Nature Neuroscience December 2001

© 2001 Nature Publishing Group http://neurosci.nature.com

contents


volume 4 no 12

december 2001

http://neurosci.nature.com

editorial Brain structure and function are determined by the interplay of genes and environment. Thompson and colleagues now report that the amount of gray matter in several brain regions, including language areas and frontal cortex, is more similar between identical twins than between fraternal twins. These heritable differences in brain structure were correlated with measures of cognitive performance, suggesting a possible link to general cognitive ability. See pages 1153 and 1253.

Great expectations.........................................................................................1151

news and views Genes, brain and cognition.............................................................................1153 Robert Plomin and Stephen M. Kosslyn SEE ARTICLE, PAGE 1253

Spreading synapsins.......................................................................................1155 Venkatesh Murthy SEE ARTICLE, PAGE 1187

Synaptic connectivity and computation..........................................................1157 Anthony M. Zador SEE ARTICLE, PAGE 1230

Synchronicity: when you’re gone I’m lost without a trace?..............................1159 Anthony D. Wagner SEE ARTICLES, PAGE 1259

book review Yes, but am I free?..........................................................................................1161 Neurophilosophy of Free Will by Henrik Walter, translated by Cynthia Klohr REVIEWED BY ADINA L. ROSKIES

brief communications Regulation of hindbrain organizer FGF8. Page 1175.

Induction of photoreceptor-specific phenotypes in adult mammalian iris tissue.............................................................................1163 M Haruta, M Kosaka, Y Kanegae, I Saito, T Inoue, R Kageyama, A Nishida, Y Honda and M Takahashi Melanopsin in cells of origin of the retinohypothalamic tract...........................1165 J J Gooley, J Lu, T C Chou, T E Scammell and C B Saper Does bouton morphology optimize axon length?............................................1166 J C Anderson and K A C Martin Passive eye displacement alters auditory spatial receptive fields of cat superior colliculus neurons....................................................................1167 J C Zella, J F Brugge and J W H Schnupp

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, 4 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 or (615) 850 5315, 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. $650, Canada add 7% for GST (institutional/corporate), U.S. $199, 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 pending at New York, NY. Executive Officers of Nature America Inc: Annette Thomas, President; Edward Valis, Secretary-Treasurer. Printed by Publishers Press, Lebanon Junction, KY, USA. Copyright ©2001 Nature America Inc.

nature neuroscience • volume 4 no 12 • december 2001

i


contents Memory retrieval impairment induced by hippocampal CA3 lesions is blocked by adrenocortical suppression............................................................1169 B Roozendaal, R G Phillips, A E Power, S M Brooke, R M Sapolsky and J L McGaugh


Visual stimuli activate auditory cortex in the deaf............................................1171 E M Finney, I Fine and K R Dobkins

articles The wiring problem in cortical axons. Page 1166.

Distinct regulators control the expression of the mid-hindbrain organizer signal FGF8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1175 W Ye, M Bouchard, D Stone, X Liu, F Vella, J Lee, H Nakamura, S Ang, M Busslinger and A Rosenthal A receptor is required for response to the sugar trehalose in taste neurons of Drosophila.....................................................................................1182 A Dahanukar, K Foster, W M van der Goes van Naters and J R Carlson Synapsin dispersion and reclustering during synaptic activity...........................1187 P Chi, P Greengard and T A Ryan SEE NEWS AND VIEWS, PAGE 1155

TNFα contributes to the death of NGF-dependent neurons during development.......................................................................................1194 V Barker, G Middleton, F Davey and A M Davies

Auditory responses and eye position. Page 1167.

Wallerian degeneration of injured axons and synapses is delayed by a Ube4b/Nmnat chimeric gene..........................................................................1199 T G A Mack, M Reiner, B Beirowski, W Mi, M Emanuelli, D Wagner, D Thomson, T Gillingwater, F Court, L Conforti, F S Fernando, A Tarlton, C Andressen, K Addicks, G Magni, R R Ribchester, V H Perry and M P Coleman GABAB receptor activation enhances mGluR-mediated responses at cerebellar excitatory synapses.........................................................................1207 M Hirono, T Yoshioka and S Konishi Long-term depression in the nucleus accumbens: a neural correlate of behavioral sensitization to cocaine..................................................................1217 M J Thomas, C Beurrier, A Bonci and R C Malenka Endogenous nicotinic cholinergic activity regulates dopamine release in the striatum................................................................................................1224 F Zhou, Y Liang and J A Dani Differential synaptic processing separates stationary from transient inputs to the auditory cortex...........................................................................1230 M Atzori, S Lei, D I P Evans, P O Kanold, E Phillips-Tansey, O McIntyre and C J McBain

A putative receptor for the sugar trehalose. Page 1182.

SEE NEWS AND VIEWS, PAGE 1157

Inducible, pharmacogenetic approaches to the study of learning and memory.........................................................................................................1238 M Ohno, P W Frankland, A P Chen, R M Costa and A J Silva Inferotemporal neurons represent low-dimensional configurations of parameterized shapes.....................................................................................1244 H Op de Beeck, J Wagemans and R Vogels Genetic influences on brain structure..............................................................1253 P M Thompson, T D Cannon, K L Narr, T van Erp, V Poutanen, M Huttunen, J Lönnqvist, C Standertskjöld-Nordenstam, J Kaprio, M Khaledy, R Dail, C I Zoumalan and A W Toga SEE NEWS AND VIEWS, PAGE 1153

Human memory formation is accompanied by rhinal–hippocampal coupling and decoupling................................................................................1259 J Fell, P Klaver, K Lehnertz, T Grunwald, C Schaller, C E Elger and G Fernández SEE NEWS AND VIEWS, PAGE 1159

Protecting axons from Wallerian degeneration. Page 1199.

errata.......................................................................................................1265 classifieds...................................................................................see back pages


ii



editorial

Great expectations The invention of the vacuum cleaner, the washing machine and the dishwasher were supposed to free housewives from drudgery. Instead, commonly accepted standards for housekeeping became higher with each new ‘labor-saving’ device, while the time spent on housework remained the same or even increased. New technologies create new expectations, and electronic publishing is unlikely to be an exception. In the last decade, many journals have taken advantage of the opportunities presented by the internet. Two-thirds of all journals are now available both online and in print. Over 1000 peerreviewed journals are published solely on the web, approximately 20% of them in the life sciences1,2. Despite this rapid evolutionary change, online publication has recently become a lightning rod for a variety of more revolutionary ideas about scientific publishing3. However, many of the concerns discussed in this context, such as library journal pricing, restrictions on access to the literature and streamlining the peer review system, are only loosely related to one another. Separating these arguments would improve the clarity of the debate. One component of this discussion is the belief that the web should greatly reduce the costs of publishing a research article. However, it cannot be taken for granted that switching from paper to electronic publishing alone would reduce publishers’ expenses. For most journals, the main cost of publication under the present system is not paper or distribution, but skilled labor. The amount of editorial input varies considerably among journals, but in general, high-profile journals spend more on the peer review process, including the costs of considering papers that are ultimately rejected. Such filtering adds value to the highly selected papers that appear in prestigious journals, but having papers reviewed at multiple journals also increases the overall cost of scientific publication. Accepted papers then must be copy edited and formatted for publication, and this process is largely independent of whether the content eventually appears on paper or as a pdf file on a computer screen. Printing and distribution are a surprisingly low percentage of the cost of publication1. Electronic submission can save money on administrative labor and the costs of shipping manuscripts to referees, but these expenses are typically even lower than the cost of printing. Because most of the expenses of producing a journal article do not scale with the number of copies produced, a journal’s circulation is a strong determinant of the publisher’s cost per copy. However, revenues (from both subscriptions and advertising) do increase as more copies are sold, explaining why many prestigious journals are cheaper than their more specialized counterparts. Web publication also brings with it substantial new expenses for establishing and maintaining electronic archives, along with any additional searching or linking services that the publisher may choose to provide. In the future, sophisticated search options will be greatly facilitated if electronic articles can be tagged with metanature neuroscience • volume 4 no 12 • december2001

data according to broadly accepted standards. Current web markup languages, such as HTML, label words according to their position and appearance on the page, but newer languages such as XML permit tagging for content, so that search engines can identify a string of words as “article title” or “gene” or “keyword”. Taking advantage of these new capabilities should improve the usefulness of scientific journals, but it is not likely to reduce their cost. Reducing publishers’ costs significantly would require saving money on labor. One possible solution to this problem would involve major changes to the current system of peer review. Some electronic journals have experimented with automated peer review, in which the author supplies key words and a database program randomly assigns referees who are described with similar key words. More radical proposals include posting unreviewed papers on a web site and allowing readers to comment online, or selecting papers based on the number of hits they receive from browsing scientists. Such approaches would undoubtedly result in substantial cost savings, but they also run the risk of forfeiting quality control. Perhaps greater efficiency can be realized without loss of quality. Another effect of electronic publishing is that it has become easier to test new business models, because the web lowers barriers to entry into the publishing business. New electronic-only journals such as Journal of Vision and the Biomed Central (BMC) journals are peer reviewed and provide free access to primary research articles. J. Vision charges a processing fee to authors, and the BMC journals plan to do so from January 2002. A more radical attempt at quality control is the ‘Faculty of 1000’ initiative4, a soon-to-be-launched subscription service that will separate the filtering from the publishing function, by rating individual articles from other publishers based on the recommendations of wellknown scientists. It remains to be seen which of these approaches will prove economically viable. The current system of scientific publication has evolved over time in response to a variety of selective pressures, and like a biological system, aspects of it are undoubtedly shaped by history rather than function. Because there is no central authority to impose a solution, even revolutionary ideas will have to compete with other models for acceptance among authors, readers, referees, librarians and publishers. In a time of rapid change, perhaps the best strategy will be to occupy as many ecological niches as possible. Whether the future evolution of scientific publishing will be gradual or catastrophic remains to be seen, but it seems likely that the changes will include more competition among a larger number of publishers, which should be good news for scientists. 1. Tenopir, C. & King, D. W. Nature 413, 672–674 (2001). 2. http://dsej.arl.org/dsej/2000/mogge.html. 3. http://www.nature.com/nature/debates/e-access/index.html 4. http://www.facultyof1000.com

1151


news and views

Genes, brain and cognition © 2001 Nature Publishing Group http://neurosci.nature.com

Robert Plomin and Stephen M. Kosslyn By making maps of the differences in cortical gray matter volume between twins, Thompson et al. describe which brain regions are strongly determined by genetic factors; they further investigate how these brain differences correlate with measures of cognitive performance.

The 1990s were declared the “Decade of the Brain” for good reason, but the present decade might yield even more fundamental discoveries as neuroscience begins to capitalize on developments in genetics. The report by Thompson et al.1 in this issue represents an important step forward because it bridges these two fields. The authors used magnetic resonance imaging (MRI) to create threedimensional maps of gray matter and then computed correlations between these measures and general cognitive ability (‘g’), derived from diverse cognitive tests for 40 individuals. What makes this study special is that the subjects were twins—10 pairs of monozygotic (MZ or identical) twins and 10 pairs of dizygotic (DZ or fraternal) twins—allowing the authors to estimate the genetic contribution to individual differences in gray matter volume in various brain regions. The new study1 focuses on the influence of naturally occurring genetic variation on normal interindividual variation, that is, the standard deviation found for nearly any characteristic assessed sensitively enough. Heredity is not only about passing species-general characteristics from parent to offspring, but also about transmitting variation in such characteristics (Fig. 1). Indeed, inheritance of variation is the mainspring of evolution, and thus a central focus of genetics. In contrast, most neuroscience research focuses on universal characteristics. Although perspectives are not right or wrong, just more or less useful for particular purposes, the species-universals perspective and the individual-differences perspective can arrive at different answers because they ask different questions. Robert Plomin is in the Institute of Psychiatry, Social, Genetic & Developmental Psychiatry Research Center, 111 Denmark Hill, London SE5 8AF, UK. Stephen Kosslyn is in the Department of Psychology, Harvard University, William James Hall, 33 Kirkland Street, Cambridge, Massachusetts 02138, USA. e-mail: [email protected] or [email protected]

This distinction is in essence the difference between means and variance, which have no necessary connection, either descriptively or etiologically. Despite its name, the analysis of variance (the most widely used statistical test in science) is actually an analysis of mean effects, with individual differences included in the ‘error term’. Most speciesuniversal research is experimental in the sense that it manipulates an independent variable—such as genes, lesions, drugs or tasks—and asks whether the manipulation can have an effect. Individual differences research, in contrast, is correlational in the sense that it investigates factors that do have an effect in the world outside the laboratory. Not all genetic research informs us about the basis for naturally occurring differences within a species. For example, although knocking out a gene can have major effects, such experiments do not imply that the gene has anything to do with the variation responsible for hereditary transmission of individual differences within a species. In contrast, quantitative genetic methods such as the twin method used by Thompson et al.1 are rooted in the study of naturally occurring variation. Although 99.9% of the human DNA sequence is identical for all people, the 0.1% that differs—3 million base pairs—is ultimately responsible for the ubiquitous hereditary differences found for nearly all complex dimensions and disorders, including cognitive abilities and disabilities2. As the new study1 demonstrates, valuable information can be gained by examining individual differences instead of averaging across groups and treating the differences as error. Indeed, such studies can provide a crucial bridge between neuroscience and genetics, leading to new insights not only about how genes affect cognition but also about how the brain works3. A full understanding of the relationships among genes, brain and cognition needs to encompass events at both levels of analysis (Fig. 1) and discover the links between them.


