-brain coverage. Functional images were acquired with a gradient echo sequence using a singleshot spiral trajectory through k-space30, flip angle = 85°, TE = 40 ms, TR = 2000 ms, a 64 × 64 matrix size and a 20 × 20 cm field of view for a functional resolution of 3.125 × 3.125 × 4 mm and an inter-frame interval of 2 s. The first 3 frames were discarded to allow for signal equilibration, giving a total of 314 frames used in analysis. 282

Data processing and analysis. Preprocessing, volumetry and analysis were performed using BrainVoyager 3.5 (Brain Innovation, Frankfurt, Germany), implemented on a 450 MHz Pentium II platform. Functional data were corrected for within-frame time of acquisition, high-pass filtered with a cut-off at 0.0175 Hz to remove slow drifts in signal intensity (period >1 min), coregistered with the 3-dimensional anatomical images, transformed into standard stereotactic space31 and resampled at a resolution of 3 × 3 × 3 mm. Data were smoothed using a Gaussian kernel of 4 mm full width at half maximum to accommodate anatomical and functional-anatomical variation between subjects. Individual subjects’ data were averaged together for group analysis. During the generation of statistical maps, data were further interpolated to 1 × 1 × 1 mm resolution to facilitate estimation of the volume of activation in each area. To identify unimodally responsive areas, we used a voxelwise t-test to compare the BOLD signal two to eight s after a transition within a given modality to the signal two to eight s after transitions in the other modalities. We used an initial threshold t > 3.70 for individual 1-mm3 voxels (n = 130 frames, uncorrected p = 0.00032). To identify multimodally responsive areas, we used a voxelwise correlation to the predicted hemodynamic response to brief stimulus events spaced 14 s apart, based on previous studies4,6. We used an initial threshold of r > 0.20, equivalent to t > 3.61 (n = 314 frames, uncorrected p = 0.00036). The total volume of each statistical map was 1,492,559 mm3; hence, at the thresholds used, ∼400 voxels in each entire map would be expected to show activation due to type I errors. These voxels would also be expected to display a certain degree of clustering, as they were interpolated from 3 × 3 × 3 mm-resolution, smoothed functional data. We therefore required a conservative minimum cluster size of 200 contiguous 1-mm3 voxels for all reported activations. We initially examined the event-related average response of each area to each transition direction separately (transitions from A to B were considered separately from transitions from B to A) to rule out areas whose responses, ostensibly to transitions per se, merely reflected the different physical characteristics of the stimuli presented (for instance, intensity, contrast). In such a scenario, the areas observed might show a large or rapid activation for the transition from A to B but only a small or gradual deactivation for the opposite transition from B to A. The average response would seem to be an activation for all transitions, even though it truly reflected only the different properties of the two stimuli. We excluded this possibility by analyzing the responses of all areas to the two kinds of transitions separately, as in Fig. 5. For all unimodal and multimodal areas presented, we verified activation in response to transitions from A to B as well as B to A. In addition, for each multimodal area, we examined the event-related average response to transitions in each modality as in Fig. 3. Areas that did not respond to transitions in all three sensory modalities were discarded from the list of multimodal activations. This additional step was necessary to exclude areas that merely showed an exceptionally strong response to transitions in one of the three modalities rather than a consistent response across all three.

ACKNOWLEDGEMENTS This study was supported by grants to K.D.D. from the Whitehall Foundation and the Medical Research Council of Canada.

RECEIVED 23 AUGUST; ACCEPTED 22 DECEMBER 1999 1. Vallar, G. Spatial hemineglect in humans. Trends Cogn. Sci. 2, 87–97 (1998). 2. Halligan, P. W. & Marshall, J. C. Neglect of awareness. Conscious. Cogn. 7, 356–380 (1998). 3. Mesulam, M. M. From sensation to cognition. Brain 121, 1013–1052 (1998). 4. Pollmann, S., Wiggins, C. J., Norris, D. G., von Cramon, D. Y. & Schubert, T. Use of short intertrial intervals in single-trial experiments: a 3T fMRI-study. Neuroimage 8, 327–339 (1998). 5. Buckner, R. L. Event-related fMRI and the hemodynamic response. Hum. Brain Mapp. 6, 373–377 (1998). 6. Aguirre, G. K., Zarahn, E. & D’Esposito, M. The variability of human, BOLD hemodynamic responses. Neuroimage 8, 360–369 (1998). 7. Sergent, J., Ohta, S. & MacDonald, B. Functional neuroanatomy of face and object processing. A positron emission tomography study. Brain 115, 15–36 (1992).




articles

8. Kim, Y. H. et al. The large-scale neural network for spatial attention displays multifunctional overlap but differential asymmetry. Neuroimage 9, 269–277 (1999). 9. Robinson, C. J. & Burton, H. Somatotopographic organization in the second somatosensory area of M. fascicularis. J. Comp. Neurol. 192, 43–67 (1980). 10. Dong, W. K. et al. Nociceptive responses of trigeminal neurons in SII-7b cortex of awake monkeys. Brain Res. 484, 314–324 (1989). 11. Maldjian, J. A. et al. Mapping of secondary somatosensory cortex activation induced by vibrational stimulation: an fMRI study. Brain Res. 824, 291–295 (1999). 12. Tong, F., Nakayama, K., Vaughan, J. T. & Kanwisher, N. Binocular rivalry and visual awareness in human extrastriate cortex. Neuron 21, 753–759 (1998). 13. Kawashima, R., O’Sullivan, B. T. & Roland, P. E. Positron-emission tomography studies of cross-modality inhibition in selective attentional tasks: closing the ‘mind’s eye’. Proc. Natl. Acad. Sci. USA 92, 5969–5972 (1995). 14. Alain, C., Woods, D. L. & Knight, R. T. A distributed cortical network for auditory sensory memory in humans. Brain Res. 812, 23–37 (1998). 15. Driver, J. & Mattingley, J. B. Parietal neglect and visual awareness. Nat. Neurosci. 1, 17–22 (1998). 16. Ro, T., Cohen, A., Ivry, R. B. & Rafal, R. D. Response channel activation and the temporoparietal junction. Brain Cogn. 37, 461–476 (1998). 17. Friedrich, F. J., Egly, R., Rafal, R. D. & Beck, D. Spatial attention deficits in humans: a comparison of superior parietal and temporal-parietal junction lesions. Neuropsychology 12, 193–207 (1998). 18. Ikeda, A. et al. Cognitive motor control in human pre-supplementary motor area studied by subdural recording of discrimination/selection-related potentials. Brain 122, 915–931 (1999). 19. Coull, J. T. Neural correlates of attention and arousal: insights from electrophysiology, functional neuroimaging and psychopharmacology. Prog. Neurobiol. 55, 343–361 (1998).


20. Yamaguchi, S. & Knight, R. T. P300 generation by novel somatosensory stimuli. Electroencephalogr. Clin. Neurophysiol. 78, 50–55 (1991). 21. Swick, D. & Knight, R. T. in The Attentive Brain (ed. Parasuraman, R.) 144–162 (MIT Press, Cambridge, Massachusetts, 1998). 22. Knight, R. T., Scabini, D., Woods, D. L. & Clayworth, C. C. Contributions of temporal-parietal junction to the human auditory P3. Brain Res. 502, 109–116 (1989). 23. Yamaguchi, S. & Knight, R. T. Anterior and posterior association cortex contributions to the somatosensory P300. J. Neurosci. 11, 2039–2054 (1991). 24. Knight, R. T. Decreased response to novel stimuli after prefrontal lesions in man. Electroencephalogr. Clin. Neurophysiol. 59, 9–20 (1984). 25. Opitz, B., Mecklinger, A., Friederici, A. D. & von Cramon, D. Y. The functional neuroanatomy of novelty processing: integrating ERP and fMRI results. Cereb. Cortex 9, 379–391 (1999). 26. Knight, R. T. & Nakada, T. Cortico-limbic circuits and novelty: a review of EEG and blood flow data. Rev. Neurosci. 9, 57–70 (1998). 27. Tulving, E., Markowitsch, H. J., Kapur, S., Habib, R. & Houle, S. Novelty encoding networks in the human brain: positron emission tomography data. Neuroreport 5, 2525–2528 (1994). 28. Katayama, J. & Polich, J. Stimulus context determines P3a and P3b. Psychophysiology 35, 23–33 (1998). 29. Mesulam, M. M. A cortical network for directed attention and unilateral neglect. Ann. Neurol. 10, 309–325 (1981). 30. Glover, G. H. & Lee, A. T. Motion artifacts in fMRI: comparison of 2DFT with PR and spiral scan methods. Magn. Reson. Med. 33, 624–635 (1995). 31. Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain (Thieme Medical, New York, 1988).

283


articles

The neural mechanisms of topdown attentional control J. B. Hopfinger1, M. H. Buonocore2 and G. R. Mangun3 1

Center for Neuroscience and Department of Psychology, One Shields Ave.,University of California, Davis, Davis, California 95616, USA

2

Department of Radiology, University of California, Davis Medical Center, 4701 X St., Sacramento, California 95817, USA

3

Center for Cognitive Neuroscience, LSRC Bldg., Rm. B203, Duke University, Durham, North Carolina 27708, USA


Correspondence should be addressed to G.R.M. ([email protected])

Selective visual attention involves dynamic interplay between attentional control systems and sensory brain structures. We used event-related functional magnetic resonance imaging (fMRI) during a cued spatialattention task to dissociate brain activity related to attentional control from that related to selective processing of target stimuli. Distinct networks were engaged by attention-directing cues versus subsequent targets. Superior frontal, inferior parietal and superior temporal cortex were selectively activated by cues, indicating that these structures are part of a network for voluntary attentional control. This control biased activity in multiple visual cortical areas, resulting in selective sensory processing of relevant visual targets.