One exciting finding from the Thompson et al. study1 is the high heritability for gray matter volume in several cortical regions. The remarkably high correlations (about 0.95) for MZ twins mean that MZ co-twins are virtually identical in their volume of gray matter. The same measures for DZ twins, who like any brother and sister are only 50% similar genetically, are much less correlated. Although previous twin studies reported high heritability for brain region volumes assessed by MRI (reviewed in ref. 4), the present study1 goes beyond mere size to the more specific measure of gray matter volume, thus ruling out differences in white matter volume. Gray matter consists of neural cell bodies, whereas white matter consists of axons. Connections among neurons reflect, at least in part, the results of learning—which might be expected to differ among individuals as a result of experience. In contrast, the new findings1 suggest that density of neurons may not be easily modified by experience. Studies of individual differences have much greater demands for statistical power than studies of mean differences. Statistical power refers to the likelihood of detecting a true difference (more accurately, of rejecting the null hypothesis). A rule of thumb is to consider the power required to detect a true result of a specified effect size 80% of the time (in other words, in four of five studies). Ten pairs of MZ twins, as used by Thompson et al., confers 80% power to detect a correlation only if the correlation is greater than 0.70 (one-tailed test, p < 0.05). If correlations for gray matter density are as high as 0.95 for MZ twins, as suggested by this study (and in studies of brain volume as well5), they can be detected reliably with just 10 twin pairs. However, MZ twins could be similar not simply because they have identical genes, but also because they were raised (and continue to live) in similar environments. To remove the coarsest contributions of common environment, heritability estimates are based on the difference in correlations for MZ and DZ twins. The essence of any esti1153



news and views

ferent brain regions are mate of heritability is to intercorrelated, as is likesubtract the correlation Distributions Levels Cognitive examples Genes ly. For example, an MRI for DZ twins from that study of total volume of for MZ twins and double 13 brain regions found the difference. that the brain regions Trying to detect such Species Language intercorrelated substana difference in correlauniversals learning Non-varying tially and that a general tions more than doubles factor (first unrotated the demands for statistiprincipal component in a cal power. For example, factor analysis) accounted even if heritability is Rare severe Severe retardation for 48% of the variance5. 0.90 based on an MZ Single disorders Early-onset Alzheimer's correlation of 0.90 and a Thus the simple correlaDZ correlation of 0.45, tions between gray matter power is less than 40% to volume in different brain detect the heritability regions and ‘g’ should be Common Mild retardation Multiple mild disorders Learning disabilities with 10 pairs of each type considerably higher than of twin. This means that suggested by Thompson a true heritability of 90% et al. Moreover, simple would not be detected as correlations would probNormal Specific cognitive abilities significant more than ably show that all brain QTLs variation General cognitive ability half the time. For more regions correlate with ‘g’, typical heritabilities of not just the frontal region. Amy Center 0.50 (such as MZ and DZ Further analyses of these Fig. 1. Levels of analysis from species to individual differences. Interposed between correlations of 0.75 and these extremes are rare severe disorders, often caused by a single gene necessary and data could also examine 0.50, respectively), 80 sufficient for the disorder, and common mild disorders, called ‘complex disorders’ whether gray matter denpairs of each type of twin because they are influenced by multiple genes and environmental factors. Many sity correlates positively are needed to achieve researchers now believe that common mild disorders are often merely the quantita- with different cognitive 80% power. Comparing tive extreme of the same factors that create normal variation. In other words, there abilities, not just with a heritabilities—for exam- may be no common disorders, just dimensions of normal variation. Genes in such composite ‘g’ score. That ple, asking whether heri- multiple-gene (polygenic) systems are called quantitative trait loci (QTLs) because is, the correlation between dimensions (quantitative continua) rather than disorders gray matter volume and tability differs for brain they are likely to result in 15 regions—again raises the (qualitative dichotomies) . the ‘g’ composite could be ante substantially. The due to certain abilities Thompson et al. sample (such as verbal abilities) of 40 twin individuals is large for a neucorrelating highly and other abilities (such Moreover, multivariate genetic analysis, roimaging study; studying many hunas spatial abilities) correlating less well. In which investigates the extent of genetic dreds of individuals is daunting and may contrast, the hypothesis of ‘genetic g’— basis for associations between variables, require multi-site collaborative efforts. that the same genetic factors affect diverse shows that most of the genetic action on The second important feature of the cognitive abilities—leads to the predicdiverse cognitive abilities involves ‘g’8. A new study1 is that it shows an association tion that gray matter volume should corkey issue for neuroscience is to underrelate not just with a ‘g’ composite but stand the brain mechanisms that medibetween individual differences in gray with all cognitive abilities. ate this genetic effect. matter volume in the frontal cortex and The old workhorse of the twin design Studies of total brain volume antici‘g’ or general cognitive ability. The con(comparing MZ and DZ twins) can be pated the interesting finding by Thompcept of ‘g’ is controversial; not all used to ask questions that go beyond estison et al. of an association between gray researchers are comfortable with the idea mating heritability. For example, the twin matter volume and ‘g’. In 14 studies of that a single factor may influence all design can trace the developmental course about 700 individuals, correlations types of intelligence6. Although ‘g’ is not of genetic and environmental influences. between brain volume and ‘g’ are about the whole story of cognitive abilities— One of the most fascinating findings 0.40 (ref. 4), indicating that individuals group factors representing specific abiliabout ‘g’ is that its heritability increases with larger brain volumes have higher ‘g’ ties are also important level of almost linearly from infancy (about 20%) scores. These correlations are similar in analysis—trying to tell the story without to childhood (about 40%) to adulthood magnitude to the correlations found for ‘g’ loses the plot entirely. The Thompson and old age (about 60%)9. Does the herfrontal gray matter volume in the new et al. results suggest that ‘g’ is not simply study1. However, Thompson et al. undera statistical abstraction that emerges from itability of gray matter follow a similar factor analyses of psychometric tests; it developmental course? estimate the extent to which gray matter also has a biological substrate in the In addition, a multivariate genetic volume in each brain region correlates brain. Dozens of studies, including more analysis suggests that the association with ‘g’. They report partial regressions or than 8,000 parent–offspring pairs, 25,000 between total brain volume and intellicorrelations that indicate the association pairs of siblings, 10,000 twin pairs and gence is substantially mediated geneticalbetween each brain region and ‘g’ indehundreds of adoptive families, all conly 5 . Although the extent to which pendent of other brain regions. Such verge on the conclusion that genetic facanalysis will miss associations with ‘g’ to correlations between brain and cognition tors contribute substantially to ‘g’ 7 . the extent that gray matter volumes in difare genetic must be assessed rather than 1154




news and views

assumed, given the high heritability of gray matter volume in the new paper1, it seems likely that its association with ‘g’ is also mediated genetically rather than environmentally. Multivariate analysis can also help with the next step: discovering what underlies this association and what other aspects of brain anatomy and physiology give rise to individual differences in ‘g’. For example, could differences in the number of specific types of receptors or the density of neuromodulatory pathways be responsible for the observed correlations with intelligence? Magnetic resonance spectroscopy provides measures of metabolic byproducts that can serve as markers for some of these variables. Although it is possible that a single fundamental brain characcteristics such as frontal gray matter volume is responsible for g, it seems more likely that many brain processes are involved. However, so far, the pickings are slim other than brain volume measures. For example, although EEG alpha peak frequency10, EEG coherence (which has been taken as a measure of brain interconnectivity11) and peripheral nerve conduction velocity12 are all highly heritable, these measures do not relate to ‘g’13. Thus, ‘g’ does not seem to involve speedier brains, at least as assessed by these physiological measures. Although event-related brain potential

(ERP) measures yield widely varying heritability estimates across cortical sites, measurement conditions and age, some researchers have reported that ERP (especially the P-300 component) is related to ‘g’14. Other researchers have reported correlations between ‘g’ and brain functioning as assessed by positron emission tomography, single photon emission tomography and functional MRI13, but we are not aware of genetic studies using these techniques. Finding high heritability for ‘g’-related brain measures paves the way for molecular genetic studies to harvest the fruits of the Human Genome Project. Armed with such information, we are poised to identify the specific DNA variation responsible for high heritability. However, identifying specific genes associated with complex traits has proven more challenging than expected, largely because many genes are probably involved, each with small effects7. Nevertheless, finding specific genetic variation is a high priority for research because it will provide a very sharp scalpel for dissecting pathways relating genes, brain and cognition. 1. Thompson, P. et al. Nat. Neurosci. 4, 1253–1258 (2001). 2. Plomin, R., DeFries, J. C., McClearn, G. E. & McGuffin, P. Behavioral Genetics 4th edn. (Worth, New York, 2001).

Spreading synapsins Venkatesh N. Murthy Fluorescent synapsins were used to study the dissociation– reassociation cycle of this synaptic vesicle protein in situ, and how this process relates to regulation of exocytosis.

Three decades ago, Greengard and colleagues identified an abundant brain protein that is a substrate for the cAMPdependent protein kinase1. In the ensuing years, this family of proteins, called synapsins, has been investigated intensely. Somewhat surprisingly, their precise role in synaptic transmission is still unclear. Now, Chi and colleagues2 elegantly combine fluorescence microscopy The author is in the Department of Molecular and Cellular Biology, Harvard University, 16 Divinity Avenue, Cambridge, Massachusetts 02138, USA e-mail: [email protected]

with molecular biology to provide new insight into the involvement of synapsins in neurotransmitter release. Synapsins are abundant at nerve terminals and are highly conserved, and their biochemical properties are regulated by activity. For this reason, investigators have anticipated that synapsins are critical in synaptic transmission. Synapsins have been implicated in a variety of functions—synaptic vesicle clustering, mobilization and even exocytosis—based on their dynamic affinity for synaptic vesicles 3–6 . Mice with two of the three synapsin genes knocked out are viable, but have abnormal synaptic transmission4.


3. Kosslyn, S. & Plomin, R. in Psychiatric Neuroimaging Research: Contemporary Strategies (eds. Dougherty, D., Rauch, S. L. & Rosenbaum, J. F.) 491–515 (American Psychiatric Press, Washington, DC, 2001). 4. Vernon, P. A., Wickett, J. C., Banzana, P. G. & Stelmack, R. M. in Handbook of Intelligence (ed. Sternberg, R. J.) 245–264 (Cambridge Univ. Press, 2000). 5. Pennington, B. C. et al. J. Cogn. Neurosci. 12, 223–232 (2000). 6. Gardner, H. Frames Of Mind: The Theory of Multiple Intelligences (Basic, New York, 1983). 7. Plomin, R., DeFries, J. C., Craig, I. W. & McGuffin, P. (eds.) Behavioral Genetics in a Postgenomic World (APA Books, Washington, DC, in press). 8. Plomin, R. Nat. Rev. Neurosci. 2, 136–141 (2001). 9. McClearn, G. E. et al. Substantial genetic influence on cognitive abilities in twins 80+ years old. Science 276, 1560–1563 (1997). 10. Posthuma, D., Neale, M. C., Boomsma, D. I. & de Geus, E. J. C. Behav. Genet. (in press). 11. van Beijsterveldt, C. E., Molenaar, P. C. M., de Geus, E. J. C. & Boomsma, D. I. Behav. Genet. 28, 443–453 (1998). 12. Rijsdijk, F. V. & Boomsma, D. I. Behav. Genet. 27, 87–98 (1997). 13. Deary, I. J. Looking Down on Human Intelligence: From Psychometrics to the Brain (Oxford Univ. Press, 2000). 14. van Beijsterveldt, C. E. & Boomsma, D. I. Hum. Genet. 94, 319–330 (1994). 15. Plomin, R., Owen, M. J. & McGuffin, P. Science 264, 1733–1739 (1994).

Although it remains to be seen whether removing all three synapsin genes has a more profound effect on survival, knockout mice alone may not reveal subtle regulatory roles; mechanistic studies are important in this regard. Previous experiments using biochemical methods have suggested the following sequence of events. At rest, synapsins are associated with synaptic vesicles and, perhaps, with any actin filaments that may be present in presynaptic sites 1 . Synapsins do not have a membrane-spanning sequence; therefore, their observed association with synaptic vesicles must arise from binding to vesicle components. Synapsins also form homo- and heterodimers, which may assist in crosslinking neighboring vesicles. During action potential stimulation, synapsins dissociate from vesicles and disperse into the cytosol7–9. Synapsin dissociation from vesicles, controlled by phosphorylation, frees the vesicles to move toward the active zone to replenish spent vesicles. Upon termination of stimulation, synapsins are dephospho1155


news and views

Fig. 1. The relationship of the synapsin–synaptic vesicle cycle to exocytosis during nerve activity. Vesicles are initially clustered with synapsin bound in the unphosphorylated state. Stimulation with action potentials results in release of vesicles and phosphorylation of synapsins, which dissociates from vesicles and diffuses throughout the axon. After the termination of stimulation, vesicles are endocytosed, and dephosphorylated synapsin reassociates with the vesicles.

Synaptic bouton

Clustered


Axon shaft

Active zone

P

P

Stimulation P P

P

P

P

P P

P P

P

P P

P

P

P

P

P

P

Re-clustering P

P

P

P

=

Synaptic vesicle

= Synapsin

P = Phosphate

rylated, and they reassociate with vesicles. Phosphorylation of synapsins can occur at several sites by a variety of kinases, including calcium calmodulin-dependent kinases (CaMK) I, II and IV, MAP kinases and protein kinase A. Despite a wealth of biochemical evidence, a real-time view of synapsin dynamics has been lacking until now. Chi and colleagues 2 examined synapsin Ia in living hippocampal synapses by tagging it with enhanced green fluorescent protein (EGFP). At rest, EGFP-synapsin localized to synaptic vesicle clusters, but upon stimulation, it lost its clustered appearance and dispersed along the axon (Fig. 1). Synapsin dispersion was faster than dispersion of an integral vesicle membrane protein or of vesicles labeled with the fluorescent dye FM4-64. Therefore, the redistribution of synapsin is due to its dissociation from vesicles and its movement into axonal regions. EGFP-synapsins carrying mutations in three CaMK phosphorylation sites (serine to alanine substitutions) dispersed more slowly upon stimulation. Concomitantly, vesicle exocytosis was also slowed, presumably because of reduced mobility of vesicles. The key finding that phosphorylation of synapsins alters dispersion rates and vesicle exocytosis rates in a correlated way led Chi and colleagues to suggest that 1156

synapsin dissociation rate is causally related to vesicle mobilization. As the rate of unbinding of synapsins from P P vesicles is slowed, vesicle P movement toward the active zone is also slowed, resulting in reduction of the rate of exocytosis. The strong evidence for dispersion preceding vesicle mobilization would be further bolstered if it could be shown that dispersion occurs even in the absence of exocytosis. For example, synapses treated with tetanus toxin (which essentially abolishes evoked vesicle release) should Bob Crimi exhibit synapsin dispersion rates similar to control synapses upon stimulation. A further important finding was that the effects of mutations at CaMK sites 2/3 are more severe when expressed in synapsin I/II-null neurons than in wildtype neurons. In contrast, CaMK site 1 mutation has similar effects on vesicle mobilization rates whether expressed in null background or in wild type. The authors conclude that phosphorylation at site 1 influences direct binding of synapsins to vesicles, whereas phosphorylation at sites 2 and 3 regulates interaction of synapsins among themselves. What happens to the dispersed synapsins once stimulation is terminated? EGFP-synapsins return to the synaptic vesicle cluster at a rate slower than the observed dispersion rate (t 1/2 of about 100 seconds versus about 15 seconds). To recluster, synapsins need to regain their strong affinity for synaptic vesicles, which is thought to be due to the removal of the phosphate on specific serine residues. However, this idea is not supported by Chi and colleagues’ data 2 , which indicate not only that reclustering occurs in all their mutants, but also that the reclustering is quantitatively similar. This means either that other phosphorylation sites are involved in governing the kinetics of reassociation of synapsins with vesicles, or that the phosphorylation state of synapsins

does not regulate the reassociation. Interestingly, synapsin reclustering and endocytosed vesicle reclustering follow a similar time course 10. This suggests that synapsins rebind to vesicles as they transit from their sites of endocytosis back to the vesicle cluster. Therefore, the rate of synapsin reclustering might be controlled in part by the rate of vesicle reclustering. A simple experiment could address this hypothesis: if synapsin dispersion is induced in the absence of exocytosis (and therefore minimal subsequent endocytosis), the reclustering of synapsin is independent of vesicle traffic. This experiment would determine whether reclustering could be accounted for by diffusional return of synapsin to the synaptic vesicle cluster, or whether it is governed by the traffic of vesicles back to the vesicle cluster. The value of Chi and colleagues’ study arises in part from its unexpected results and the resulting predictions. First, all mutant EGFP-synapsins can dissociate from vesicles. A previous biochemical investigation of synapsin phosphorylation and vesicle affinity found that dissociation of synapsin from vesicles strongly depended on phosphorylation at the CaMK site 1 (ref. 9). In the present study, synapsins that lack all CaMK phosphorylation sites could dissociate from vesicles upon stimulation, although the rate of dissociation was slower. Thus, there must be another switch controlling the affinity of synapsins for vesicles, and Chi and colleagues remind us of other phosphorylation sites and the ATP binding site in the central domain. Examination of the dynamics of EGFP-synapsin with mutations in these sites will be informative. A second surprising finding is that the reclustering dynamics are not altered by the mutations studied. Perhaps the rebinding of synapsins to vesicles following cessation of activity is not rate limiting for reclustering, and therefore is not visible to the assay used. Also, there may be a stronger effect of mutations on dispersion and reclustering in inhibitory synapses. It is likely that Chi and colleagues studied excitatory synapses, which are numerically dominant in




news and views

the type of cultures they used. An earlier electrophysiological study found that the depletion of releasable vesicles occurred more readily in inhibitory synapses from synapsin I null mice11. Real-time visualization of synapsins and other proteins, as developed by Chi and colleagues2, adds a powerful analytical method to the existing array of techniques used to study presynaptic function. It permits analysis of the dynamics of presynaptic molecules in situ while monitoring synaptic function. Imaging biochemistry inside living

synapses in real time will no doubt facilitate analysis of mechanisms in vesicle trafficking. 1. Greengard, P., Valtorta, F., Czernik, A. J. & Benfenati, F. Science 259, 780–785 (1993). 2. Chi, P., Greengard, P. & Ryan, T. A. Nat. Neurosci. 4, 1187–1193 (2001). 3. Li, L. et al. Proc. Natl. Acad. Sci. USA 92, 9235–9239 (1995). 4. Rosahl, T. W. et al. Nature 375, 488–493 (1995). 5. Hilfiker, S. et al. Nat. Neurosci, 1, 29–35 (1998).

Synaptic connectivity and computation Anthony M. Zador

A new study finds two classes of synapses between layer 2/3 neurons in auditory cortex, and suggests they may be involved in processing transient versus sustained acoustic stimuli What endows a cortical circuit with its unique identity? How does a bit of cortex implement the computation that it must perform? The simple answer, of course, is that function arises from the pattern of synaptic connection between neurons and the strengths of these connections. This view motivates much research on synaptic function and plasticity, and is enshrined in formal neural network models of computation. There is, however, remarkably little experimental evidence detailing how a particular matrix of synaptic connectivity gives rise to a particular computation. Atzori and colleagues 1 advance an intriguing hypothesis that begins to bridge this gap between cortical computation and synaptic mechanism. Using dual whole-cell patch-clamp recordings, they examined the properties of synaptic connections between pairs of coupled neurons in layer 2/3 of acute slices of auditory cortex. In the rat, these layer 2/3 neurons (along with neurons in layer 4) receive direct input from the auditory thalamus; layer 2/3 The author is at Cold Spring Harbor Laboratory, 1 Bungtown Rd., Marks Bldg., Cold Spring Harbor, New York, 11724, USA. e-mail: [email protected]

neurons in turn make connections both to other layer 2/3 neurons, and to layer 5 neurons. Atzori et al. found that these layer 2/3 connections fell into two classes, ‘weak’ and ‘strong,’ which differed in a number of important characteristics, including average amplitude, failure rate and paired pulse ratio. Most notably, these connections differed in their temporal dynamics, as assessed by the response to a sustained train of action potentials. Strong connections decayed during the train, while weak connections retained their efficacy at a constant, albeit lower, level throughout the train. Strong connections thus gave their most robust responses to the transient portion of stimuli, while weak connections responded equally well to both transient and sustained stimuli. By contrast, synaptic characteristics between pairs of layer 2/3 neurons in the barrel cortex fell into a third category, which might be called ‘very strong.’ The authors noted an interesting parallel between these two synaptic classes identified in auditory cortex and the two types of firing patterns—‘transient’ and ‘sustained’—with which thalamic inputs to the auditory cortex respond to acoustic stimuli. They sug-


6. Humeau, Y. et al. Neuroscience 21, 4195–4206 (2001). 7. Sihra, T. S., Wang, T. K., Gorelick, F. S. & Greengard, P. Proc. Natl. Acad. Sci. USA 86, 8108–8112 (1989). 8. Torri-Tarelli, F., Bossi, M., Fesce, R., Greengard, P. & Valtorta, F. Neuron 9, 1143–1153 (1992). 9. Hosaka, M., Hammer, R. E. & Sudhof, T. C. Neuron 24, 377–387 (1999). 10. Li, Z. & Murthy, V. N. Neuron 31, 593–605 (2001). 11. Terada, S., Tsujimoto, T., Takei, Y., Takahashi, T. & Hirokawa, N. J. Cell Biol. 145, 1039–1048 (1999).

gest that the two classes of synaptic connection might provide two distinct, parallel channels within the cortex for processing information about these two kinds of auditory stimuli, just as the magnocellular and parvocellular pathways provide separate pathways in the visual system for processing different kinds of visual stimuli. The core of the present findings relates to the physiological properties underlying excitatory coupling between neuronal pairs of layer 2/3 of the cortex. According to the classical quantal model 2 , synaptic transmission is a probabilistic process in which the presynaptic terminal is coupled to its postsynaptic target through a set of n release sites. When an action potential invades the presynaptic terminal, each of the sites releases a vesicle of neurotransmitter with a probability p (and therefore fails to release with probability 1 – p); the postsynaptic response due to the vesicle is given by the quantal size q (which has units of mV or pA). The total postsynaptic response R following an action potential is thus given by the simple equation: R=npq

(1)

Together, the product of the three quantal variables n, p and q governs the average synaptic response R, with weak synapses having a smaller product than strong synapses. Although the quantal model was originally developed to describe events at the neuromuscular junction, the same framework, with relatively minor modifications, has been able to account for transmission at central synapses as well. One important difference is quantitative. At the neuromuscular junction, the number of release sites is typically 1157


quite large (n ∼104), whereas the dynamics in both vertical and number of release sites mediathorizontal excitatory pathing the coupling between a pair ways 12 . These changes might of neurons in the cortex be due to changes in the tem(including both neocortex and poral dynamics of individual hippocampus) is much smaller, synapses, or to a change in often near one. The smaller the relative contribution of value at central synapses has populations with different made it easier to study not characteristic dynamics13. The just the aggregate statistical possibilities for temporal properties of release sites, as processing offered by dynamat the neuromuscular juncic synapses are only beginning tion, but differences among to be explored in the context Statistical n = 4 Statistical them as well. of formal neural network n=1 Strong and weak connecmodels14. Fig. 1. Number of release sites varies among different synapses. tions differ in their release Atzori and colleagues1 have probability. Part of this differproposed a thought-provoking ence may be due to differences and testable hypothesis—that in the number of release sites, n, get (as in autaptic cultures), have difthe two classes of synaptic connection between those two connections (Fig. 1); ferent release probabilities7,8. between layer 2/3 neurons in auditory even small differences in the number of cortex provide a substrate for differenCentral synapses differ in their temrelease sites among synaptic populatial processing of transient versus susporal dynamics as well. Synaptic effications can alter the interpretation of tained acoustic stimuli. It should be cy during a train of action potentials can experimental results. In the hippocamemphasized that, by virtue of the prepaincrease or decrease, depending on the pus, the well-studied Schaffer collateral ration they used (in vitro recording in properties of the synapse and the temconnection between neurons in regions acute slices), their experimental results poral dynamics of the input spike train. CA3 and CA1 is usually mediated by provide no direct support for this idea, The complexity of these synaptic only a single release site3, whereas in the not even correlative. Testing this dynamics arises from the interplay of a hypothesis will require an experimental host of physiologically distinct mechaneocortex a single axon from one neuapproach that can link synaptic and sennisms, including paired pulse facilitaron may make several contacts—as sory physiology. The potential payoff for tion, paired pulse depression and many as a dozen—onto its target 4 . A such challenging experiments is the post-tetanic potentiation, operating on complete failure of synaptic transmisopportunity to understand how netcharacteristic time scales ranging from sion following an action potential is an works of cortical neurons implement milliseconds to seconds or more10. Most exponentially rare event when the their computations. synaptic coupling between two neurons forms of short-term plasticity are mediinvolves multiple release sites: for n ated by changes in release probability. 1. Atzori, M. et al. Nat. Neurosci. 4, 1230–1237 release sites, each with a probability p Indeed, there is an inverse relation (2001). of release and 1 – p of failure, the probbetween the initial synaptic release prob2. del Castillo, J. & Katz, B. J. Physiol. (Lond.) ability that all n sites will fail simultaability and the amount of short-term 124, 560–573 (1954). neously is given by (1 – p)n, a quantity facilitation at single release sites (that is, 3. Sorra, K. E. & Harris, K. M. J. Neurosci. 13, sites with high release probability that diminishes rapidly with increasing 3736–3748 (1993). depress 5 ). This is consistent with the n. Thus the difference in the failure rate 4. Markram, H. Cereb. Cortex 7, 523–533 (1997). between weak, strong and very strong tendency of strong, high probability synapses may arise in part through difconnections in layer 2/3 of auditory cor5. Dobrunz, L. E. & Stevens, C. F. Neuron 18, 995–1008 (1997). ferences in the number of release sites tex to show depression in response to 1 6. Markram, H., Wang, Y. & Tsodyks, M. n, rather than through differences in sustained stimulation . Proc. Natl. Acad. Sci. USA 95, 5323–5328 the release sites themselves. Heterogeneity in the temporal (1998). Strong and weak connections may dynamics of synaptic responses pro7. Murthy, V. N., Sejnowski, T. J. & Stevens, C. F. differ not only in the number of release vides a rich substrate for cortical Neuron 18, 599–612 (1997). sites n, but also in the release probabilcircuits to implement different compu8. Rosenmund, C., Clements, J. D. & ity p at each release site. Recent studies tations. Synaptic dynamics are themWestbrook, G. L. Science 262, 754–757 (1993). of central synapses have revealed selves subject to plasticity. Changes in 9. Hessler, N. A., Shirke, A. M. & Malinow, R. remarkable heterogeneity among release temporal processing in the auditory Nature 366, 569–572 (1993). sites. Heterogeneity of release probabilcortex have been observed following 10. Koch, C. Biophysics of Computation (Oxford ity has been demonstrated by a wide perturbations of sensory experience11, Univ. Press, New York, 1999). range of experimental techniques, but the cellular and synaptic mecha11. Kilgard, M. P. & Merzenich, M. M. Nat. including minimal stimulation5, paired nisms underlying these changes have Neurosci. 1, 727–731 (1998). 6 7 not been examined. In the somatosenneuronal recording , and optical and 12. Finnerty, G. T., Roberts, L. S. & Connors, B. W. sory (whisker) system of the rat, pharmacological 8,9 methods. Even Nature 400, 367–371 (1999). sensory deprivation leads to a reorganisynapses within an ostensibly homoge13. Poncer, J. C. & Malinow, R. Nat. Neurosci. 4, 989–996 (2001). zation of the pattern of cortical responneous population, such as those arising 14. Natschlager, T., Maass, W. & Zador, A. M. siveness, and to corresponding changes from a single presynaptic axon and terNetwork 12, 75–87 (2001). in the average characteristics of synaptic minating on a single postsynaptic tarBob Crimi


news and views

1158



news and views


Synchronicity: when you’re gone I’m lost without a trace? Anthony D. Wagner

Recordings from the human medial temporal lobe suggest that synchronization of oscillations between rhinal cortex and hippocampus may contribute to building declarative memories.