Selective attention enables us to focus awareness on objects and events that are relevant to our immediate goals. Spatial attention, the selective direction of visual attention toward a location, can occur covertly, without overt movements of the head or eyes. Theoretically, mechanisms of covert, voluntary spatial attention can be decomposed into elementary mental operations: disengaging attention from the current focus, orienting attention to a new locus and selectively modulating new stimulus inputs are three stages proposed in current models1,2. Attentional disengagement and voluntary orienting can be considered aspects of top-down attentional control, whereas subsequent selective modulation of sensory inputs reflects the result of this top-down control on sensory information processing. Studies in neurological patients and physiological studies in humans and animals implicate a network of cortical and subcortical regions that support visual selective attention 3–6. Neuroimaging studies identify areas involved in spatial attention in the frontal, parietal, temporal and occipital lobes as well as in subcortical structures7–11. However, although neuroimaging studies implicate various brain regions in selective attention, inherent limitations of classical neuroimaging analysis impair efforts to relate neural structures/networks to specific attentional operations. Specifically, such methodologies use averaging of hemodynamic responses over seconds or minutes, a time course that is too long to permit direct viewing of brain activity related to the subcomponents of task performance. As a result, human neuroimaging studies to date are only partly successful in distinguishing between neural activity related to top-down attentional-control processes and activity related to selective sensory and motor processing. Recordings of event-related neuroelectric potentials (ERPs) allow the selective averaging of responses to different classes of events and, thus, serve to index mechanisms of attentional control separately from selective stimulus processing during spatial attention12. However, although ERPs provide information about the timing of neural processes, the limited spatial resolution of this approach hinders identification of the neural structures involved in attentional control. Advances in functional magnetic resonance imaging (fMRI) 284

analysis13,14 allowed us to combine the spatial resolution necessary for localization of neural activity, which this technique provides, with neuroimaging methods that selectively extract components of hemodynamic activity15 correlated with distinct aspects of complex-task performance. Here we used event-related fMRI methods to determine which brain regions were involved in attentional-control processes and to distinguish these from areas involved in subsequent selective processing of target stimuli. This was accomplished by separately convolving the onsets of the cue and target stimuli with synthetic hemodynamic response functions. Subjects oriented attention covertly (without moving their eyes) to spatial locations based on instructive cues shown at fixation (Fig. 1). Subsequently, reversing checkerboard stimuli were presented at both sides of the visual field, and subjects were required to discriminate whether the checkerboard at the cued location contained some gray checks or only black and white checks and to respond accordingly. All stimuli in the opposite (uncued) hemifield were to be ignored. Based on previous studies implicating various brain regions in attentional control5,7,8,11, we hypothesized that attentional-control processes in response to an instructive cue directing spatial attention would activate a network of regions in frontal and parietal cortex. Additionally, based on ERP studies in humans12, regions of occipital cortex corresponding to portions of visual space indicated by the cue should show spatially specific changes in activity in anticipation of selective processing of target stimuli. We also expected lateralized changes in activity of occipitotemporal regions in response to the bilateral target stimuli to reflect spatial attention7–11. In line with these predictions, we found distinct patterns of neural activation for attention-directing cues as compared with those for subsequent target stimuli, thereby distinguishing topdown attentional control from selective modulations of target processing in sensory structures.

RESULTS Subjects discriminated black and white checkerboards from those containing some gray checks at the cued location with a mean nature neuroscience • volume 3 no 3 • march 2000


articles

Fig. 1. Stimuli and timing. For simplicity, we show here the inverse black/white contrast of the actual stimulus screen; subjects saw a black background with white outline boxes and fixation cross, and the cues were overlapping isoluminant yellow and blue arrows (pointing in opposite directions). Subjects were told which color arrow to attend for the entire session and were required to covertly orient their attention to the visual field location indicated by that arrow. After a variable interstimulus interval (1000 or 8160 ms), the target checkerboards appeared bilaterally, and subjects were required to perform a discrimination of the checkerboard stimuli at the attended location only.

Cue 500 ms +

ISI 1000 ms (17%) or 8160 ms (83%) +

Target

Time

750 ms 4 Hz reversal


+

accuracy of 83.9% correct responses. This corresponded to a mean d′ score of 2.99. We investigated circuitry involved in attentional control by identifying regions activated in response to the cue. Multiple cortical areas were activated following cue presentation, but before target presentation (Fig. 2; Table 1). These regions were activated regardless of the cue direction: brain regions activated by rightward- and leftward-pointing cues were highly similar (with the exception of the extrastriate cortex; see below). Regions of the inferior parietal lobule (in the lateral intraparietal sulcus, or IPS), the superior parietal lobule (SPL) and the posterior cingulate cortex (PC) in both hemispheres were activated in response to the instructive cues. Regions of the lateral and medial superior frontal lobes, including the frontal eye fields (FEF), were also activated bilaterally. These

Leftward L SFG L MFG

L SPL L IPS

L MFG L SFG

7910 ms +

frontal activations tended to be larger in the left hemisphere, regardless of cue direction. In addition, cue stimuli generated activity along the superior temporal sulcus (STS). With the exception of the superior parietal lobule, none of these regions were activated in separate sensory control scans in which cues were viewed passively or were irrelevant to the task (and, therefore, did not instruct the subjects to direct attention to a single location). Posterior occipital regions of cortex were activated in response to the cues, but these regions were also activated by cues in the sensory-control scans, indicating that they represented simple sensory processing of the cues. Additionally, the cues produced bilateral activations in the region of Cues the insula near the putamen. Rightward In contrast, the target stimuli evoked neural activity in brain areas that were largely distinct from those activated by the L SFG attention-directing cues (Fig. 3; Table 2). Bilateral activations in the supplementary R MFG L MFG R MFG motor area (SMA) extending into regions of the midcingulate gyrus and activations R SPL surrounding the central sulcus were found L SPL R SPL R IPS R IPS in response to targets whether attention L IPS was directed to the right or left. Ventrolateral prefrontal areas were also bilaterally activated by the targets. None of these regions were activated by the cues. Only L MFG L SPL L SPL the superior parietal lobule was found to L SFG

L IPS

L IPS

L LOG

L LOG L STS/STG

L STS/STG

L SPL

L SPL

L SFG

L SFG L Post. Cingulate

ITI

L Post. Cingulate

Fig. 2. Activity related to attentional control. Data for brain regions significantly activated in response to the cue stimuli were overlaid onto a brain rendered in 3D. Left column, activations to cues instructing subjects to orient attention to the left visual field location. Right column, activations to cues instructing subjects to attend the right visual field location. Top panels, dorsal view of the brain (frontal pole at top); middle and bottom panels, lateral and medial views, respectively, of the left hemisphere. Labels indicate the brain regions referred to in Table 1. The Z-values and stereotactic coordinates for the regional maxima are given in Table 1. SFG, superior frontal gyrus; MFG, middle frontal gyrus; SPL, superior parietal lobule; IPS, intraparietal sulcus; STS, superior temporal sulcus; STG, superior temporal gyrus; LOG, lateral occipital gyrus.


285



articles

be activated in response to both targets and cues. Additionally, bilateral targets activated several regions of the ventral and dorsal occipital cortex of both hemispheres. Cue and target responses were directly statistically compared (Tables 1 and 2; Fig. 4). To investigate the consequences of lateralized spatial attention on stimulus processing, the event-related neural activity evoked by the bilateral target stimuli was evaluated for attention to the left versus right. In line with previous neuroimaging studies in humans9,10,16–19 and related findings in animals3, attending to the left visual field increased activity in the right ventral occipital cortex, whereas attending to the right increased activity in the left occipital cortex (Fig. 5a). In addition, activity in a more dorsal region of visual cortex was also modulated by spatial attention. To relate these attention-related activations to specific visual-field areas, the vertical and horizontal meridian were mapped in two subjects to identify the borders of the early cortical visual areas20,21. Although the spatial smoothing used in the present analysis tended to blend together the attention-related activations in adjacent visual cortical areas, the contralateral attentionrelated activations in occipital cortex can be seen to cross multiple visual area borders from V2 through V4 (Fig. 5b), an observation in line with other reports10,18,22. Models of spatial attention propose that attention effects on target processing result from a gain-control mechanism that enhances the excitability of extrastriate neurons coding attended regions of visual space4,9,23. To investigate this possibility, we further examined extrastriate cortex for evidence of spatially

selective attentional activations in response to the direction of the instructive cue, but before target presentation, using the attention-related activations to targets as regions of interest. Direct statistical evaluation of the cue-left versus cue-right effects revealed relative increases of activity in extrastriate cortex contralateral to the direction of attention. These contralateral cuerelated activations overlapped closely with the extrastriate regions showing attentional modulations in response to the subsequent targets (Fig. 6). That is, there was a relative increase in activity in the visual cortical regions representing the spatial locations of the expected target stimuli before the target stimuli were actually presented. These differential activations were not related to inherent sensory features of the cues that activated separate cortical regions in control sessions.

DISCUSSION This study used event-related fMRI methods to distinguish between neural networks involved in top-down attentionalcontrol processes and those participating in the subsequent spatially selective attentional processing of target stimuli. A network of cortical areas including superior frontal, inferior parietal and superior temporal brain regions were implicated in top-down attentional control because they were found to be active only in response to instructive cues. In contrast, other regions of the cortex, including the ventrolateral prefrontal cortex, anterior cingulate and supplementary motor area, were found to be selectively activated by the target stimuli, suggesting that these

Table 1. Event-related activations to cue stimuli and statistical contrasts. Leftward-directing cue Coordinates Z-score x y z Frontal Left SFG* Left SFG (lateral)* Left MFG* Right MFG*

Rightward-directing cue Coordinates Z-score x y z

Cues > targets Coordinates Z-score p x y z (corrected)

–16 –44 –20 24

48 36 –4 0

36 16 52 56

4.56 2.96 4.36 3.96

–16 –44 –20 24

48 36 0 4

36 16 48 52

4.10 4.30 5.53 4.37

–16 –52 –32 20

48 32 16 8

36 16 48 48

7.10 6.93 7.67 4.95

p < 0.001 p < 0.001 p < 0.001 p < 0.01

–16 24 –44 –44 Right IPS* 40 36 L cingulate (posterior)* –12 R cingulate (posterior) 8

–52 –40 –64 –48 –68 –48 –44 –36

56 56 32 36 28 36 28 28

3.27 3.84 3.79 3.29 4.23 3.46 3.53 2.94

–16 20 –40 –40 40 32 –8 4

–56 –40 –64 –48 –64 –48 –44 –40

56 52 28 32 32 32 32 28

4.21 2.76 4.06 3.88 3.83 4.23 2.91 3.48

–12 8 –44 –44 44 36 0 4

–56 –40 –64 –48 –60 –44 –40 –44

56 56 32 32 36 32 44 32

4.41 4.27 7.01 7.29 6.63 7.22 7.64 7.63

p > 0.05 p > 0.05 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001

Parietal Left SPL* Right SPL* Left IPS*

Temporal Left STS/STG* Right STS/STG

–60 48

–24 –12

8 8

3.53 1.85

–56 48

–24 –12

8 8

2.39 1.76

–56 48

–24 –12

8 8

6.71 5.28

p < 0.001 p < 0.005

Occipital Left LOG* Right LOG

–36 32

–76 –80

8 8

4.05 2.82

–36 24

–76 –84

4 8

4.42 5.77

–36 28

–76 –80

4 0

8.12 8.58

p < 0.001 p < 0.001

Other regions L insula/putamen R insula/putamen

–28 28

–4 –8

12 16

4.17 2.77

–28 24

–8 0

16 16

4.56 3.96

–28 24

–8 –8

20 12

5.53 5.28

p < 0.001 p < 0.005

SFG, superior frontal gyrus; MFG, middle frontal gyrus; SPL, superior parietal lobule; IPS, intra-parietal sulcus; STS, superior temporal sulcus; STG, superior temporal gyrus; LOG, lateral occipital gyrus. Coordinates: x, left/right; y, anterior/posterior; z, inferior/superior in the reference frame of the MNI brain in SPM97. *Activity represented in Figs. 2 and 4.