Conscious memory for everyday events depends on learning mechanisms in the medial temporal lobe1,2, where neocortical inputs converge on the hippocampus by way of rhinal cortex. A key to understanding medial temporal contributions to learning is determining how these regions interact during the building of memories. One proposed mechanism for functional integration across different brain regions is gamma-band phase synchronization, in which distinct populations of neurons fire at around 40 Hz and in synchrony3,4. In this issue, Fell and colleagues 5 report that intracranial electroencephalograms from human rhinal cortex and hippocampus tend to demonstrate greater synchrony while subjects learn words later remembered than words later forgotten. This brain–behavior correlation suggests that rhinal-hippocampal interactions may contribute to effective memory formation. Lesion evidence from humans and experimental animals indicates that the medial temporal lobe circuit is necessary for declarative memory. It is well established that damage to these regions decreases the ability to consciously remember events that occur after the neural insult1,2. Moreover, recent functional imaging and electroencephalographic (EEG) studies in humans implicate medial temporal lobe computations in mnemonic encoding6. For example, the degree of rhinal, posterior parahippocamThe author is in the Department of Brain & Cognitive Sciences and Center for Learning & Memory, Massachusetts Institute of Technology, NE20-463, Cambridge, Massachusetts 02139 and the Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, USA. e-mail: [email protected]

pal and hippocampal activation during the encoding of an experience correlates with whether the experience will be later remembered or forgotten7, with these subsequent memory effects emerging in the rhinal cortex before the hippocampus. Although considerable insights into medial temporal lobe function have emerged from these and related studies, evidence regarding the nature of human rhinal–hippocampal interactions during encoding and their relationship to effective learning has been lacking. Fell and colleagues recorded field potentials from the seizure-free rhinal (perirhinal and entorhinal) and hippocampal regions of patients with intractable epilepsy while the patients were attempting to learn individually presented words (Fig. 1). After encoding, the patients were asked to freely recall the words that had been studied, and EEG data acquired during learning were sorted by whether the words were subsequently recalled or forgotten. Rhinal–hippocampal interactions were indexed separately for later remembered and later forgotten trials by assessing the phase synchronization of gamma-frequency oscillations in the EEG signals from these regions. Critically, Fell and colleagues observed that the constancy of the phase lag between rhinal and hippocampal gamma oscillations differed depending on whether the words were later remembered or forgotten. Gamma synchronicity was initially greater for remembered words from 100 to 300 ms and from 500 to 600 ms following word onset. These changes reflected a decrease in the phase differences between rhinal and hippocampal oscillations during these periods (Fig. 1). Later during stimulus processing, decreased synchronicity was observed from 1,000 to 1,100 ms after the onset of subsequently


recalled words. These synchronization differences partially overlapped in time with transient reductions in gamma power in both rhinal and hippocampal regions during effective encoding trials. Due to the correlational nature of these results, we cannot conclude that early rhinal–hippocampal gamma synchronization, later desynchronization, or decreased gamma power in these regions is necessary for declarative memory formation. Nevertheless, these novel findings suggest that more effective encoding may emerge when rhinal and hippocampal neurons synchronously oscillate and then desynchronize, and further suggest that decreased gamma power in these regions during encoding may aid learning. Fell and colleagues propose that increased gamma phase coupling may reflect a change in the functional connectivity between rhinal and hippocampal regions that is important for initiating encoding, for instance by facilitating the transmission of information between these regions. Subsequent desynchronization may mark termination of these regional interactions following information transfer. The authors further suggest that the negative correlation between gamma power and effective encoding may reflect adverse consequences of noise-like ambient gamma activity that could interfere with stimulusspecific activity and thus encoding. Alternatively, they suggest that decreases in gamma could reflect the suppression of components of the rhinal–hippocampal circuit and that failure of such suppression may hinder encoding. Although these interpretations are speculative, they provide important directions for further investigation that undoubtedly will continue to unravel the mysteries of memory formation in the medial temporal lobe. Fell and colleagues results raise a number of fundamental questions regarding rhinal–hippocampal synchronization. First, how do these changes in synchrony emerge? One possibility is that they emerge directly through the dynamics of the medial temporal circuit. However, it is also possible that an attentional signal beyond the medial temporal lobe may serve as a ‘pacemaker’ through inputs that entrain rhinal and hippocampal neurons. Functional MRI studies of encoding consistently show greater prefrontal cortical activation during learning trials that are later better remembered7,8. A number of theorists have posited that encoding depends on interactions among prefrontal, posterior neocortical and medial temporal computations, with prefrontal cortex initiating a cascade of events that can modulate effective trace 1159


news and views

Fig. 1. EEG recordings from human medial temporal lobe revealed greater gamma phase synchronization and desynchronization during the encoding of words later remembered compared to words later forgotten. (a) Approximate location of Fell and colleagues’ recordings from rhinal cortex and hippocampus. The dashed black line represents the angle of electrode insertion along the long axis of the hippocampus. (b) Relationship between rhinal–hippocampal coupling and subsequent memory performance. Encoding of events that were subsequently remembered first evoked increased gamma-phase synchronization between rhinal and hippocampal regions (blue shading) and then decreased synchronization (yellow shading) relative to the encoding of events later forgotten. (Note that, for visualization purposes, the presently rendered oscillations are slower than the observed gamma frequency.)


a

encountered stimulus, such as a person you recently met at a conference, can be based on recollection of speSubsequently remembered cific details about the past encounter with the stimulus or on a general sense of stimulus familiarity. For Subsequently forgotten example, when subsequently encountering the person, you may recall her name or professional affiliation or you may simply have the subjective Time Bob Crimi sensation that the face is familiar and, hence, that you must have met her before. Recently, extensive attention has formation8,9,10. Fell and colleagues propose focused on whether rhinal and hipthat the early onset of increased rhinal–hippocampal subregions differentially subpocampal synchronization may preclude a serve recollection and familiarity. From prefrontal source, perhaps pointing to thalone perspective, the hippocampus is amic modulation of the circuit. However, thought to specifically mediate processes given the hypothesized role of prefrontal that underlie subsequent conscious recolcortex in representing goal states—reprelection of event details13,14. Within this sentations that may be on-line before stimulus presentation—and in biasing posframework, hippocampally derived traces terior processes in favor of task-relevant do not subserve memory based on item codes and pathways11, assessment of prefamiliarity in the absence of recollection. Rather, perirhinal cortex is posited to subfrontal contributions to the emergence of serve the acquisition of item traces that rhinal–hippocampal synchrony would support subsequent familiarity-based appear to be a promising direction for memory14. Fell and colleagues assessed further investigation. Second, is subsequent memory selecsubsequent memory using a free recall tively associated with changes in rhitest. Thus, their results demonstrate that nal–hippocampal synchronization in the synchronous rhinal–hippocampal activigamma range? Fell and colleagues do not ty is correlated with subsequent recollecspecify whether changes in frequency tion. However, these findings need not bands outside of gamma were associatimply that rhinal and hippocampal struced with memory performance. Prior tures subserve the same form of declaraintracranial EEG recordings in humans tive memory. That is, although rhinal have shown theta (4–8 Hz) oscillations inputs to the hippocampus are likely during spatial navigation12. These results important for successful hippocampal formation of traces that ultimately yield recconverge with animal studies that demonollection, within rhinal cortex the strate a relationship between theta rhythm resultant traces may simply support suband hippocampal place codes, with theta sequent item memory. It should prove modulation being associated with proinformative in future investigations to cessing stages that may strengthen memderive separate behavioral measures of ory representations13. To fully appreciate recollection and familiarity, and to examthe role of gamma-band synchronization, ine the relationship between each of these it may be critical to determine whether forms of declarative memory and rhimemory-related rhinal–hippocampal nal–hippocampal synchronization (and coupling is selective to this oscillatory fregamma power). quency or derives from broader coupling. Although questions remain regarding Third, what form of declarative memhow the medial temporal lobe circuit supory emerges from rhinal–hippocampal ports declarative memory formation, the coupling? Memory for a previously

b

1160

Hippocampus

Rhinal cortex

results of Fell and colleagues mark a significant advance in understanding the temporal dynamics of activity within these regions and their relationship to memory formation. Moreover, their study highlights the leverage that can be gained by assessing temporal characteristics of neuronal responses, both within and across distinct structures. This investigation may prove to be the first of many influential efforts to specify how neural coupling across brain regions affects memory behavior. Such future efforts may also emerge from the integration of scalp-recorded magnetoencephalography or EEG with fMRI. The findings by Fell and colleagues5 may well stand as a landmark along the road to specifying the neurocognitive processes that allow us to remember our past. 1. Squire, L. R. Psychol. Rev. 99, 195–231 (1992). 2. Eichenbaum, H. & Cohen, N. J. From Conditioning to Conscious Recollection: Memory Systems of the Brain (Oxford Univ. Press, New York, 2001). 3. Engel, A. K., Fries, P. & Singer, W. Nat. Rev. Neurosci. 2, 704–716 (2001). 4. Varela, F. J., Lachaux, J.-P., Rodriguez, E. & Martinerie, J. Nat. Rev. Neurosci. 2, 229–239 (2001). 5. Fell, J. et al. Nat. Neurosci. 4, 1259–1264 (2001). 6. Schacter, D. L. & Wagner, A. D. Hippocampus 9, 7–24 (1999). 7. Wagner, A. D., Koutstaal, W. & Schacter, D. L. Phil. Trans. R. Soc. Lond. B Biol. 354, 1307–1324 (1999). 8. Wagner, A. D. in Neuropsychology of Memory 3rd edn. (Guilford, New York, New York, in press). 9. Moscovitch, M. J. Cognit. Neurosci. 4, 257–267 (1992). 10. Buckner, R. L., Kelley, W. M. & Petersen, S. E. Nat. Neurosci. 2, 311–314 (1999). 11. Miller, E. K. & Cohen, J. D. Annu. Rev. Neurosci. 24, 167–202 (2001). 12. Kahana, M. J., Sekuler, R., Caplan, J. B., Kirshen, M. & Madsen, J. R. Nature 399, 781–784 (1999). 13. Louie, K. & Wilson, M. A. Neuron 29, 145–156 (2001). 14. Brown, M. W. & Aggleton, J. P. Nat. Rev. Neurosci. 2, 51–61 (2001). 15. Eldridge, L. L., Knowlton, B. J., Furmanski, C. S., Bookheimer, S. Y. & Engel, S. A. Nat. Neurosci. 3, 1149–1152 (2000).



brief communications


Induction of photoreceptor-specific phenotypes in adult mammalian iris tissue

ically expressed in the photoreceptors of the mature retina and is crucial in photoreceptor differentiation. To examine whether iris-derived cells could acquire photoreceptor-specific phenotypes as a result of ectopic expression of Crx, iris-derived cells were infected with a replication-defective recombinant adenovirus6,7. The iris-derived cells infected with Crx-transducing adenovirus expressed rhodopsin (Fig. 2a and d), whereas none of the infected cells with enhanced green fluorescent protein (EGFP)-transducing adenovirus did (Fig. 2c). Similar results were Masatoshi Haruta1, Mitsuko Kosaka2, Yumi Kanegae3, obtained by using the anti-recoverin antibody8 that detects pho3 4 4 toreceptors and subpopulation of bipolar cells (Fig. 2b and d). Izumu Saito , Tomoyuki Inoue , Ryoichiro Kageyama , These results indicate that iris-derived cells have the potential to Akihiro Nishida1, Yoshihito Honda1 and Masayo Takahashi1 differentiate into photoreceptors in response to Crx. 1 Department of Ophthalmology and Visual Sciences, Kyoto University Whereas the adenovirus infection study demonstrated the rhodopsin-inductive effect of Crx in the iris-derived cells, it was Graduate School of Medicine, Sakyo-ku, Kyoto 606-8507, Japan 2 Japan Science and Technology Corporation/TOREST, Okayama Technology not clear whether these rhodopsin-expressing cells were differCenter, 5301 Haga, Okayama 701-1221, Japan entiated from mitotic progenitor-like cells or postmitotic cells 3 Laboratory of Molecular Genetics, Institute of Medical Science, University of that were committed to the neural fate. To examine whether Crx Tokyo, Minato-ku, Tokyo 108-8639, Japan can induce rod generation in dividing cells, Crx was misex4 Institute for Virus Research, Kyoto University, Sakyo-ku, Kyoto 606-8507, Japan pressed with a retrovirus, which was infectious only to mitotic cells. The iris-derived cells infected with the control retrovirus Correspondence should be addressed to M.T. ([email protected]) CLIG9 (Fig. 3a), which regulates expression of only EGFP, were large and flat, and none of them expressed rhodopsin (Fig. 3c–e). Published online: 12 November 2001, DOI: 10.1038/nn762 In contrast, the iris-derived cells infected with the CLIG-Crx (Fig. 3b), which regulates expression of both Crx and EGFP, were We show that iris tissue in the adult rat eye, which is embryonismall and round, characteristic of the rod photoreceptors in cally related to the neural retina, can generate cells expressing monolayer culture (Fig. 3f–h). Most EGFP-positive cells also differentiated neuronal antigens. In addition, the Crx gene transexpressed rhodopsin, although none of the uninfected (EGFPfer induced the specific antigens for rod photoreceptors in the negative) cells did. These results showed that Crx induction is iris-derived cells, which was not seen in the adult hippocampusnecessary for the iris-derived cells to express rhodopsin under derived neural stem cells. Our findings demonstrate a remarkthese culture conditions. Moreover, as retrovirus-infected cells able plasticity of adult iris tissue with potential clinical represent the population of mitotic cells at the time of infection, applications, as autologous iris tissue can be feasibly obtained our findings suggest that it is possible to expand the cell source with peripheral iridectomy. for rod photoreceptors before virus infection. During vertebrate eye development, the inner layer of the As a control, we examined if neural stem cells have the capacoptic cup differentiates into the neural retina and irisity to differentiate into rod photoreceptors. A neural stem cell line pigmented epithelium. This common developmental origin derived from the adult rat hippocampus10 is one of the few proven prompted us to test whether iris tissue could give rise to retinal neurons. When the iris tissue of adult rats was plated and mainto have the potential to self-renew and differentiate into neurons tained in serum-free media containing basic fibroblast growth and glial cells. When grafted to the retina, these neural stem cells factor, many cells migrated out of the iris tissue and proliferatare well integrated into the host retina, where they differentiate ed as a monolayer of cells (Fig. 1a). Some of these iris-derived into neurons and glial cells, but cannot acquire rhodopsin cells became immunoreactive for neurofilament 200 when transimmunoreactivity11–13. To determine whether Crx gene transfer to ferred to an environment that promotes retinal cell differentiathe hippocampus-derived neural stem cells can also induce rod tion1 (Fig. 1b). However, none of these cells expressed rhodopsin, photoreceptors, these cells were infected with recombinant adenovirus. EGFP-positive cells were detected when infected with a specific marker for rod photoreceptors2. EGFP-transducing adenovirus, but none of the cells acquired Both extrinsic cues and intrinsic properties regulate the choice rhodopsin immunoreactivity when infected with Crx-transducof photoreceptor cell fates3. Crx4,5 is the homeobox gene specifing adenovirus (data not shown). These results suggest that adult rat hippocampus-derived neural stem cells a b are intrinsically restricted in their response to Crx under these culture conditions. The common developmental origin of the iris-pigmented epithelium and neural retina may account for the iris-derived cells expressing rhodopsin in response to a Crx cue. Single pigmented ciliary margin cells, but not single pigmented iris cells, can proliferate in vitro to form spherical colonies of cells that can differentiate into rod photoreceptors1,14. To test if the ciliary tissue could also give rise to retinal neurons in adherent monolayFig. 1. Iris-derived cells expressing neural antigens. (a) Iris tissue cultured in the er culture, we transferred the ciliary-derived cells propresence of basic fibroblast growth factor for 3 days. (b) Some of the iris-derived liferated as an adherent monolayer of cells to a cells expressed neurofilament 200 (green) when cultured in a differentiating envidifferentiating environment. None of these ciliaryronment. Nuclei in cells stained with DAPI (4′, 6′-diamidino-2phenylinedole; blue). derived cells showed rhodopsin immunoreactivity. Scale bar, 200 mm (a), 50 mm (b). nature neuroscience • volume 4 no 12 • december 2001

1163



a

a

b


b

c

f

d

g

e

h

d

Fig. 2. Induction of rod photoreceptor-specific antigens with Crx gene transfer. (a, b, d) Iris-derived cells infected with Crx-transducing adenovirus expressed rhodopsin (a, red; 523 of 5011, 10.6 ± 1.2%) and recoverin (b, red; 336 of 2854, 11.8 ± 2.5%). (c) Iris-derived cells infected with EGFP-transducing adenovirus expressed enhanced green fluorescent protein (green; 1896 of 5394, 35.4 ±11.4%); however, none of the infected cells expressed rhodopsin. (a–c) Nuclei in cells stained with DAPI (blue). Scale bar, 25 µm.

However, when these cells were infected with Crx-transducing adenovirus, they became immunoreactive for rhodopsin (data not shown). These results indicate that iris- and ciliary-derived cells behave similarly under the culture conditions in this study. Rod photoreceptors fail to develop in dissociated-cell cultures of the neonatal neural retina, although they develop in large numbers in pellet cultures15. The spherical colonies obtained from ciliary margin cells may provide a differentiating environment similar to that of pellet cultures for neonatal retinal cells. If the iris-derived cells could be cultured to yield spherical colonies, such cells may also give rise to photoreceptors without the need for homeobox gene transfer. Clinically, using iris tissue as the source of retinal regeneration offers the ability to feasibly obtain autologous tissue with peripheral iridectomy, a long established method of ophthalmic surgery. Obtaining autologous ciliary tissue, on the other hand, necessitates such a traumatic surgical intervention as to make this approach unrealistic. The present study shows that iris tissue in the adult mammalian eye retains a remarkable plasticity to give rise to cells expressing neuronal antigens. In addition, iris-derived cells can express photoreceptor-specific antigens with Crx gene transfer, raising the possibility that these cells constitute a potential source of retinal transplantation in patients with retinal degenerative diseases or damaged retinas. (For Methods, see the supplementary information page of Nature Neuroscience on line.) Note: Supplementary methods are available on the Nature Neuroscience web site (http://neurosci.nature.com/web_specials).

ACKNOWLEDGEMENTS We thank C.L. Cepko and T. Furukawa for Crx cDNA and F.H. Gage for adult

1164

c

Fig. 3. Induction of rhodopsin expression from mitotic progenitor-like cells. (a, b) Structures of the recombinant retrovirus CLIG and CLIGCrx. The internal ribosomal entry sequence (IRES) allows for bicistronic expression. LTR, long terminal repeat. (c–e) Infection with the control retrovirus CLIG. The virus-infected cells (EGFP positive, 212 of 2158, 9.8%) were flat and large. None of them expressed rhodopsin. (f–h) Infection with the retrovirus CLIG-Crx. The virus-infected cells (EGFP positive, 187 of 2106, 8.9%) were small and round. Most of the virus-infected cells expressed rhodopsin (179 of 187, 96%). (e, h) Nuclei in cells stained with DAPI (blue). Scale bar, 50 µm. hippocampus-derived neural stem cells, J.F. McGinnis and R.J. Elias for antirecoverin antibody and C.J. Barnstable for the information on the antirhodopsin antibody. This study was supported by a Grant-in-Aid from the Ministry of Education, Science, Sports and Culture of Japan (No. 13671834, 13210077) and a grant from Japan Science and Technology Corporation.

RECEIVED 17 JULY; ACCEPTED 11 OCTOBER 2001 1. Tropepe, V. et al. Science 287, 2032–2036 (2000). 2. Barnstable, C. J. Nature 286, 231–235 (1980). 3. Cepko, C. L., Austin, C. P., Yang, X., Alexiades, M. & Ezzeddine, D. Proc. Natl. Acad. Sci. USA 93, 589–595 (1996). 4. Furukawa, T., Morrow, E. M. & Cepko, C. L. Cell 91, 531–541 (1997). 5. Chen, S. et al. Neuron 19, 1017–1030 (1997). 6. Miyake, S. et al. Proc. Natl. Acad. Sci. USA 93, 1320–1324 (1996). 7. Kanegae, Y., Makimura, M. & Saito, I. Jpn. J. Med. Sci. Biol. 47, 157–166 (1994). 8. McGinnis, J. F. et al. J. Neurosci. Res. 55, 252–260 (1999). 9. Hojo, M. et al. Development 127, 2515–2522 (2000). 10. Palmer, T. D., Takahashi, J. & Gage, F. H. Mol. Cell Neurosci. 8, 389–404 (1997). 11. Takahashi, M., Palmer, T. D., Takahashi, J. & Gage, F. H. Mol. Cell Neurosci. 12, 340–348 (1998). 12. Nishida, A. et al. Invest. Ophthalmol. Vis. Sci. 41, 4268–4274 (2000). 13. Young, M. J., Ray, J., Whiteley, S. J., Klassen, H. & Gage, F. H. Mol. Cell Neurosci. 16, 197–205 (2000). 14. Ahmad, I., Tang, L. & Pham, H. Biochem. Biophys. Res. Commun. 270, 517–521 (2000). 15. Watanabe, T. & Raff, M. C. Neuron 4, 461–467 (1990).





Melanopsin in cells of origin of the retinohypothalamic tract Joshua J. Gooley, Jun Lu, Thomas C. Chou, Thomas E. Scammell and Clifford B. Saper Department of Neurology and Program in Neuroscience, Harvard Medical School, Beth Israel Deaconess Medical Center, Boston, Massachusetts 02215, USA Correspondence should be addressed to C.B.S. ([email protected])

Published online: 19 November 2001, DOI: 10.1038/nn768

All known eukaryotic organisms exhibit physiological and behavioral rhythms termed circadian rhythms that cycle with a near-24hour period; in mammals, light is the most potent stimulus for entraining endogenous rhythms to the daily light cycle. Photic information is transmitted via the retinohypothalamic tract (RHT) to the suprachiasmatic nucleus (SCN) in the hypothalamus, where circadian rhythms are generated, but the retinal photopigment that mediates circadian entrainment has remained elusive. Here we show that most retinal ganglion cells (RGCs) that project to the SCN express the photopigment melanopsin. The phase of circadian rhythms in rodents is shifted most effectively by light ranging from 480–511 nm, consistent with an opsin-based photopigment1–3. However, mice lacking rods and cones have normal circadian entrainment, suggesting that a novel photopigment mediates phase-shifting in response to light4. Recently, melanopsin, an opsin-based photopigment, was localized to the RGC layer of rodents and primates5. We therefore tested whether RGCs that express melanopsin project to the SCN. We injected the right SCN of 10 rats with FluoroGold (FG) to retrogradely label the retinohypothalamic RGCs. Four of the injections were confined to the SCN and did not include the optic chiasm or optic tract (Fig. 1a). In these animals, FG labeled a distinct subset of widely distributed RGCs, corresponding to type III or W cells, as previously reported6. For in situ hybridization, we used a 957-base-pair mouse melanopsin riboprobe5. Melanopsin transcript occurred in a pattern similar to that previously described5, with a scattered population of cells showing intense hybridization, predominantly in the RGC layer (Fig. 1b). In doubly labeled sections, 74.2 ± 0.3% (mean ± s.e.m.) of retrogradely labeled RGCs also expressed melanopsin mRNA (Fig. 1c), with a similar percentage of double labeling in eyes ipsiFig. 1. Colocalization of retrogradely labeled FluoroGold (FG) and melanopsin transcript in retinal ganglion cells of rat. (a) The suprachiasmatic nucleus (asterisk) was injected by glass micropipette with 3 nl of 5% FG, resulting in retrograde labeling of the contralateral suprachiasmatic nucleus (arrow) due to reciprocal innervation. The injection avoided the optic chiasm. (b) In situ hybridization for melanopsin localized with NTB-2 emulsion autoradiography, demonstrating a group of three intensely labeled cells (arrows) in the ganglion cell layer. Light diffuse labeling over all three cellular layers5 was similar to labeling seen with sense probe. (c) All three intensely labeled RGCs were retrogradely labeled with FG (arrows). 3v, third ventricle; oc, optic chiasm; scn, suprachiasmatic nucleus; gcl, ganglion cell layer; inl, inner nuclear layer; onl, outer nuclear layer. Scale bar, 200 µm (a), 50 µm (b, c). nature neuroscience • volume 4 no 12 • december 2001

lateral and contralateral to the FG injection. Although the extent of retrograde labeling differed between cases, approximately 70% of RGCs that were intensely labeled for melanopsin mRNA were also retrogradely labeled. Both calculations are likely to underestimate the actual percentage of colocalization, because technical factors limit the efficiency of the combined labels. Therefore, most RGCs that project to the SCN express melanopsin, and a majority of melanopsin-containing RGCs project to the SCN. These observations suggest that RGCs that contain melanopsin are particularly well poised to provide photic information to the SCN. Melanopsin in these retinohypothalamic RGCs may therefore mediate the photic entrainment of circadian rhythms in mice lacking rods and cones. Although a high percentage of RHT RGCs express melanopsin, RHT cells may also receive other photic signals through rods and cones in intact animals. In addition, the photopigments cryptochrome 1 and 2 have been localized to RGCs of the mouse retina7. Further experiments will be necessary to determine whether cryptochromes are involved in circadian photic entrainment. However, melanopsin may now be considered a primary candidate photopigment for mediating circadian entrainment. Acknowledgements This work was supported by USPHS grants HL60292, MH62589 and HL07901. Competing interests statement The authors declare that they have no competing financial interests.