286




articles

areas may be more involved in Targets selective stimulus processing Attend left Attend right and/or response mechanisms. The activation in the inferior SMA/cingulate SMA/cingulate parietal lobule (specifically, IPS) in response to the cues, but not targets, R preCG suggests that this region of parietal L preCG cortex is involved in attentional-conL preCG R preCG trol processes. This finding supports R postCG prior evidence from blocked-design L postCG R postCG imaging studies, which inferred the L postCG R SPL L SPL role of inferior parietal cortex based L SPL on activity averaged across blocks of R SPL trials8,11,24,25, as well as work in animals26, but provides the first direct L postCG L postCG L preCG L preCG evidence that these regions of the inferior parietal lobule are specifiL SPL L SPL cally linked to attentional control in humans. The present findings in inferior parietal cortex suggest a role for this brain region in attentional control mechanisms, which may L CUN L CUN include shifting attention 7,24, or L IFG L IFG working memory processes engaged to support task performance27. In contrast, regions in superior parietal cortex, also proposed to be SMA/cingulate SMA/cingulate involved in attentional orienting7, were not only activated by the cue stimuli, but were also (and to a greater extent) activated in response to the target stimuli. This finding agrees with neuroimaging studies L LG/FG L LG/FG reporting that regions of SPL are active both during shifts of attention in the visual periphery and when attention is focused on target stimuli Fig. 3. Activity related to target processing. Brain regions significantly activated in response to the target but no attentional shifts are stimuli, overlaid onto a 3D rendering of a brain, as in Fig. 2. Left column, activations to target stimuli when 28 required . These findings fit well attention was focused on the left visual field location. Right column, shows activations to target stimuli with neurological lesion studies when attention was focused on the right visual field location. The labels indicate the brain regions referred finding that inferior regions of pari- to in Table 2. The Z-values and stereotactic coordinates for the regional maxima are listed in Table 2. SMA, etal cortex (extending into the tem- supplementary motor area; preCG, pre-central gyrus; postCG, post-central gyrus; IFG, inferior frontal gyrus; poroparietal junction) are more MFG, middle frontal gyrus; SPL, superior parietal lobule; CUN, cuneus; LG, lingual gyrus; FG, fusiform gyrus. critical than superior parietal regions for normal performance in spatial cueing tasks29. Neuroimaging studies in humans reveal activations in the posthe cortex in the vicinity of the FEF was activated by the cues, but terior STS during the performance of attentional tasks, suggestnot by the target stimuli, suggesting its participation in attentionaling that the STS may be involved in attentional processing in control processes in the absence of eye movements. Given that eye humans11,24, a role supported by induction of attentional neglect movements needed to be suppressed to both cue and target presentations, the present pattern cannot be attributed solely to supin lesions of the STS in monkeys30. However, it has remained pression of oculomotor responses33. unclear whether the STS is involved in attentional control circuitry, or instead is involved in attention-related processing of Regions of the dorsolateral prefrontal cortex (superior frontal relevant target stimuli. We demonstrated that a region of the STS gyrus) that were anterior to the FEF activations were also actianterior to the temporal-parietal junction was selectively activated to the cues but not the targets. These regions of frontal corvated by the attention-directing cues, suggesting a role in attentex have been associated with working memory processes in tional control circuitry. studies of spatial working memory34–38. In the present task, such The activations to the attention-directing cues in the vicinity of processes may have activated these regions as subjects encoded the frontal eye field in the present study suggest a role for this brain the to-be-attended location in working memory. region in attentional control. The FEF is proposed to be involved In contrast, although ventrolateral prefrontal regions are in attentional processes, specifically as they relate to overt eye movesometimes considered to be involved in retention of information ments5,31. Imaging studies, however, suggest that FEF may be in working memory34,35, in the present study, the strongest veninvolved in covert shifts of attention as well24,32. In the present study, trolateral prefrontal activations were not observed in response to nature neuroscience • volume 3 no 3 • march 2000

287



articles

the cue stimuli when one might expect working memory processes to be engaged. Rather, these cortical regions were activated upon presentation of the target stimuli. This finding may be consistent with a role in inhibitory filtering of information from the ignored hemifield when the bilateral targets appeared36–38. Previously, the supplementary motor area and the anterior cingulate cortex were hypothesized to be part of attentional control systems2,8. The present finding that these areas were activated in response to the target stimuli but not the attention-directing cues, however, suggests a role in either the selective analysis of target features or decisional processes engaged once the relevant stimulus is presented, such as selecting appropriate motor actions based on target features39,40. As predicted by the previous literature, target-evoked activity in posterior lingual and fusiform regions was enhanced in the hemisphere contralateral to the direction of attention. This supports the idea that spatial attention-related increases in regional blood flow in extrastriate cortex are related to enhanced processing of target stimuli falling within the attentional spotlight, as suggested by prior ERP4,41,42 and functional imaging studies 9,10,16–19 in humans and single-unit studies in monkeys3,43. In addition, a more dorsal region of visual cortex was also enhanced by the direction of attention. This region may be area V3a, which has an upper visual field representation in dorsal visual cortex44 and which is activated in response to flashing checkerboard patterns18. Priming of visual neurons by top-down attentional control signals is suggested by the finding that regions of visual cortex

coding to-be-attended locations were differentially activated in response to the instruction to orient attention by the cue. Singleunit studies in non-human primates show an increase in background firing rates of neurons coding an attended region of space in the absence of stimuli45, and proposals based on neuroimaging data suggest that increased background firing rates may mediate attentional facilitation in visual cortex46,47. The present data show a precise spatial correspondence between areas of visual cortex activated by top-down processes and those showing subsequent selective modulations of target processing. Moreover, these priming effects in visual cortex were shown to follow the retinotopic mapping of the visual fields onto visual cortex as attention was shifted from left to right locations by the cue instructions. In summary, a top-down attentional control system was isolated using event-related fMRI methods. This network included the superior frontal cortex, inferior parietal cortex, superior temporal cortex and portions of the posterior cingulate cortex and insula. The present evidence indicates that this top-down control system modulated activity in extrastriate cortex as a function of where spatial attention was directed in the visual field. These spatially selective effects in extrastriate cortex, observed before the presentation of the target stimuli, resulted in modulations in the activity evoked by subsequent target stimuli in precisely corresponding early cortical areas. Hence, it seems that effects of spatial attention in visual cortex before target onset may reflect changes in sensory-neural excitability as a function of top-down control. These findings provide direct evidence for clear distinctions between the neural mech-

Table 2. Event–related activations to target stimuli and statistical contrasts.

Attended left target Coordinates x y z Frontal SMA/cingulate** Left preCG** Right preCG**

Z-score

Attended right target Coordinates x y z

Z-score

Targets > cues Coordinates x y z

Z-score

p (corrected)

Left IFG** Left insula/IFG** Right IFG/MFG Right insula/IFG

4 –40 32 40 –56 –36 44 32

8 –16 –16 –4 12 28 12 24

52 56 52 52 32 4 24 4

7.04 7.26 5.64 6.37 5.01 5.06 4.82 4.61

4 –36 36 36 –52 –36 44 32

8 –16 –20 –4 12 28 12 24

52 56 56 52 32 4 24 4

6.10 7.43 4.00 4.04 3.97 4.39 4.73 4.87

0 –36 36 32 –56 –36 52 32

4 –16 –16 4 8 24 4 28

52 56 60 60 32 4 40 0

8.67 8.51 8.34 7.11 6.49 7.30 5.28 7.64

p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.005 p < 0.001

Parietal Left postCG** Right post CG** Left SPL** Right SPL**

–36 28 –24 24

–36 –44 –64 –60

56 56 48 60

4.53 6.54 3.76 4.87

–36 24 –20 20

–40 –48 –68 –60

56 56 44 60

5.85 4.52 4.16 4.11

–28 28 –16 12

–32 –48 –72 –68

64 56 48 52

6.03 7.29 5.37 5.00

p < 0.001 p < 0.001 p < 0.001 p < 0.01

–28 –12 –24 –28 24 36 12 20 28

–68 –64 –48 –60 –76 –68 –60 –56 –48

32 8 0 –4 28 16 0 4 –8

4.24 7.42 4.40 3.76 7.81 6.29 7.75 7.45 4.03

–28 –12 –24 –28 24 36 8 24 24

–68 –68 –44 –60 –76 –64 –60 –64 –40

24 0 –4 –4 28 12 8 0 0

7.48 8.13 5.98 7.70 3.86 3.03 6.89 2.05 2.29

–28 –8 –24 –28 24 36 4 20 24

–72 –64 –44 –60 –76 –68 –64 –56 –48

28 4 8 –4 28 20 8 4 –16

7.77 9.31 5.19 8.36* 7.8 6.07 9.04 7.78* 7.86

p < 0.001 p < 0.001 p < 0.005 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001

Occipital Left CUN** Left LG/FG**

Right CUN Right LG/FG

SMA, supplementary motor area; preCG, pre-central gyrus; postCG, post-central gyrus; IFG, inferior frontal gyrus; MFG, middle frontal gyrus; SPL, superior parietal lobule; CUN, cuneus; LG, lingual gyrus; FG, fusiform gyrus. *Not a regional maximum; selected from regional maximum of target-related activations. **Activations shown in Figs. 3 and 4

288



articles

Cues > targets

Targets > cues SMA/cingulate

L SFG R MFG

L MFG

L preCG

R preCG

L postCG L IPS

L SPL

R IPS

L MFG


L SFG

L preCG L IPS

L postCG L SPL

L LOG

L CUN L IFG

L STS/STG

L Post. cingulate

R postCG R SPL

L SFG

SMA/cingulate

L LG/FG

Fig. 4. Significant differences between cue and target processing. Areas significantly more active for cues than targets (left column) and those active more for targets than for cues (right column) projected on a 3Drendered brain. Labels indicate the brain regions referred to in Tables 1 and 2. The Z-values and stereotactic coordinates for the regional maxima are listed in Tables 1 and 2. SFG, superior frontal gyrus; MFG, middle frontal gyrus; SPL, superior parietal lobule; IPS, intra-parietal sulcus; STS, superior temporal sulcus; STG, superior temporal gyrus; LOG, lateral occipital gyrus. SMA, supplementary motor area; preCG, pre-central gyrus; postCG, post-central gyrus; IFG, inferior frontal gyrus; CUN, cuneus; LG, lingual gyrus; FG, fusiform gyrus.

anisms of attentional control and the resulting modulation of sensory signals. The challenge for the future is to relate activity in each of these identified brain regions to the specific neurocomputational operations they subserve during visual spatial attention.