RECEIVED 11 OCTOBER; ACCEPTED 1 NOVEMBER 2001 1. Takahashi, J. S., DeCoursey, P. J., Bauman, L. & Menaker, M. Nature 308, 186–188 (1984). 2. Provencio, I. & Foster, R. G. Brain Res. 694, 183–190 (1995). 3. Yoshimura, T. & Ebihara, S. J. Comp. Physiol. A 178, 797–802 (1996). 4. Freedman, M. S. et al. Science 284, 502–504 (1999). 5. Provencio, I. et al. J. Neurosci. 20, 600–605 (2000). 6. Moore, R. Y., Speh, J. C. & Card, J. P. J. Comp. Neurol. 352, 351–366 (1995). 7. Miyamoto, Y. & Sancar, A. Proc. Natl. Acad. Sci. USA 95, 6097–6102 (1998).

a

b

c 1165



Does bouton morphology optimize axon length? Institute for Neuroinformatics, University of Zurich and ETH Zurich, Winterthurerstr. 190, 8057 Zurich, Switzerland Correspondence should be addressed to J.C.A. and K.A.C.M ([email protected] and [email protected])


The total length of cortical axons could be reduced if the parent axons maintained straight trajectories and simply connected to dendritic shafts via spine-like terminaux boutons and to dendritic spines via bead-like en passant boutons. Cortical axons from cat area 17 were reconstructed from serial electron micrographs and their bouton morphology was correlated with their synaptic targets. En passant or terminaux boutons did not differ in the proportion of synapses they formed with dendritic spines and shafts, and thus, the two morphological variants of synaptic bouton do not contribute directly to optimizing axon length. The bulk of the brain consists of axonal ‘wiring.’ In the gray matter of the neocortex, each cubic millimeter contains about 4 km of axon1, so removing axonal zigzags is not trivial: if only 0.1 mm were pruned off each axon in cat area 17, it would save 3 km of ‘wire.’ It has been proposed that one role of dendritic spines is to help optimize the length of axonal wire2,3. The idea is that instead of zigzagging through the neuropil to contact their specific target dendrites, axons simply grow in economically straight trajectories and form en passant synaptic boutons, leaving to dendrites the task of emitting spines to ‘catch’ the passing axons. This gives spiny dendrites the active role in selecting particular axons during development and learning4. New evidence pointing to the involvement of dendritic spines in forming connections has come from reports of motile spines5 and of new spines appearing during long-term potentiation6–8. However, one-fifth of cortical neurons lack spines and thus cannot exploit such a mechanism, so some degree of axonal zigzagging would seem inevitable. In all these discussions of spines, it has been completely overlooked that cortical axons too can produce spine-like structures, called terminaux boutons, whose dimensions match those of dendritic spines. Do terminaux boutons exist to prevent zigzags in the parent axons by connecting to dendritic shafts by means of axonal ‘spines’? Evidence in support of this comes from the pyramidal cells a b c of layer 6, which, unusually for pyramidal Bouton morphology and target specificity 80 cells, form synapses mainly with dendritic n = 43 n = 49 70 shafts9 and whose axons bear mainly ter60 minaux boutons 9,10 . This is the mirror50 image of the pattern for spiny cells in other 40 30 cortical layers whose axons bear mainly 20 en passant boutons and form synapses 10 mainly with spines. These observations sug0 Spines - Dendrites gest an ‘only connect’ hypothesis3 in which Dendrites Spines terminaux rmin Terminaux En passant es dendrite both dendritic and axonal spines are speFig. 1. Reconstructions from serial electron microscope sections of segments of axon and cializations that allow axons to maintain summary histograms. (a) Axon, light blue; target dendritic shafts, red; target spines, orange; economically straight trajectories. We testdendritic postsynaptic densities, yellow. (b) Axon, light blue; spine, transparent brown; posted this hypothesis by correlating the target synaptic density, yellow; unlabeled myelinated axon, transparent mauve. Scale bars (a, b), 1 µm. (dendritic spine/shaft) with the bouton type (c) Histograms of all identified targets of the two bouton types for all 92 synapses. Electron in the same local region of neuropil. microscope reconstructions created using Nuages and Blue Moon Rendering Tools. Percent


John C. Anderson and Kevan A. C. Martin

We examined the axons of 3 spiny neurons, which were recorded in area 17 of anesthetized cats (protocols approved by the Veterinary Department of the Canton of Zurich; for details, see ref. 10). The neurons were filled intracellularly with horseradish peroxidase. Two were layer 3 pyramidal cells with extensive axon collaterals in layers 2, 3 and 5. The third cell was a spiny stellate cell from layer 4A whose axon arborized in layers 2, 3 and 4. We selected collateral segments located in layer 3 that had a mix of both bouton types. Eighty-five boutons were serially sectioned, photographed in the electron microscope (EM), and reconstructed together with their targets. The morphological type of each bouton was determined from the serial EM reconstructions. The synaptic targets of the labeled boutons were positively identified using established criteria. Spines and dendritic shafts were the only synaptic targets and were readily distinguished. Forty-five of the boutons were en passant and formed 49 synapses, and 40 boutons were terminaux and formed 43 synapses. As is typical for spiny cortical neurons, spines were the major target. Of the 34 synapses formed by the spiny stellate axon, 24 (70%) were with spines and similarly, of 58 synapses formed by the two pyramidal cell axons, 37 (64%) were spines. In one reconstruction of a 20-µm length of the spiny stellate axon (Fig. 1a), all but one of the boutons were located in a tight cluster, where they formed synapses with dendritic shafts and spines. This cluster illustrates the morphological variety of terminaux boutons and the size of their necks relative to the parent axon. The target spines were of the same dimension as the presynaptic boutons. At the center of the cluster was a single en passant bouton. Terminaux boutons were formed even when the parent axon actually touched the target. In the example in Fig. 1b, the synaptic target (a spine) of the spiny stellate axon was directly on the path of the parent axon. Yet, instead of forming an en passant bouton, the axon formed a terminaux bouton, whose slender neck had to wrap around a myelinated axon (approximately 0.8 µm diameter) to reach its target spine. Even when the targets were aligned along the same path as the axon, multiple synapses were rarely formed. In the only two cases found, two closely spaced en passant boutons formed two synapses with the dendritic shafts of smooth neurons, before the trajectories of the axon and dendrites deviated (see supplementary reconstructions, available on the Nature Neuroscience web site). In the analysis of the distribution of all the targets of the 3 axons (Fig. 1c), spines formed 63% of the targets of terminaux boutons and 69% of en passant boutons. Dendritic shafts formed

1166





the remainder of the targets. The distributions for the two bouton types were not significantly different (chi-square test). The morphology of the bouton is not correlated with the type of synaptic target. It seems that axons and dendrites have no trouble finding each other, regardless of bouton type or whether the target is dendritic spine or shaft. Thus, the ‘only connect’ hypothesis fails to account for the data. It is tempting to write off the differences between the two bouton types as being of any particular significance. Are they simply an expression of some developmental quirk that produces one or the other kind of bouton? However, the morphological similarity of the terminaux boutons to dendritic spines prompts another interpretation. Perhaps, as with dendritic spines, which compartmentalize calcium11–13, the fine necks of these axonal spines also prevent calcium from diffusing rapidly into the parent axon during an action potential? Because each bouton would retain more residual calcium after each impulse, synapses formed by terminaux boutons might show more presynaptic facilitation than the en passant boutons. The layer 6 pyramidal synapses, which are formed mainly by terminaux boutons, show strong facilitation due to an increased probability of transmitter release14. If bouton morphology does indeed influence synaptic dynamics, it will be a revision of the present view that cortical axons exist only to connect.

Passive eye displacement alters auditory spatial receptive fields of cat superior colliculus neurons J. Curtis Zella1, John F. Brugge1 and Jan W. H. Schnupp1,2 1 Department of Physiology and Waisman Center,

627 Waisman Center, University of Wisconsin, Madison, Wisconsin 53705

2 Laboratory of Physiology, Oxford University, Parks Road, OX1 3PT

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


The superior colliculus (SC) is thought to use a set of superimposed, topographically organized neural maps of visual, auditory, somatosensory and motor space to direct the eyes toward novel stimuli1,2. Auditory spatial response fields (SRFs) of SC neurons may change when an animal moves its eyes, presumably to compensate for the resulting misalignment of visual and auditory sensory spatial reference frames3–6, but the mechanisms responsible for these SRF changes remain unknown. Here we report that passive deviation of the eye in anesthetized, paralyzed animals can profoundly affect the auditory responsiveness of SC neurons, but seems insufficient by itself to provide adaptive shifts of auditory SRFs. In awake animals, changes in eye position either shift the rostral edge or the center of auditory SRFs in the SC 3–6, or modulate the overall strength of auditory responses while not systematically shifting the SRF5,6. It is unknown whether these SRF changes are mediated through efference copy of eye movement commands, through sensory feedback from the oculomotor plant, or through some combination of the two. nature neuroscience • volume 4 no 12 • december 2001

Note: Supplementary reconstructions are available on the Nature Neuroscience web site (http://neuroscience.nature.com/web_specials).

ACKNOWLEDGEMENTS We thank T. Binzegger for help with calculations and reconstructions. Additional support from EU (QULG3-1999-01064) and HFSP (RG0123/2000-B) grants to K.A.C.M.

RECEIVED 3 AUGUST; ACCEPTED 24 OCTOBER 2001 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.

Braitenberg, V. & Schüz, A. Anatomy of the Cortex (Springer, Berlin, 1991). Peters, A. & Kaiserman-Abramof, I. R. Am. J. Anat. 127, 321–356 (1970). Swindale, N. V. Trends Neurosci. 4, 240–241 (1981). Gray, E. G. Trends Neurosci. 5, 5–6 (1982). Fischer, M., Kaech, S., Knutti, D. & Matus, A. Neuron 20, 847–854 (1998). Engert, F. & Bonhoeffer, T. Nature 399, 66–70 (1999). Maletic-Savatic, M., Malinow, R. & Svoboda, K. Science 283, 1923–1927 (1999). Toni, N., Buchs, P.-A., Nikonenko, I., Bron, C. R. & Muller, D. Nature 402, 421–425 (1999). McGuire, B. A., Hornung, J.-P., Gilbert, C. D. & Wiesel, T. N. J. Neurosci. 4, 3021–3033 (1984). Martin, K. A. C. & Whitteridge, D. J. Physiol. (Lond.) 353, 463–504 (1984). Muller, W. & Connor, J. A. Nature 354, 73–76 (1991). Guthrie, P. B., Segal, M. & Kater, S. B. Nature 354, 76–80 (1991). Majewska, A., Tashiro, A. & Yuste, R. J. Neurosci. 20, 8262–8268 (2000). Tarczy-Hornoch, K., Martin, K. A. C. & Stratford, K. J. Cereb. Cortex 9, 833–843 (1999).

Extraocular proprioceptive signals reach the SC8,9, but modeling studies have so far emphasized the importance of efference copy and visual feedback in SC motor function (for example, see ref. 10). Furthermore, the motor aspects of SC function can operate accurately when proprioceptive feedback is abolished11. We assessed the involvement of proprioceptive feedback in sensory processing within the SC by testing whether passive eye displacement alters auditory SRFs in anesthetized (∼1% halothane, 66% N2O), paralyzed (pancuronium bromide, 1 mg/kg every 3 h) cats in complete darkness. In this preparation, neither efference copy nor visual input could contribute to any observed changes. Experimental protocols were approved by the University of Wisconsin Institutional Animal Care and Use Committee. Twenty neurons from 11 cats were tested for the effects of passive eye movements. The position of the contralateral (left) eye was manipulated by tension on four sutures (6-0 silk) passed through the nasal, temporal, ventral and dorsal margin of the sclera. The sutures were attached to a mechanical device which held the eye securely either in its straight-ahead (control) position, or displaced by ∼23° in a contralateral (temporal) and downward direction. This displacement was chosen for maximal proprioceptive activation (stretching both the superior and medial recti, and possibly the superior oblique, while relaxing the inferior and lateral recti). Tension on the sutures was maintained for the control and the deviated conditions. We used short (10–100 ms) Gaussian noise bursts delivered in virtual acoustic space (VAS) and standard extracellular recordings12 to map auditory SRFs of single SC neurons. SRFs were constructed from 3 to 5 randomly interleaved stimulus presentations at each of 324 different VAS directions. In some cases, we repeated SRF measurements in the eyedeviated and the control position several times, and at different sound intensities. Recording sites were confirmed histologically to be distributed evenly throughout the central two-thirds of the intermediate and deep layers of the SC. With the eyes in their control position, SC neurons exhibited circumscribed 1167



Elev S tart

Elev S tart

Elev S tart Elev


Elev Start

C296 Fig. 1. Contraction of SRF resulting 252 spikes Control 90 a from passive eye displacement. To 2 60 1 .3 30 the left of each SRF is a dot raster 0 .7 0 taken for all azimuthal positions 0 -3 0 along an elevation of +10º (arrow) 180 135 90 45 0 -4 5 -9 0 -1 3 5 -1 8 0 A zim S tart through the SRF. Right, ‘raw’ spatial 1 response profile data. (Diameters of the filled circles are proportional to C296 the mean spike count evoked at the Eye displaced 65 spikes 90 b virtual sound direction indicated by 2 60 1 .3 the axes.) The SRFs in the center 30 0 .7 0 were generated from the raw data 0 -3 0 by interpolation (Delaunay triangu180 135 90 45 0 -4 5 -9 0 -1 3 5 -1 8 0 A zim S tart lation), smoothed (30° wide running 2 averaging filter) and plotted using an equal area projection. Zero C 2 96 c 379 spikes Return degrees azimuth is at the midline, 90 2 negative directions to the cat’s 60 1 .3 30 left. All data were recorded 0 .7 0 0 in complete darkness. Number -3 0 below each SRF marks the order in 1 80 1 35 90 45 0 -4 5 -9 0 -1 35 -1 80 A zim S tart 3 which the SRF was recorded. The black contour line delineates the d ‘best area’ (>75% of maximum). The Eye displaced C296 cross indicates SRF centroid direc- –180∫ 81 spikes 90 tion. To test for statistical signifi2 60 1 .3 cance, the mean-squared-difference 30 0 .7 0∫ 0 between two SRFs was compared 0 -3 0 against a bootstrapped null hypothe180 135 90 45 0 -4 5 -9 0 -1 3 5 -1 8 0 A zim S tart sis distribution. Z-scores outside 180∫ 4 0 10 20 30 40 50 0021-3 the interval (–2.33, 2.33) indicate ms statistical significance (α = 0.01). Z0 1 .6 scores (a) versus (b), Z = 9.97; (b) Spikes/stimulus versus, (c), Z = 12.26; (c) versus (d), Z = 13.44; (a) versus (c), Z = 1.42; (b) versus (d), Z = –0.0126. Z-scores, shown for each intensity. Not shown because of space constraints are ‘return’ SRFs obtained at 20 and 35 dB, which are very similar to ‘controls.’ (For supplementary Methods, see the Nature Neuroscience web site.)

SRFs with a defined ‘best area’ and a gradient of response strength radiating from it, as described by others13,14. In 8 units (40%), static eye displacement resulted in a significant change in SRF size and in overall spike count (Figs. 1–3). When the eye was returned to its resting position (return), the SRF returned toward its original control configuration. These changes could be observed over a range of stimulus intensities (Fig. 2). Half the units that exhibited a significant dependence on eye position responded with an increase of SRF size and magnitude, whereas the other half showed a marked decrease Eye Displaced

Control 10 dB

Z = 0.08

3

25 dB

5

Z = 5.65

2

45 dB

4

Z = 7.70

1

6 9939-15

Fig. 2. Expansion of spatial response field (SRF) resulting from passive eye displacement obtained at three sound intensities above the neuron’s acoustic threshold. SRF representation as in Fig. 1. To compare changes in SRF across intensity, data were normalized to the maximum response at each intensity. Maximal spike counts, 10 dB, 0.81; 25 dB, 2.66; 45 dB, 1.43.

1168

(Fig. 1). Systematic shifts in the centroids of the SRF were not observed (Fig. 3). SRFs were stable over hours of recording, and effects were independent of the order in which the SRFs were obtained. Eye displacement alone generated neither evoked activity nor changes in background firing level. Six additional neurons exhibited reversible changes in SRFs, but the results were not statistically significant. The remainder showed no systematic relationship to changes in eye position. The finding that passive eye displacement produced no systematic shifts in SRF centroids suggests that mechanisms underlying the compensatory shifts of auditory SRFs described previously in awake monkey SC3,4 may not have been engaged in our preparation. Our data are, however, very similar to previous observations showing a modulation of overall responsiveness of SC5,6 and IC7 neurons in awake animals which is not accompanied by SRF shifts. We could rule out corollary discharge or efference copy of motor signals due to voluntary gaze shifts, as well as visual input, as our animals were anesthetized, paralyzed and in the dark. Thus, we conclude that periorbital proprioceptive feedback seems capable of profoundly influencing auditory responsiveness, providing a modulating influence over brainstem circuits underlying auditory SRFs. These modulations of auditory SRF size and changes in overall response magnitude may, in the alert animal, work in concert with other mechanisms underlying compensatory shifts in the location of an SRF. Note: Supplementary Methods are available on the Nature Neuroscience web site (http://neuroscience.nature.com/web_specials).




Fig. 3. Population results. (a) Shifts in spatial response field (SRF) centroids with changes in eye position. Data shown for 12 neurons. (Eight neurons were excluded because their relatively poor spatial tuning made centroid direction an unreliable measurement.) Some neurons were tested at multiple sound levels, leading to a total of 17 data sets. Crosses, control and eye return conditions; triangles, eye displacement condition. Displacement direction shown by arrow. (b) Histograms summarizing distribution of centroid shifts shown in (a) and Z-scores for all SRFs in our sample (see Fig. 1 legend). Yellow shading denotes significant differences between control and eye-displaced conditions. (c) Azimuthal functions through the SRF center for four neurons showing SRF expansion or contraction with relatively small shifts in the response peak. Red, control or return; blue, eye displaced; open, original condition; closed, repeat.

a

b

10 Azimuth

10

N = 37

N = 64

Count

Count

5 0

5

5 10 Elevation –20

c

1

–10

0 Degrees

EL 10o

10

20

9903-12

0.8

0 –20

2

–10

0 * Z-Score

EL 10o

10

20

0021-3

1.5

RECEIVED 8 JUNE; ACCEPTED 9 OCTOBER 2001

0.6 Average Spike Count


15o

1

0.4 0.2

0.5

0

0

2

EL 10o

0021-2

2.5

EL 20o

9939-15

2

1.5

1.5

1

1 0.5

7.

0.5

0

0 –180

–92.57

0

92.57

180

1. 2. 3. 4. 5. 6.

–180

–92.57

0

92.57

180

Azimuth (deg)

ACKNOWLEDGEMENTS Supported by NIH grants DC00116 and HD03352 and a Defeating Deafness/Dunhill Research Fellowship to J.W.H.S.

Memory retrieval impairment induced by hippocampal CA3 lesions is blocked by adrenocortical suppression Benno Roozendaal1, Russell G. Phillips2, Ann E. Power1, Sheila M. Brooke2, Robert M. Sapolsky2 and James L. McGaugh1 1 Center for the Neurobiology of Learning and Memory, and Department of

Neurobiology and Behavior, University of California, Irvine, California 92697-3800, USA 2 Department of Biological Sciences, Stanford University, Stanford, California

94305-5020, USA Correspondence should be addressed to B.R. ([email protected])


There is evidence that in rats, partial hippocampal lesions or selective ablation of the CA3 subfield can disrupt retrieval of spatial memory1 and that hippocampal damage disinhibits hypothalamic-pituitary-adrenocortical (HPA)-axis activity, thereby elevating plasma levels of adrenocorticotropin and corticosterone2,3. Here we report evidence that attenuation of CA3 lesioninduced increases in circulating corticosterone levels with the nature neuroscience • volume 4 no 12 • december 2001

8. 9. 10. 11. 12. 13. 14.

Jay, M. F. & Sparks, D. L. J. Neurophysiol. 57, 22–34 (1987). Stein, B. E. & Meredith, M. A. (MIT Press, Cambridge, 1993). Jay, M. F. & Sparks, D. L. Nature 309, 345–347 (1984). Jay, M. F. & Sparks, D. L. J. Neurophysiol. 57, 35–55 (1987). Peck, C. K., Baro, J. A. & Warder, S. M. Exp. Brain Res. 103, 227–242 (1995). Hartline, P. H., Pandey Vimal, R. L., King, A. J., Kurylo, D. D. & Northmore, D. P. M. Exp. Brain. Res. 104, 402–408 (1995). Groh, J. M., Trause, A. S., Underhill, A. M., Clark, K. R. & Inati, S. Neuron 29, 509–518 (2001). Abrahams, V. C. Prog. Brain. Res. 50, 325–334 (1979). Nelson, J. S., Meredith, M. A. & Stein, B. E. J. Neurophysiol. 62, 1360–1374 (1989). Quaia, C., Lefevre, P. & Optican, L. M. J. Neurophysiol. 82, 999–1018 (1999). Guthrie, B. L., Porter, J. D. & Sparks, D. L. Science 221, 1193–1195 (1983). Brugge, J. F., Reale, R. A. & Hind, J. E. J. Neurosci. 16, 4420–4437 (1996). Middlebrooks, J. C. & Knudsen, E. I. J. Neurosci. 4, 2621–2634 (1984). King, A. J. & Hutchings, M. E. J. Neurophysiol. 57, 596–624 (1987).

synthesis inhibitor metyrapone, administered shortly before water-maze retention testing, blocks the impairing effects of the lesion on memory retrieval. These findings suggest that elevated adrenocortical activity is critical in mediating memory retrieval deficits induced by hippocampal damage. Male Sprague–Dawley rats (10–12 weeks old) from Charles River Laboratories (Wilmington, Massachusetts) were housed individually with food and water ad libitum (lights on 7.00– 19.00 h). All procedures were done in compliance with NIH guidelines and were approved by UC Irvine’s Institutional Animal Care and Use Committee. Rats were anesthetized with sodium pentobarbital (50 mg/kg, intraperitoneally), and a low dose of kainic acid (0.12 µg per 3.0 µl of phosphate buffer, pH 7.4; Sigma, St. Louis, Missouri) was infused bilaterally over 180 seconds at the following coordinates: –3.3 mm from bregma; ±2.0 mm from midline; –3.5 mm from skull surface4. For sham lesions, the needle tip was inserted but no infusion was given. Lesion size was determined in 35-µm brain slices by measuring the length and width of damage with a calibrated ocular grid. Areas of damage in seven coronal sections equidistantly spaced along the rostrocaudal axis were used to determine the volume of damage. In agreement with previous evidence5, damage as identified by this technique was restricted to the pyramidal neurons of CA3 region (Fig. 1). Right and left hemispheric lesion volumes were averaged to calculate total lesion volume. The four groups of lesioned rats did not differ in total lesion volumes (ANOVA, F3,56 = 0.42; p = 0.74; data not shown). Water-maze training commenced 8–10 days after surgery and consisted of 4 trials on each of 2 consecutive days. Training and testing were conducted between 10.00 and 14.00 hours, at the 1169




a

Fig. 1. Kainic acid infusions induce selective damage to the CA3 field with pyramidal neuron loss and gliosis. (a) Lower magnification. Scale bar, 500 µm. (b) Higher magnification. Scale bar, 200 µm. DG, dentate gyrus; CA1, CA3, Ammon’s horn. Arrowheads point to damaged area.

b

nadir of the rat circadian cycle for corticosterone. Four starting positions were spaced equally around a black circular water tank (25°C; diameter, 1.83 m; height, 58 cm). An escape platform (20 × 25 cm) was submerged in a fixed location at a depth of 2.5 cm below the water surface. After mounting the platform, the rats were retained on it for 10 seconds and were then placed in a holding cage for 20 seconds between trials. A two-way ANOVA of training escape latencies indicated that the sham-operated and CA3-lesioned rats did not differ in acquisition performance (F1,672 = 1.08; p = 0.30; Fig. 2a). The lesioned animals were also not impaired on the first trial of the second day of training (which is a retention test), indicating that on day 1, they had acquired procedural aspects of the task (swimming away from the sides of the tank) without acquiring explicit spatial information. Memory for the platform location was tested on day 3 with a 60-second free-swim probe trial using a novel starting position. Ninety minutes before the probe trial, they received an injection of vehicle or metyrapone, a corticosterone-synthesis inhibitor that prevents corticosterone secretion above basal levels, even in response to major stressors6,7. Metyrapone (35 mg/kg, subcutaneously; Sigma) was dissolved in 40% polyethylene glycol and 60% saline. Vehicle-injected control rats spent significantly more time in the target quadrant than chance level (Fig. 2b). Retention of vehicle-treated rats with CA3 damage was impaired relative to that of sham-operated controls, as indicated by less time spent in the target quadrant

a

Fig. 2. Effect of metyrapone (35 mg/kg) injections 90 min before the retention test on CA3 lesion-induced changes in water-maze spatial performance and plasma corticosterone. (a) Training escape latencies (mean ± s.e.m.) on a water-maze spatial task. (b) Time spent in the target quadrant (mean ± s.e.m.) on the probe trial (ANOVAs, lesion effect, F1,65 = 6.80, p = 0.01; metyrapone effect, F1,65 = 1.06, p = 0.31; lesion × metyrapone effect, F1,65 = 4.50, p = 0.04). (c) Latencies (mean ± s.e.m.) to cross the platform location on the probe trial (ANOVAs, lesion effect, F1,65 = 6.32, p = 0.01; metyrapone effect, F1,65 = 5.41, p = 0.02; lesion × metyrapone effect, F1,65 = 6.96, p = 0.01). (d) Plasma corticosterone levels (mean ± s.e.m.) as assessed immediately after the probe trial (ANOVAs, lesion effect, F1,65 = 8.01, p = 0.006; metyrapone effect, F1,65 = 7.63, p = 0.008; lesion × metyrapone effect, F1,65 = 14.03, p = 0.0004). Double asterisk, p < 0.01 CA3 lesion compared to sham lesion; Single diamond, p < 0.05; double diamond, p < 0.01 metyrapone compared to vehicle; single circle, p < 0.05; double circle, p < 0.01 corticosterone compared to vehicle. veh, vehicle; cort, corticosterone (n = 13–20/group).