METHODS Stimuli and task parameters: cueing study. Each trial began with a cue that consisted of two overlapping isoluminant arrows presented at fixation (one blue and one yellow, pointing in opposite directions, each 1.9° tall × 0.96° wide; duration, 500 ms). To eliminate differential activations based on the physical differences in right versus left cues, half the subjects were instructed to orient attention based on the direction of the blue arrow, and the other half were required to use the yellow arrow. The relevant arrow pointed randomly to the left or right. Target locations were demarcated by white outline rectangles (8.1° wide × 7.2° tall, centered 5.6° above the horizontal meridian and 7.7° lateral to the vertical meridian) that remained on the screen throughout all runs. Target stimuli appeared within the white outline rectangles and consisted of black and white checkerboards (8.1° × 7.2°, composed of 0.9° square nature neuroscience • volume 3 no 3 • march 2000

checks) that reversed at 4 Hz over a 750-ms duration (so that each stimulus comprised 3 checkerboards lasting 250 ms each). A random one-half of target stimuli contained between 3 and 9 gray checks (randomly located) in place of white checks during one phase of the reversals (the second 250-ms period). Subjects were required to discriminate the target stimuli at the cued location only, pressing a button with one hand if gray checks were present, or pressing a button with the other hand if there were no gray checks (hand of response was counter-balanced across subjects). On 17% of trials, the target stimuli occurred 1000 ms after the offset of the cue. On the remaining 83% of trials, targets occurred 8160 ms following the cue. In order to distinguish the hemodynamic response to the cues from that to the targets, only the longer ISI data were analyzed and reported. The interval from the offset of the targets to the onset of the next trial cue was 7910 ms. Stimuli were delivered through an MRI-compatible goggle system (Resonance Technology, Northridge, California). Subjects were trained before the MRI session to ensure that the task was clearly understood and that subjects were able to perform the task without eye movements. In the scanner, they were instructed to covertly orient attention at the moment the cue was presented and to diligently maintain fixation throughout all runs; this was verified in control studies in two subjects using signal-averaged electrooculography outside the scanner. Speed of response was not emphasized, in order to minimize head-motion artifacts that might result from requiring a speeded response. Head motion was minimized by using a bite bar attached to the head coil. Six healthy adults (ages 22–34; 4 female) participated and gave written informed consent before all sessions. All procedures were approved by the UC Davis human subjects review committee. Each subject performed 5 runs of the task (396 s per run) while MRI data was acquired.

Stimuli and parameters: stimulation of meridia and control sessions. In two of the subjects, visual field mapping and cue control sessions were performed. Meridian stimuli for visual area mapping were 9.0° × 2.7° checkerboard patterns aligned on the vertical or horizontal meridia and composed of alternating 0.9° square black and white checks, reversing at 15 Hz. Stimuli were presented in alternating 22-s blocks of right versus left meridia and in a separate run, upper versus lower meridia. The alternating pattern of vertical and horizontal meridia activations in visual cortex were used to identify the borders between V1, V2, VP and V4 (refs. 44, 47). For the cue sensory-control session, the cues used in the main experiment were flashed every 750 ms (duration, 500 ms) for 22 s alternating with 22-s no-stimulus blocks. A synthetic hemodynamic response function was convolved with the box-car function that represented the periodic alternation in the experimental conditions to yield the model response function that included a temporal derivative. Low-frequency drifts were modeled and filtered with a high-pass filter cutoff of 88 s, and the data were temporally smoothed by convolving the time series with a 4-s full width at half maximum (FWHM) Gaussian kernel. In another cue control, the cue and target stimuli were identical to that used in the main attention study except that the cues were not instructive as to where to orient attention. (Subjects had to 289


articles

Selective attention (targets) (single subject)

a

Att. left > att. right L

R

Scanning procedures. Functional images were acquired with a Signa Advantage 1.5 Tesla whole-body MRI system (General Electric, Waukesha, Wisconsin), with an elliptical end-capped quadrature radio frequency and local gradient head coil (Medical Advances, Milwaukee, Wisconsin). Images were acquired using T2*-weighted gradient-recalled echo, echo planar imaging (EPI) in the coronal plane with a repetition time (TR) of 2.75 seconds, an effective echo time (TE) of 40 ms and a flip angle (FA) of 90°. Twenty-two interleaved slices were collected with a 22-cm field of view (FOV), 64 × 64 matrix, slice thickness of 6 mm and an interslice gap of 2 mm, yielding a voxel size of 3.43 × 3.43 × 6.00 mm3. Fourier image reconstruction included N/2 ghost correction using image phase correction48. Neural activation was detected via the blood oxygenation level dependent (BOLD) contrast mechanism15, which provided differences in signal intensity based upon differences in tissue T2* and perfusion. High-

a

Targets

Att. left > att. right

(n = 6 subs)

4

2

0

Cued left > cued right

Cues (n = 6 subs)

4

2

R

0

Cued right > cued left

Z value

Cuneus Lingual Fusiform

0

290

3 2 1

1

L

R

L

R

Attend left

Image processing. The first 16 images at each slice location (initial 44 s of acquisition) were discarded from the analysis. The remaining 128 images at each slice were used for the time-series analyses. Images were realigned and corrected for movement artifacts. The anatomical T2 images were coregistered with the proton density images and the functional images for each subject. Each subject’s T2 (and coregistered proton density scans) were then

4

2

L

resolution proton density and T2-weighted fast spin echo (FSE) images were also obtained in the coronal plane during the same scanning session. The anatomical scans were acquired with the following parameters: TR = 4.2 s; TE = 17 and 136 ms; echo train = 8; FOV = 22 cm, matrix = 512 × 256 (interpolated to 512 × 512 for display), slice thickness of 6 mm, with an interslice gap of 2 mm. In each EPI acquisition, 144 images were acquired at each of the 22 slice locations, for a total acquisition time of 396 s (6 min, 36 s).

4 3

Selective attention

Attend right

Z value

Z value

b

0

V1 V2 VP V4

Att. right > att. left

L

2

R

1

R

3

Visual areas L

Cuneus Lingual Fusiform L

R

b

4

0

L

Z value

6

2

5

1

5

3

Att. right > att. left 6

4

detect gray checks in both hemifields). During these sessions, target stimuli (250 ms) were presented on only 22% of trials with an ISI of 300 ms, whereas the other 78% of trials (catch trials) consisted only of the cue stimuli; only these catch trials were analyzed, using event-related methods (see below).

Z value


Fig. 5. Selective processing of target stimuli and retinotopy (single subject). (a) A T2-weighted coronal slice (y = –72) showing the activations due to selective visual attention on target processing for one subject. The left column shows the regions showing greater activity when attention was focused on the left visual field. The right column shows those areas showing greater activity when attention was focused on the right. Attention-related activations in the contralateral fusiform and lingual gyri are outlined in black (coordinates of left hemisphere maximum, –32, –60, –8; Z = 5.26; right hemisphere maximum, 20, –68, –8; Z = 5.39). A more dorsal region in the cuneus (not outlined in black) also showed enhanced activity contralateral to the direction of attention (left hemisphere maximum = –32, –64, 28, Z = 4.02; right hemisphere maximum = 24, –72, 28, Z = 6.27). (b) Left panel, ventral visual areas for the same subject shown at top, determined using the meridia scans (see Methods) and traced onto contrast-inverted, T2-weighted MRI scans (y = –72). Right panel, effect of attention to target stimuli on responses in lingual and fusiform gyri (shown as thick black outlines) overlaid onto the subject’s retinotopically mapped visual areas. Attention-related target processing extends from V2 through V4 in the ventral visual-processing stream.

R

0

Fig. 6. Selective attention-related activations to targets and cues (group data). Results of group analysis showing regions with differential activity due to the direction of attention, overlaid onto averaged (6 subject) proton density MRI scans (slice at y = –64). (a) Target processing was enhanced in regions of visual cortex contralateral to the attended hemifield. These significant (all p < 0.001 corrected) attention-related activations were observed in the lingual and fusiform gyri in the right (12, –64, –4; Z = 5.86) and left (–28, –60, –8; Z = 6.15) hemispheres, and also in dorsal regions of the cuneus in the right (24, –72, 28; Z = 6.04) and left (–28, –68, 24; Z = 5.30) hemispheres. (b) Contralateral to the direction of attention, multiple regions of extrastriate cortex showed responses to the cue stimuli that differed based on the direction of attention, including the lingual and fusiform gyri (12, –64, –4, Z = 4.51; –20, –64, –4, Z = 4.25) as well as a more dorsal region of the cuneus (24,–68, 28, Z = 3.87; –28, –68, 24, Z = 4.61). The maxima of these activations were 0–1 cm from those in response to targets and were all significant at p < 0.001, uncorrected. (Because regions of interest analyzed for cue responses were specified a priori by analyses for attention effects on target responses, we give uncorrected p values for cue responses.) nature neuroscience • volume 3 no 3 • march 2000


articles


spatially normalized to a standard stereotactic space49 (using, however, the canonical Montreal Neurological Institute template in the SPM97 package), with the origin at the anterior commisure and the x–y plane extending through the anterior and posterior commisures. The functional images were then spatially normalized into stereotactic space, resampled to 4 × 4 × 4 mm3 voxels and spatially smoothed with an 8-mm isotropic Gaussian kernel. Event-related analyses. The responses of interest in the event-related design were modeled by convolving a vector containing the onset times of the cues and targets with a synthetic hemodynamic response function composed of the sum of two gamma functions and its temporal derivative (accounting for the hemodynamic response and subsequent undershoot and for slight variations in timing)50. As a result, the present analysis emphasizes transient hemodynamic activity temporally adjacent to the cue and target events. Multiple linear regression (using a least-squares approach to estimate the parameters), as implemented in SPM97, was used to model the effects of interest and other confounding factors (such as session effects and low-frequency drifts of signal intensity) at every voxel simultaneously. Data from different sessions were proportionally scaled to a grand mean of 100 arbitrary units to account for overall differences in the intensity of whole brain volumes across the time series. Temporal smoothing was accomplished by convolving the data with a 4-s FWHM Gaussian filter (to account for temporal autocorrelations in the time series). An omnibus voxel-wise analysis provided an SPM(F), and those voxels with significant F-ratios (p < 0.001 uncorrected) were retained for further specific statistical tests. Specific hypotheses were then tested with t-tests at those voxels, and the resulting SPM{t} was transformed to the unit normal distribution resulting in the Z-scores and SPM(Z) maps in the figures and tables.

ACKNOWLEDGEMENTS This research was supported by grants from the National Institute of Mental Health (MH55714 and MH57138) and Human Frontier Science Program to G.R.M., and a National Science Foundation fellowship to J.B.H. We thank Neva Corrigan for assistance in collecting the MRI data, Jeff Maxwell for assistance in collecting the EOG data, Karl Friston and Christian Buechel for advice regarding the event-related fMRI analyses and Kevin LaBar, Kevin Wilson, Barry Geisbrecht, Marty Woldorff, Daniel Weissman, Steven Hillyard and Tamara Swaab for comments on the manuscript.

RECEIVED 27 SEPTEMBER 1999; ACCEPTED 12 JANUARY 2000 1. Pashler, H. The Psychology of Attention (MIT Press, Cambridge, Massachusetts, 1999). 2. Posner, M. I. & Petersen, S. E. The attention system of the human brain. Annu. Rev. Neurosci. 13, 25–42 (1990). 3. Moran, J. & Desimone, R. Selective attention gates visual processing in the extrastriate cortex. Science 229, 782–784 (1985). 4. Mangun, G. R. & Hillyard, S. A. Modulation of sensory-evoked brain potentials provide evidence for changes in perceptual processing during visual-spatial priming. J. Exp. Psychol. Hum. Percept. Perform. 17, 1057–1074 (1991). 5. Mesulam, M.-M. A cortical network for directed attention and unilateral neglect. Ann. Neurol. 10, 309–325 (1981). 6. Posner, M. I., Walker, J. A., Friedrich, F. A. & Rafal, R. D. Effects of parietal injury on covert orienting of attention. J. Neurosci. 4, 1863–1874 (1984). 7. Corbetta, M., Miezin, F., Shulman, G. & Petersen, S. A PET study of visuospatial attention. J. Neurosci. 13, 1202–1226 (1993). 8. Gitelman, D. R. et al. A large-scale distributed network for covert spatial attention. Brain 122, 1093–1106 (1999). 9. Heinze, H. J. et al. Combined spatial and temporal imaging of spatial selective attention in humans. Nature 392, 543–546 (1994). 10. Martinez, A. et al. Involvement of striate and extrastriate visual cortical areas in spatial attention. Nat. Neurosci. 2, 364–369 (1999). 11. Nobre, A. C. et al. Functional localization of the system for visuospatial attention using positron emission tomography. Brain 120, 515–533 (1997). 12. Harter, M. R., Miller, S. L., Price, N. J., LaLonde, M. E. & Keyes, A. L. Neural processes involved in directing attention. J. Cogn. Neurosci. 1, 223–237 (1989). 13. Buckner, R. et al. Detection of cortical activation during averaged single trials of a cognitive task using functional magnetic resonance imaging. Proc. Natl. Acad. Sci. USA 93, 14302–14303 (1996). 14. McCarthy, G., Luby, M., Gore, J. & Goldman-Rakic, P. Infrequent events transiently activate human prefrontal and parietal cortex as measured by functional MRI. J. Neurophysiol. 77, 1630–1634 (1997).


15. Ogawa, S. et al. Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging. Proc. Natl. Acad. Sci. USA 89, 5951–5955 (1992). 16. Mangun, G. R., Hopfinger, J., Kussmaul, C., Fletcher, E. & Heinze, H. J. Covariations in ERP and PET measures of spatial selective attention in human extrastriate cortex. Hum. Brain Mapp. 5, 273–279 (1997). 17. Mangun, G. R., Buonocore, M., Girelli, M. & Jha, A. ERP and fMRI measures of visual spatial selective attention. Hum. Brain Mapp. 6, 383–389 (1998). 18. Tootell, R. B. et al. The retinotopy of visual spatial attention. Neuron21, 1409–1422 (1998). 19. Brefczynski, J. A. & DeYoe, E. A. A physiological correlate of the spotlight of visual attention. Nat. Neurosci. 2, 370–374 (1999). 20. Engel, S. A., Glover, G. H. & Wandell, B. A. Retinotopic organization in human visual cortex and the spatial precision of functional MRI. Cereb. Cortex 7, 181–192 (1997). 21. Sereno, M. I. et al. Borders of multiple visual areas in humans revealed by functional magnetic resonance imaging. Science 268, 889–893 (1995). 22. Woldorff, M. et al. Retinotopic organization of early visual spatial attention effects as revealed by PET and ERPs. Hum. Brain Mapp. 5, 280–286 (1997). 23. Treue, S. & Maunsell, J. H. R. Attentional modulation of visual motion processing in cortical areas MT and MST. Nature 382, 539–541 (1996). 24. Corbetta, M. et al. A common network of functional areas for attention and eye movements. Neuron, 21, 761–773 (1998). 25. Coull, J. T. & Nobre, A. C. Where and when to pay attention: The neural systems for directing attention to spatial locations and to time intervals as revealed by both PET and fMRI. J. Neurosci. 18, 7426–7435 (1998). 26. Bushnell, M. C., Goldberg, M. E. & Robinson, D. L. Behavioral enhancement of visual responses in monkey cerebral cortex. I. Modulation in posterior parietal cortex related to selective visual attention. J. Neurophysiol. 46, 755–772 (1981). 27. LaBar, K. S., Gitelman, D. R., Parrish, T. B. & Mesulam, M. -M. Neuroanatomic overlap of working memory and spatial attention networks: A functional MRI comparison within subjects. Neuroimage 10, 695–704 (1999). 28. Corbetta, M. Frontoparietal cortical networks for directing attention and the eye to visual locations: Identical, independent, or overlapping neural systems? Proc. Natl. Acad. Sci. USA 95, 831–838 (1998). 29. Friedrich, F. J., Egly, R., Rafal, R. & Beck, D. Spatial attention deficits in humans: a comparison of superior parietal and temporal-parietal junction lesions. Neuropsychology 12, 193–207 (1998). 30. Watson, R. T., Valenstein, E., Day, A. & Heilman, K. M. Posterior neocortical systems subserving awareness and neglect. Arch. Neurol. 51, 1014–1021 (1994). 31. Henik, A., Rafal, R. & Rhodes, D. Endogenously generated and visually guided saccades after lesions of the human frontal eye fields. J. Cogn. Neurosci. 6, 400–411 (1984). 32. Rosen, A. C. et al. Neural basis of endogenous and exogenous spatial orienting: a functional MRI study. J. Cogn. Neurosci. 11, 135–152 (1999). 33. Petit, L. et al. PET study of the human foveal fixation system. Hum. Brain. Mapp. 8, 28–43 (1999). 34. Jonides, J. et al. Spatial working memory in humans as revealed by PET. Nature 363, 623–625 (1993). 35. Kojima, S. & Goldman-Rakic, P. S. Delay-related activity of prefrontal neurons in rhesus monkeys performing delayed response. Brain Res. 248, 43–50 (1982). 36. D’Esposito, M., Ballard, D., Aguirre, G. K. & Zarahn, E. Human prefrontal cortex is not specific for working memory: a functional MRI study. Neuroimage 8, 274–282 (1998). 37. Jonides, J., Smith, E. E., Marshuetz, C., Koeppe, R. A. & Reuter-Lorenz, P. A. Inhibition in verbal working memory revealed by brain activation. Proc. Natl. Acad. Sci. USA 95, 8410–8413 (1998). 38. Smith, E. E. & Jonides, J. Neuroimaging analyses of human working memory. Proc. Natl. Acad. Sci. USA 95, 12061–12068 (1998). 39. Corbetta, M. in The Attentive Brain (ed. Parasuraman, R.) 95–122 (MIT Press, Cambridge, Massachusetts, 1998). 40. Posner, M. I. & DiGirolamo, G. J. in The Attentive Brain (ed. Parasuraman, R.) 401–423 (MIT Press, Cambridge, Massachusetts, 1998). 41. Luck, S. J. et al. Effects of spatial cuing on luminance detectability: Psychophysical and electrophysiological evidence for early selection. J. Exp. Psychol. Hum. Percept. Perform. 20, 887–904 (1994). 42. Van Voorhis, S. T. & Hillyard, S. A. Visual evoked potentials and selective attention to points in space. Percept. Psychophys. 22, 54–62 (1977). 43. McAdams, C. J. & Maunsell, J. H. R. Effects of attention on orientation-tuning functions of single neurons in macaque cortical area V4. J. Neurosci. 19, 431–441 (1999). 44. Tootell, R. et al. Functional analysis of human MT and related visual cortical areas using magnetic resonance imaging. J. Neurosci. 15, 3215–3230 (1995). 45. Luck, S. J., Chelazzi, L. Hillyard, S. A. & Desimone, R. Neural mechanisms of spatial selective attention in areas V1, V2 and V4 of macaque visual cortex. J. Neurophysiol. 77, 24–42 (1997). 46. Chawla, D., Rees, G. & Friston, K. J. The physiological basis of attentional modulation in extrastriate visual areas. Nat. Neurosci. 7, 671–676 (1999). 47. Kastner, S., Pinsk, M. A., De Weerd, P., Desimone, R. & Ungerleider, L. G. Increased activity in human cerebral cortex during directed attention in the absence of visual stimulation. Neuron 22, 751–761 (1999). 48. Buonocore, M. H. & Gao, L. Ghost artifact reduction for echo planar imaging using image phase correction. Magn. Reson. Med. 38, 89–100 (1997). 49. Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988). 50. Friston, K. J. et al. Event-related fMRI: Characterizing differential responses. Neuroimage 7, 30–40 (1998).

291


articles

Voluntary orienting is dissociated from target detection in human posterior parietal cortex


Maurizio Corbetta1,2,3, J. Michelle Kincade1,4, John M. Ollinger2, Marc P. McAvoy2 and Gordon L. Shulman1 1

Department of Neurology and Neurological Surgery, Washington University School of Medicine, 4525 Scott Avenue, St. Louis, Missouri 63110, USA

2

Mallinckrodt Institute of Radiology, Washington University School of Medicine, 4525 Scott Avenue, St. Louis, Missouri 63110, USA

3

Department of Anatomy and Neurobiology, Washington University School of Medicine, 4525 Scott Avenue, St. Louis, Missouri 63110, USA

4

Department of Psychology, Washington University School of Medicine, 4525 Scott Avenue, St. Louis, Missouri 63110, USA Correspondence should be addressed to M.C. ([email protected])

Human ability to attend to visual stimuli based on their spatial locations requires the parietal cortex. One hypothesis maintains that parietal cortex controls the voluntary orienting of attention toward a location of interest. Another hypothesis emphasizes its role in reorienting attention toward visual targets appearing at unattended locations. Here, using event-related functional magnetic resonance (ER-fMRI), we show that distinct parietal regions mediated these different attentional processes. Cortical activation occurred primarily in the intraparietal sulcus when a location was attended before visual-target presentation, but in the right temporoparietal junction when the target was detected, particularly at an unattended location.

Acute structural damage to the right temporoparietal cortical junction (TPJ; inferior parietal lobule and superior temporal gyrus) in humans produces a complex clinical syndrome characterized by the inability to attend and respond to objects positioned in the left visual field (unilateral visual neglect)1–3. Some symptoms of neglect may reflect a deficit in reorienting attention toward new stimuli in the visual field opposite to the lesion (contralesional field)4–7. Patients with parietal lesions can detect visual stimuli in the contralesional field when correctly cued to their locations, but they are slow or fail to detect the same stimuli when attending to other locations4. This deficit is more severe after right than left parietal lesions5, and is localized to the TPJ7. These findings suggest that the posterior parietal cortex near TPJ may be critical for reorienting the focus of attention toward visual stimuli appearing at unattended locations (reorienting hypothesis). Other data suggest that posterior parietal cortex near/along the intraparietal sulcus (IPs), which separates the superior from the inferior parietal lobule, is involved in voluntarily directing attention to a spatial location (voluntary orienting hypothesis). Neurons in the IPs increase firing rate when a monkey attends to a location while preparing a response8–11. Human functional brain imaging shows activations in the IPs (and superior parietal lobule) when observers voluntarily pay attention to and detect peripheral visual stimuli, with or without concurrent eye movements12–16. It is unknown to what extent areas in human and monkey IPs are homologous. These complementary functional anatomical theories (reorienting to targets in the TPJ and voluntary orienting in the IPs) make specific predictions about which regions should be activated while attending to a spatial location and, subsequently, when detecting a visual target there. If IPs is preferentially 292

involved in voluntary orienting, then it should be activated when an observer attends to a location before presentation/detection of a visual target. If TPJ is necessary for reorienting to a visual target, then its activation should follow the presentation/detection of the target, particularly when it is presented at an unattended location. We tested these predictions using ER-fMRI and an ANOVA-based procedure17. This method has two important characteristics: it can separate the responses to events presented within the same cognitive trial, and it is sensitive to differences in both magnitude and timing of responses. Unlike other published ER-fMRI methods18–21, this method makes no assumptions about the shape of the underlying response function.

RESULTS Normal observers were given a cue indicating the most likely location of a subsequent target stimulus they were required to detect, according to a protocol modified from a published procedure4. The stimulus display consisted of a central fixation cross flanked on either side by square boxes. The length of each arm of the central fixation cross subtended 16 minutes of visual angle. The boxes (size, 1°) were placed at 3.3° of visual angle to either side of the fixation spot. Accurate fixation of the central cross-hair was emphasized throughout the experiment. At the beginning of a trial, a cue arrow pointing to the left or right box was superimposed on the fixation cross. The arrow indicated the most likely location of a subsequent target stimulus, and leftward or rightward arrows were equally probable. The cue arrow remained on the screen for one MR frame (2360 ms; cue period). The sequence of events following presentation of the cue arrow depended on the type of trial. On a cue trial (20% of the nature neuroscience • volume 3 no 3 • march 2000



articles

Fig. 1. Display, trial types and MR design. Each trial lasted between 4 and 7 MR frames, and each MR frame was 2.36 s long. MR frames are indicated by elongated rectangles below displays. In a cue trial, a cue arrow was presented for 1 MR frame (cue period) at fixation (black rectangle) followed by an intertrial interval (ITI) period signaled by a change in the color of the fixation point (from green to red). The ITI period lasted for 2, 3 or 4 MR frame duration (white rectangles). A two-frame ITI is shown. In a valid trial, the cue period was identical. During the test period (2 MR frames or 4.72 s; crossed rectangles), after a randomly selected time between 1500–3000 ms, a 100-ms target stimulus (asterisk) was flashed in the box cued by the arrow. Subjects indicated target detection with a key press. The ITI period followed. Invalid trial, same as valid trials except that the target was flashed at the uncued box location. Noise trial, same as valid trials, except that no target was flashed during the target period.

trials), the trial ended immediately after cue presentation. The end of the trial was signaled by a change in the color of the fixation cross from green to red. During a cue trial, a subject shifted attention toward the location indicated by the cue and maintained it there until the end of the trial. On a noise trial (20% of the trials), the cue period was followed by a test period lasting 2 MR frames (4720 ms) in which no target was presented. During noise trials, the subject presumably shifted and maintained attention on the cued location for a longer time than during a cue trial (7080 versus 2360 ms; Fig. 1). On a valid trial (44% of the trials), the cue period was followed by a test period during which a target appeared at the location indicated by the cue. The target was a white asterisk that appeared in one of the square boxes for 100 ms. On an invalid trial (16% of the trials), a target appeared during the test period at the uncued location. The cue arrow correctly predicted the target location on 73% of the trials in which a target was presented. These four trial types (cue, noise, valid, invalid) were randomly intermixed. Subjects were instructed to press a button as quickly as possible upon detection of the target and to withhold responses on cue or noise trials.

We used this protocol during whole-brain measurements of blood oxygenation level dependent (BOLD) responses on a Siemens Vision 1.5 T magnet. The time course of the BOLD response for each trial type and each trial period (cue period, test period/noise, test period/valid, test period/invalid) was estimated in each subject using linear regression. Pixel-wise and regional ANOVAs were used for appropriate statistical contrasts17 (see Methods). Behavior Reaction times for target detection were faster on valid than invalid trials (380 ms versus 426 ms, F1,11 = 21.92, p = 0.0007), indicating that subjects used the cue arrow to attend to the location of the target. No responses were recorded during noise and cue trials. Imaging During the cue period, a series of ventral and dorsal visual regions were active; these included bilateral anterior fusiform (Fus; x, y, z atlas coordinates, 35, –57, –20, right; –31, –55, –16, left), lateral occipital (LO;–31, –83, 0, left; 27, –87, 0, right)22,23,

Table 1. List of parietal regions during cue and target periods (averaged over valid and invalid targets) and showing significant validity effect. Cue Regions L pos IPS L ant IPs L vIPS “ R ant IPs R pos IPs R vIPS R IPL R STG R PC L PC

Target

x –25 –25 –23 –27 27 21 29

y –67 –57 –67 –75 –59 –65 –71

z 48 46 32 26 52 52 22

Z-score 7.58 7.55 7.42 7.23 6.75 6.62 6.01

51

–55

4

5.41

Valid – invalid

x –25 –25

y –65 –57

z 48 42

Z-score 6.94 7.27

–27 33

–77 –51

20 48

7.16 7.84

27 53

–71 –45

30 20

6.63 8.53

7 –9

–73 –71

32 40

7.71 7.37

x

y

z

Z-score

39

–47

48

4.09

53 57 7 –5

–49 –45 –75 –71

30 12 34 34

5.12 4.44 4.28 4.27

See Figs. 2 and 3 for anatomical labels. Coordinates (x, y, z) correspond to the Talairach atlas.


293


articles

b

Cue period

Percent change BOLD signal

a

d

Ant IPs MT+ LO

MR frame number



c

Pos IPs

Cue and target period

IPS/cue IPS/valid TPJ/cue TPJ/valid

cue

ITI

MR frame number

Fig. 2. BOLD responses during cue and target periods. (a) ANOVA F map transformed to Z map for the cue period (averaged over subjects, cue direction). Left, sagittal slice 25 mm left of midline. Right, coronal slice 45 mm posterior to center of atlas space. Yellow lines indicate corresponding planes of section. Parietal regions with sustained responses to cues are labeled in red: posIPs, posterior intraparietal sulcus; antIPs, anterior intraparietal sulcus; vIPs, ventral intraparietal sulcus. The lateral occipital region LO (labeled in white) showed a transient response to the cue. (b) BOLD time courses in different regions (averaged over subjects, cue direction and hemispheres) during the cue period. Response typically peaked two frames after onset of cue arrow on frame 1. ITI began on frame 4. (c) Target period (averaged over subjects, valid and invalid targets). Parietal regions with prevalent responses to targets are shown in red label: TPJ, temporoparietal junction; Precun, precuneus. Several motor-related responses are also shown: SMcx, sensory-motor cortex; Put, putamen; Cbl, cerebellum. (d) BOLD time courses in IPs and TPJ (averaged over subjects, cue direction, target field and hemispheres) during cue and target (valid) periods. Target was randomly presented around frame 3. IPs/cue, IPs response during cue period; IPs/valid, IPs response during valid-target period; TPJ/cue, TPJ response during cue period; TPJ/valid, TPJ response during valid-target period.

MT+ (–45, –69, –2, left; 45, –69, –4, right) 24,25 and ventral (vIPs), anterior (ant IPs) and posterior intraparietal sulcus (pos IPs; Fig. 2a, left; see Table 1 for coordinates of parietal foci). No significant activation was detected in the TPJ during the cue period (Fig. 2a, right). The IPs activity did not reach the surface of the inferior parietal lobule, but spread into the superior parietal lobule. The vIPs region was located at the junction of dorsal occipital and parietal cortex, just dorsal and anterior to the V3A representation26. The time course of the BOLD response during the cue period was more sustained within the intraparietal sulcus (posIPs, antIPs, vIPs) than in occipital regions (Fus, LO, MT+; ANOVA regions × frame, F56,672 = 5.18, p = 0.0001). We compared a transient time course in two occipital regions (LO, MT+) with more sustained time courses in antIPs and posIPs (Fig. 2b). The difference in response duration cannot be explained by a difference in the peak magnitude. LO and posIPs, for example, showed similar peak magnitudes on frame 3, but different response duration (ANOVA regions × frames 3 and 4 only, F1,12 = 24.3, p = 0.0003). Similarly, antIPs and MT+ had similar magnitudes, but the response was more sustained in antIPs (ANOVA regions × frames 3 and 4 only, F1,12 = 7.22, p = 0.02). Whereas transient time courses in occipital regions probably reflect visual processes related to the presentation of the foveal cue, the more sustained time courses in intraparietal cortex may 294

reflect longer times required for processing cues related to orienting toward and maintaining attention at the cued location. To further test this idea, we compared BOLD responses of these regions during the noise period, in which subjects maintained attention at a peripheral location for 4.72 seconds after the offset of the cue arrow. Across all regions active during the noise period, only antIPs and vIPs showed sustained activity (ANOVA regions × frame, F77,924 = 7.80, p = 0.0001), with responses of the left hemisphere more sustained than those of the right hemisphere (Ant IPs, F 7,84 = 3.76 p = 0.0014; vIPs, F 7,84 = 5.40, p = 0.0001). Hence, after the presentation of the cue arrow, some IPs regions (antIPs, vIPs) showed a sustained BOLD response that was maintained during the noise period, during which subjects attended to the cued location for almost five seconds while waiting for the target stimulus. During the target period, many visual and motor regions were active for both valid and invalid targets (Fig. 2c). All occipital regions that were active during the cue period also responded to target presentation, and these responses were significantly stronger for targets in the contralateral visual field. In parietal cortex, significant responses were recorded in antIPs, posIPs, vIPs, precuneus (Precun) and the TPJ, where the activation was much stronger in the right than the left hemisphere (Fig. 2c; Table 1). BOLD responses in these parietal regions during the cue and valid-target periods were compared by ANOVAs. Data nature neuroscience • volume 3 no 3 • march 2000


L valid

b

L invalid


a

R valid R invalid

MR frame number

c

L valid


articles

L invalid R valid R invalid

MR frame number


Fig. 3. BOLD responses for valid and invalid targets. (a) Coronal section through TPJ cortex (~47 mm posterior). ANOVA (validity × frame) F map transformed to Z map. Voxels that show significant validity effect (different BOLD responses for valid and invalid targets) independent of the visual field of the target. BOLD time courses (averaged over subjects) were estimated in right inferior parietal lobule (R IPL; b) and right superior temporal gyrus (R STG; c) during the target period for valid and invalid targets in left and right visual field.

for antIPs and posIPs were collapsed, as the borders between the two regions could not be readily identified. Responses of IPs (Ant + Pos) and vIPs regions were stronger during the cue period, whereas those of the TPJ and precuneus region were stronger during the target period (F21,252 = 4.29 p = 0.0001; Fig. 2d). These findings indicate that the voluntary orienting and maintenance of attention to a location primarily recruited the cortex within the IPs. In contrast, target detection recruited the TPJ (and precuneus), although a significant target response was also evident in IPs (Fig. 2d). To test whether TPJ was preferentially involved in reorienting attention toward novel unattended stimuli, the time courses of the BOLD responses during the target periods for valid and invalid trials were compared. The strongest validity effect (difference in the BOLD response for valid and invalid trials) across the whole brain was localized in the right TPJ cortex, with separate foci in the inferior parietal lobule (IPL) and superior temporal gyrus (STG; pixel-wise ANOVA Z-score = 5.12; Fig. 3a; Table 1). A regional ANOVA confirmed that the validity effect was significant and independent of the visual field of the target (ANOVA frame × validity, R IPL, F 7,84 = 8.13, p < 0.0001; R STG, F7,84 = 7.53, p < 0.0001). The response in right IPL and STG was more sustained for valid than invalid targets in each visual field (Fig. 3b and c; ANOVA, frames 5 and 6 only × validity, R IPL, F 1,12 = 5.212, p = 0.041; R STG, F1,12 = 6.51, p < 0.025). Significant validity effects were also localized bilaterally in the precuneus and near the intersection of the right IPs and postcentral sulcus. This latter region did not overlap with the IPs regions active during the cue period (vector distance, 17 mm; Table 1). Significant validity effects not discussed here were also observed in other regions outside of parietal cortex.

DISCUSSION We tested two functional-anatomical theories about the role of posterior parietal cortex in visuospatial attention. One theory, supported by studies of neglect patients with parietal lesions, proposes that the TPJ cortex is necessary for reorienting toward visual targets appearing at unattended locations. Another theory, based on single unit and imaging data, proposes that cortex along the IPs is involved in the voluntary orienting of attention toward a location. Our results provide direct confirmation of both views, showing that IPs was active before target presentation when a nature neuroscience • volume 3 no 3 • march 2000

location was voluntarily attended, independent of processes related to target detection (for instance, visual responses and their attentional modulation or motor responses). In contrast, the right TPJ responded to target presentation more strongly when the target occurred at an unattended location. Voluntary orienting of attention Two findings support a role for the IPs in the voluntary orienting and maintenance of attention to a target location. First, the presentation of a cue arrow indicating the most likely location of a subsequent visual target triggered transient responses in occipital cortex, but more sustained responses in IPs. Transient responses in occipital cortex may reflect the encoding of the cue, which was probably completed within a half second27. In contrast, sustained parietal responses may reflect the required shift toward and maintenance of attention at the cued location for the entire cue period (2360 ms). Second, when the delay after the cue offset (noise period) was extended to 4.72 seconds, forcing subjects to maintain attention at the cued location for longer, IPs was the only brain region that showed a sustained response during the delay. Our results extend findings of increased activity in extrastriate, frontal and IPs cortex without sensory stimulation when subjects attend to a specific object at a specific location during an identification task20. Those activations are thought to reflect an attentional signal-biasing activity in visual cortex before stimulus presentation28. Here we show, in a simpler detection protocol, that IPs was uniquely active when attention was oriented toward and maintained at a relevant location, suggesting that IPs is the source of spatial biases observed in extrastriate visual cortex. Cue-related activity in IPs may underlie an attentional signal that ‘marks’ a location of interest9,11 or an intentional signal that prepares a response (eye movement or hand reaching) toward that location 10. However, attending to direction of motion also drives IPs regions before the presentation of any motion target17. Activations are found in IPs during shifts in an object feature (for instance, color or shape)29 or between percepts during binocular rivalry30. These results suggest that the intraparietal cortex may be involved in visual selection beyond the selection of locations. The BOLD response in IPs was time locked to different processes in the course of a trial. Early in the trial (cue and noise 295


articles


period), the response was controlled by voluntary orienting processes, whereas the response was controlled by detection processes later in the trial (target period; Fig. 2d). The coexistence of different processes in human IPs resembled that observed in monkey IPs, where neurons also manifest activity time locked to different components of a task (for instance, visual, attentional, memory, oculomotor or reaching)9–11,31,32. Although spatial overlap between regions of activation during different protocols (visual attention versus eye movements) in previous imaging studies was used to examine colocalization of processes16,33, the enhanced temporal resolution of new ER-fMRI procedures17,20,34,35 allows a richer and more realistic view of the temporal dynamics of activation in the human brain. Target detection The right TPJ (and precuneus) was specifically engaged during target detection. Unlike other parietal regions that showed both cue and target responses (antIPs, vIPs), right TPJ and precuneus showed little if any response to the cue. The pattern of activation in the TPJ cortex fits two main features of unilateral spatial neglect. The much stronger activation on the right than the left TPJ agrees with the clinical2,3,36 and neuropsychological findings5 that neglect is more severe following right TPJ lesions. The stronger TPJ response following the presentation of a target at an invalid location is also consistent with clinical and experimental4,7 observations that neglect is worse for objects presented at unattended locations. Activation of the right TPJ may underlie the process of spatially redirecting the focus of attention toward the location of unattended stimuli. This reorienting was indexed by longer reaction times for invalid targets than for valid targets; thus, the sensitivity of the TPJ activation to target validity implies that the BOLD signal, despite its slow temporal resolution (seconds), can track neuronal processes that yield behavioral differences of a few tens of milliseconds (the difference in reaction time between valid and invalid trials). Another possibility, however, is that activity in right TPJ cortex is related to spatially nonselective neural processes triggered by reorienting to an unattended target. Interestingly, TPJ damage reduces the amplitude of P300 scalp electrical potentials, which are commonly elicited by the detection of infrequent visual, auditory and somatosensory targets during spatial and nonspatial tasks37. The right TPJ is also selectively activated when observers monitor the environment for infrequent targets (for example, vigilance38), and is the region most densely innervated by noradrenergic projections from the locus coeruleus39 that are thought to mediate vigilance and arousal. Damage to the right TPJ can cause vigilance problems in patients with unilateral neglect40. Changes in the level of vigilance have a slower time course than shifts of attention41 and, therefore, might produce stronger and more sustained right TPJ responses. A dissociation of function between intraparietal and temporoparietal cortex may explain why a verbal cue directing attention toward the contralesional field can transiently reduce neglect in some neglect patients, who typically have damage in the right TPJ region. This effect, extensively used by therapists to ameliorate unilateral visual neglect42,43, may reflect the preserved activation of the IPs, which mediates the allocation of attention by cognitive cues. Normal orienting, however, can be maintained only for a short time in neglect patients based on voluntary strategies. Typically, orienting involves both cognitive and sensory-driven mechanisms 27. This study, along with lesion analyses of patients with TPJ damage, indicates that the right TPJ region is 296

critical for visual reorienting, and dissociates this region from voluntary orienting in nearby IPs.

METHODS Subjects. Thirteen subjects (6 female, 7 male; aged 18–38) were recruited from the Washington University community following procedures approved by the local human studies committee. All subjects were strongly right-handed as measured by the Edinburgh handedness inventory44 and had normal or corrected-to-normal visual acuity and no significant abnormal neurological history. Informed consent was obtained from each subject. Before the MR session, subjects participated in one behavioral session during which they were trained to perform the task while maintaining central fixation. Eye movements were monitored with electro-oculography, and all subjects were able to perform the task without breaks of fixation (resolution, 1.5°). Eye movements were not recorded during the fMRI session. Although we cannot rule out the occurrence of small eye movements during the fMRI session, several arguments diminish the likelihood of this possibility. The visual set-up was identical to the one used in the psychophysical session, in which eye movements were monitored and found negligible. The detection task was not demanding in terms of acuity. Subjects reported no problems in maintaining fixation, in agreement with many studies involving detection tasks27. Additionally, there was no activity in the frontal eye field during the noise period, when the tendency to look at the peripheral box was strongest. Reaction times were not collected from one subject because of equipment malfunction. Apparatus. Stimuli were generated by an Apple Power Macintosh computer and projected onto a screen at the head of the bore by a Sharp LCD projector. Subjects viewed the stimuli through a mirror attached to the head coil. Subjects recorded behavioral responses by pressing an MRIcompatible fiber-optic key held in the right hand. Data analysis and fMRI scan acquisition. An asymmetric spin-echo, echoplanar imaging sequence was used to measure blood oxygenationlevel-dependent (BOLD) contrast (TR = 2.36 s, TE = 50 ms, flip angle = 90°). Each scan consisted of 128 frames during which 16 contiguous 8 mm axial slices were acquired (3.75 × 3.75 mm in-plane resolution). Anatomical images were acquired using a sagittal MP-RAGE sequence (TR = 97 ms, TE = 4 ms, flip angle = 12°, inversion time T1 = 300 ms). Functional data were realigned within and across runs to correct for head movement and coregistered with anatomical data. Whole-brain normalization was applied to equalize signal intensity across subjects. In each subject, hemodynamic responses (8 frames long) were estimated at the voxel level using the general linear model. The design matrix was defined using impulse-basis functions such that at each frame, the data were modeled as the sum of the overlapping hemodynamic response produced by each task effect and a linear trend. The use of catch trials (trials in which only the cue stimulus was present) made it possible to estimate unique responses for the cue, noise, target-valid, and target-invalid periods even though the cue response overlaps noise and target responses in each full trial17 (J.M.O. et al., Soc. Neurosci. Abstr. 24, 1178, 1998). A random-effects analysis was performed by entering the individual time points of each hemodynamic response into voxel-level ANOVAs45. These ANOVAs had two main effects, time and task. The main effect of time was used to determine which voxels were activated. The resulting F-maps were corrected for multiple comparisons using a Gaussian random fields approach46. Fstatistics at each voxel were converted to equivalent Z-statistics. These Z-maps were used to delineate regions of interest. Separate ANOVAs were then run at the regional level to determine the task effect of cue direction (left, right), noise direction (left, right), visual field of the target (left, right) and target validity (valid, invalid).

ACKNOWLEDGEMENTS This research was supported by NIH EY00379 and EY12148 (M.C.). We thank Thomas Conturo, Avi Snyder and Erbil Akbudak for technical support.

RECEIVED 27 SEPTEMBER 1999; ACCEPTED 6 JANUARY 2000 nature neuroscience • volume 3 no 3 • march 2000



articles

1. Mesulam, M. M. A cortical network for directed attention and unilateral neglect. Ann. Neurol. 10, 309–315 (1981). 2. Heilman, K. M., Watson, R. T. & Valenstein, E. in Clinical Neuropsychology (eds. Heilman, K. M. & Valenstein, E.) 243–293 (Oxford, New York, 1985). 3. Halligan, P. W. & Marshall, J. C. Toward a principled explanation of unilateral neglect. Special Issue: The cognitive neuropsychology of attention. Cognit. Neuropsychol. 11, 167–206 (1994). 4. Posner, M. I., Walker, J. A., Friedrich, F. J. & Rafal, R. D. Effects of parietal injury on covert orienting of attention. J. Neurosci. 4, 1863–1874 (1984). 5. Morrow, L. A. & Ratcliff, G. The disengagement of covert attention and the neglect syndrome. Psychobiology 16, 261–269 (1988). 6. Di Pellegrino, G. Clock-drawing in a case of left visuospatial neglect: a deficit of disengagement? Neuropsychologia 33, 353–358 (1995). 7. Friedrich, F. J., Egly, R., Rafal, R. D. & Beck, D. Spatial attention deficits in humans: a comparison of superior parietal and temporo-parietal junction lesions. Neuropsychology 12, 193–207 (1998). 8. Bushnell, M. C., Goldberg, M. E. & Robinson, D. L. Behavioral enhancement of visual responses in monkey cerebral cortex. I. Modulation in posterior parietal cortex related to selective attention. J. Neurophysiol. 46, 755–772 (1981). 9. Colby, C. L., Duhamel, J. R. & Goldberg, M. E. Visual, presaccadic, and cognitive activation of single neurons in monkey lateral intraparietal area. J. Neurophysiol. 76, 2841–2852 (1996). 10. Snyder, L. H., Batista, A. P. & Andersen, R. A. Coding of intention in the posterior parietal cortex. Nature 386, 167–170 (1997). 11. Gottlieb, J. P., Kusunoki, M. & Goldberg, M. E. The representation of visual salience in monkey parietal cortex. Nature 391, 481–484 (1998). 12. Corbetta, M., Miezin, F. M., Shulman, G. L. & Petersen, S. E. A PET study of visuospatial attention. J. Neurosci. 13, 1202–1226 (1993). 13. Nobre, A. C. et al. Functional localization of the system for visuospatial attention using positron emission tomography. Brain 120, 515–533 (1997). 14. Vandenberghe, R. et al. The influence of stimulus location on the brain activation pattern in detection and orientation discrimination-a PET study of visual attention. Brain 119, 1263–1276 (1996). 15. Gitelman, D. R. et al. Functional imaging of human right hemispheric activation for exploratory movements. Ann. Neurol. 39, 174–179 (1996). 16. Corbetta, M. et al. A common network of functional areas for attention and eye movements. Neuron 21, 761–773 (1998). 17. Shulman, G. L. et al. Areas involved in encoding and applying directional expectations to moving objects. J. Neurosci. 19, 9480–9496 (1999). 18. Zarahn, E., Aguirre, G. & D’Esposito, M. A trial-based experimental design for fMRI. Neuroimage 6, 122–138 (1997). 19. Friston, K. J., Josephs, O., Rees, G. & Turner, R. Nonlinear event-related responses in fMRI. Magn. Reson. Med. 39, 41–52 (1998). 20. Kastner, S., Pinsk, M. A., De Weerd, P., Desimone, R. & Ungerleider, L. G. Increased activity in human visual cortex during directed attention in the absence of visual stimulation. Neuron 22, 751–761 (1999). 21. Chawla, D., Reese, G. & Friston, K. J. The physiological basis of attentional modulations in visual areas. Nat. Neurosci. 2, 671–676 (1999). 22. Dupont, P. et al. The kinetic occipital region in human visual cortex. Cereb. Cortex 7, 283–292 (1997). 23. Mendola, J. D., Dale, A. M., Fischl, B., Liu, A. K. & Tootell, R. B. The representation of illusory and real contours in human cortical visual areas revealed by functional magnetic resonance imaging. J. Neurosci. 19, 8560–8572 (1999).


24. Watson, J. D. et al. Area V5 of the human brain: evidence from a combined study using positron emission tomography and magnetic resonance imaging. Cereb. Cortex 3, 79–94 (1993). 25. Tootell, R. B. H. et al. Functional analysis of human MT and related visual cortical areas using magnetic resonance imaging. J. Neurosci. 15, 3215–3230 (1995). 26. Tootell, R. B. H. et al. The retinotopy of visual spatial attention. Neuron 21, 1409–1422 (1998). 27. Posner, M. I. Orienting of attention. Q. J. Exp. Psychol. 32, 3–25 (1980). 28. Desimone, R. & Duncan, J. Neural mechanisms of selective visual attention. Annu. Rev. Neurosci. 18, 193–222 (1995). 29. Le, T. H., Pardo, J. V. & Hu, X. 4T-fMRI study of nonspatial shifting of selective attention: cerebellar and parietal contributions. J. Neurophysiol. 79, 1535–1548 (1998). 30. Lumer, E. D., Friston, K. J. & Rees, G. Neural correlates of perceptual rivalry in the human brain. Science 280, 1930–1934 (1998). 31. Colby, C. L., Gattass, R., Olson, C. R. & Gross, C. G. Topographic organization of cortical afferents to extrastriate visual area PO in the macaque: a dual tracer study. J. Comp. Neurol. 238, 1257–1299 (1988). 32. Gnadt, J. W. & Andersen, R. A. Memory related motor planning activity in posterior parietal cortex of macaque. Exp. Brain Res. 70, 216–220 (1988). 33. Culham, J. C. et al. Cortical fMRI activation produced by attentive tracking of moving targets. J. Neurophysiol. 80, 2657–2670 (1998). 34. Courtney, S. M., Ungerleider, L. G., Keil, K. & Haxby, J. V. Transient and sustained activity in a distributed neural system for human working memory. Nature 386, 608–611 (1997). 35. Cohen, J. D. et al. Temporal dynamics of brain activation during a working memory task. Nature 386, 604–607 (1997). 36. Mesulam, M.-M. Large-scale neurocognitive networks and distributed processing for attention, language, and memory. Ann. Neurol. 28, 597–613 (1990). 37. Knight, R. T. & Scabini, D. Anatomic bases of event-related potentials and their relationship to novelty detection in humans. J. Clin. Neurophysiol. 15, 3–13 (1998). 38. Pardo, J. V., Fox, P. T. & Raichle, M. E. Localization of a human system for sustained attention by positron emission tomography. Nature 349, 61–64 (1991). 39. Foote, S. L. & Morrison, J. H. Extrathalamic modulation of cortical function. Annu. Rev. of Neurosci. 10, 67–96 (1987). 40. Robertson, I. H., Mattingley, J. B., Rorden, C. & Driver, J. Phasic alerting of neglect patients overcomes their spatial deficit in visual awareness. Nature 395, 169–172 (1998). 41. Parasuraman, R., Warm, J. S. & See, J. E. in The Attentive Brain (ed. Parasuraman, R.) 221–256 (MIT Press, Cambridge, Massachusetts, 1998). 42. Weinberg, J. et al. Visual scanning training effects on reading-related tasks in acquired brain-damage. Arch. Phys. Med. Rehabil. 58, 479–486 (1977). 43. Antonucci, G. et al. Effectiveness of neglect rehabilitation in a randomized group study. J. Clin. Exp. Neuropsychol. 17, 383–389 (1995). 44. Raczkowski, D., Kalat, J. W. & Nebes, R. Reliability and validity of some handedness questionnaire items. Neuropsychologia 12, 43–47 (1974). 45. Braver, T. S. et al. A parametric study of prefrontal cortex involvement in human working memory. Neuroimage 5, 49–62 (1997). 46. Worsley, K. J. et al. A unified statistical approach for detemining significant signals in images of cerebral activation. Hum. Brain Mapp. 4, 58–73 (1995).

297

Nature Neuroscience March 2000.pdf

Recommend Documents