1170

c

(Fisher’s, p < 0.01). Lesioned rats also had longer latencies to cross the platform location (Fisher’s, p < 0.01; Fig. 2c). CA3 lesions did not influence total path lengths on the probe trial (Fisher’s, p = 0.32; data not shown). These results indicate that the CA3 lesions impaired retention of spatial information and that the probe trial performance impairment was not due to attenuation of motor performance or swimming pattern. Corticosterone levels were determined by radioimmunoassay in plasma from trunk blood collected within 90 seconds after the probe trial (see ref. 6, for Methods). Corticosterone levels were significantly elevated in vehicle-injected rats with CA3 damage relative to sham-operated controls (Fisher’s, p < 0.01; Fig. 2d). Consistent with previous findings6,7, metyrapone treatment did not affect plasma corticosterone levels or retention performance in sham-operated controls. However, metyrapone blocked the lesion-induced elevation of corticosterone levels (Fisher’s, p < 0.01; Fig. 2d) and blocked the retention impairment induced by CA3 lesions, as assessed both by time spent in the target quadrant (Fisher’s, p < 0.05; Fig. 2b) and by platform-crossing latencies (p < 0.01; Fig. 2c). Metyrapone did not affect total path lengths (Fisher’s, p = 0.23 and p = 0.29, respectively; data not shown). Metyrapone is known to induce anxiolytic and/or sedative effects and may also directly influence adrenal steroid receptors, which could potentially alter water-maze probe trial performance7. However, such alternative explanations of the findings were excluded, as metyrapone did not block retention impairment in CA3-lesioned rats given corticosterone supplementation (2.0 mg/kg, subcutaneously; Sigma) 30 minutes before the probe trial. Time spent in the target quadrant (two-tailed ttest, p < 0.05; Fig. 2b) and platform-crossing latencies (p < 0.01; Fig. 2c) were both impaired in corticosterone-injected rats relative to those of rats given a vehicle injection (4% ethanol in saline), and their performance resembled that of CA3-lesioned

b

d





rats without metyrapone treatment. These findings strongly suggest that the metyrapone-induced attenuation of retention impairment is due to inhibition of corticosterone synthesis. The retention-probe trial findings confirm previous reports that CA3 lesions impair retention of water-maze spatial training (H.-A. Steffenach, E.I. Moser & M.-B. Moser, Soc. Neurosci. Abstr. 25, 649.4, 1999) 5 and elevate plasma corticosterone2,3. We found that normalizing corticosterone levels at the time of the probe trial is sufficient to block the retention impairment induced by CA3 damage. These findings strongly suggest that CA3 damage-induced retention deficits of this magnitude are due to influences on memory retrieval mediated by HPA-axis dysregulation. We reported previously that acute exposure to stress of glucocorticoids shortly before retention testing impairs memory retrieval6,8. Importantly, the plasma corticosterone elevation obtained with CA3 lesions was highly comparable to levels produced by doses of acutely administered corticosterone that induce memory retrieval impairment in intact rats6,8. Furthermore, our finding that the CA3 lesions did not impair acquisition performance is congruent with previous evidence indicating that glucocorticoids do not impair acquisition or immediate recall6,8. The effects of hippocampal damage on hypersecretion of glucocorticoids are temporary and dissipate after several weeks (rodents) or months (primates) 9 . It is not yet known whether the cognitive deficits induced by lesioning of the CA3 recover with the same time course. The present findings may have implications for understanding the complex relationship between glucocorticoids, hippocampal integrity and memory function. Prolonged exposure to stress levels of

Visual stimuli activate auditory cortex in the deaf Eva M. Finney, Ione Fine and Karen R. Dobkins Psychology Department-0109, University of California, San Diego, La Jolla, California 92093, USA Correspondence should be addressed to K.R.D. ([email protected])


glucocorticoids can induce atrophy of CA3 pyramidal neurons and reduce hippocampal volume, changes associated with cognitive impairments10–12. Our findings suggesting that such cognitive impairments result directly from elevated glucocorticoid levels or HPA-axis dysregulation may contribute to the development of new strategies in the treatment of memory disorders following hippocampal damage or sustained hypercortisolemia.

ACKNOWLEDGEMENTS We thank Q. Griffith, G. Hui and J. Buranday for technical assistance. Supported by USPHS NIMH Grant MH12526 (J.L.M.), NIH RO1 MH53814 (R.M.S.) and the Adler Foundation (R.M.S.).

RECEIVED 24 JULY; ACCEPTED 29 OCTOBER 2001 1. Moser, M.-B. & Moser, E. I. J. Neurosci. 15, 7535–7542 (1998). 2. Fischette, C. T., Komisaruk, B. R., Edinger, H. M. & Siegel, A. Brain Res. 195, 105–109 (1980). 3. Herman, J. P. et al. J. Neurosci. 9, 3072–3082 (1989). 4. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates (Academic, New York, 1997). 5. McLaughlin, J. et al. Proc. Natl. Acad. Sci. USA 97, 12804–12809 (2000). 6. de Quervain, D. J.-F., Roozendaal, B. & McGaugh, J. L. Nature 394, 787–790 (1998). 7. Roozendaal, B., Bohus, B. & McGaugh, J. L. Psychoneuroendocrinology 21, 681–693 (1996). 8. de Quervain, D. J.-F., Roozendaal, B., Nitsch, R. M., McGaugh, J. L. & Hock, C. Nat. Neurosci. 3, 313–314 (2000). 9. Jacobson, L. & Sapolsky, R. Endocr. Rev. 12, 118–134 (1991). 10. Sapolsky, R. M. J. Neurosci. 6, 2240–2244 (1986). 11. Arbel, I., Kadar, T., Silbermann, M. & Levy, A. Brain Res. 657, 227–235 (1994). 12. McEwen, B. S. Ann. Rev. Neurosci. 22, 105–122 (1999).

Previous brain imaging studies have demonstrated responses to tactile and auditory stimuli in visual cortex of blind subjects, suggesting that removal of one sensory modality leads to neural reorganization of the remaining modalities1–3. To investigate whether similar ‘cross-modal’ plasticity occurs in human auditory cortex, we used functional magnetic resonance imaging (fMRI) to measure visually evoked activity in auditory areas of both early-deafened and hearing individuals. Here we find that deaf subjects exhibit activation in a region of the right auditory cortex, corresponding to Brodmann’s areas 42 and 22, as well as in area 41 (primary auditory cortex), demonstrating that early deafness results in the processing of visual stimuli in auditory cortex.

Fig. 1. Visual stimuli activate auditory cortex in the deaf. Shown is an anatomical scan averaged across all deaf and hearing subjects. Auditory regions of interest (ROIs, green regions) and voxels activating differentially in deaf versus hearing subjects in response to the visual motion stimulus (colors defined in scale bar) are shown on axial (left), coronal (middle) and sagittal (right) sections of an averaged anatomical brain, transformed into the standard stereotaxic space of Talairach and Tournoux5. The area of visual responsiveness falls within Brodmann’s areas 41, 42 and 22 in the right auditory ROI. Crosshairs highlight a voxel within the area of main effect that maps to Brodmann’s area 41 (primary auditory cortex). Scale bar indicates the functional intensity (FIT) value, or magnitude of activation. L, left; R, right; A, anterior; P, posterior side of brain. nature neuroscience • volume 4 no 12 • december 2001

1171




Fig. 2. Mean activation in deaf versus hearing subjects to visual stimuli. Mean functional intensity (FIT) values, corresponding to magnitude of activation, for deaf (white bars) and hearing (gray bars) subjects within the area of main effect shown in Fig. 1. Visual stimuli produced significant activation in deaf but not hearing subjects. For comparison, auditory-evoked activity in this region of hearing subjects is also shown. Error bars denote standard errors of the means.

Six profoundly deaf and six hearing subjects were tested (3 females in each group; deaf, 27.0 ± 5.7 years old; hearing, 26.8 ± 2.6 years old). All subjects were right handed with normal or corrected-to-normal vision. All protocols were conducted in compliance with the University of California at San Diego’s Human Subjects Committee Institutional Review Board. Using fMRI (AFNI software4, 1.5 T BOLD; 28 slices; TR, 4 s; voxel size, 3 × 3 × 6 mm), we defined an auditory region of interest (ROI) by measuring responses elicited by auditory stimuli (music sequences) in our hearing subjects. In Fig. 1, auditory ROIs (green regions) are plotted on an anatomical scan averaged across all deaf and hearing subjects, after transforming individual anatomies into standard Talairach and Tournoux coordinate space5. Based on Talairach and Tournoux coordinates, auditory stimuli were found to activate regions in both right and left auditory cortex, including Brodmann’s areas 41, 42 and 22, although, consistent with known hemispheric asymmetries for music processing6, the total volume of the right auditory ROI (26.1 cm3) was larger than that of the left (14.5 cm3). Analysis of visually evoked fMRI responses was limited to within these functionally defined auditory ROIs. Our visual stimulus consisted of a moving dot pattern (size, 10° diameter; speed, 7°/s; dot luminance, 590 candelas/m2 against a black background; dot size, 0.2°; dot density, 2.7%; percent dots moving coherently, 87%). On alternate runs, the stimulus was presented in either the right or left visual field (15° eccentric to a central fixation spot). The two subject groups performed comparably on a dimming task on the motion stimulus to control for attentional state (deaf, 91.5 ± 12.2% correct; hearing 94.0 ± 2.9%; deaf, 609 ± 107 ms reaction time; hearing, 654 ± 255 ms). Visually evoked activity within the right and left auditory ROIs was computed by correlating fMRI signal amplitude in individual ROI voxels to a reference function corresponding to the time course of the visual stimulus, after first correcting for individual subject movements in six dimensions by realigning images to a single reference image. Visually evoked activity significantly differed between deaf and hearing subjects in the right auditory ROI (Fig. 1, colors defined in scale bar, main effect of subject group, F1,10 = 11.12, p = 0.0038), encompassing a volume of 0.95 cm3. Although differences were also observed in the left auditory ROI, the region of effect was extremely small (0.054 cm3) and did not survive stringent statistical standards for safeguarding against false positives. Within the right ROI region of main effect, the visual stimulus caused significant activation in deaf subjects (Fig. 2), with a mean functional intensity (FIT, which reflects the magnitude of activation) value of 2.26 ± 1.37 (p = 0.0049), whereas no significant activation occurred in hearing subjects (mean FIT, –0.31 ± 1.30, p = 0.59). In comparison, the mean FIT value produced by 1172

auditory stimuli in these same voxels within hearing subjects was 6.90 ± 2.65 (p = 0.0069). Based on Talairach and Tournoux coordinates, this region of visual activation in the deaf corresponds to Brodmann’s areas 42 and 22 (secondary and association auditory areas, respectively), which includes part of the planum temporale. In addition, several voxels (0.22 cm3, ∼23% of the total region of effect) fell within area 41 (primary auditory cortex), which encompasses the medial portion of Heschl’s gyrus. We also cross-checked these coordinates against probabilistic atlases7,8 and confirmed that our region of effect included both A1 and the planum temporale. There was no main effect of visual field or interaction between visual field and subject group within this region. In a second version of these experiments, we collected fMRI responses when subjects were instructed to ignore the motion stimulus and instead perform a dimming task on the fixation spot. Here, deaf and hearing subjects differed significantly (F1,10 = 4.09, p = 0.036) within a region of the right auditory ROI (0.54 cm3, a subset of the region of difference obtained from the attend condition) that mapped onto area 42. Deaf subjects exhibited significant activity in this region (FIT coefficient, 2.85 ± 2.91, p = 0.031) and, as before, hearing subjects did not (FIT coefficient, 0.175 ± 1.43, p = 0.78). The smaller region of effect observed under the ignore condition is consistent with the general tendency for sensory cortical areas to activate less strongly to ignored than attended stimuli9,10. Nonetheless, the fact that visual activation in auditory cortex was observed in deaf subjects even when the motion stimulus was ignored attests to the robustness of the cross-modal plasticity effect. Related to the present findings, results from previous fMRI and positron emission tomography studies have suggested that the auditory regions in which we found visual activation in the deaf include areas that may also be involved in visual language processing in both deaf and hearing subjects. Specifically, Brodmann’s areas 42 and 22 in deaf subjects are activated to visual images of sign language11,12, and these same areas are activated in hearing subjects during a silent lip reading task13 (and during auditory speech tasks13). In addition, there has been previous suggestion that these auditory areas may be used for processing purely visual (that is, non-linguistic) stimuli in the deaf. However, these earlier studies used techniques with poor spatial resolution, such as electroencephalography, that could not distinguish whether visual responses in deaf subjects originated from auditory cortex or nearby visual areas14. In sum, the results of the present study using fMRI demonstrate the recruitment of auditory cortex in the deaf for the processing of purely visual stimuli. The cross-modal plasticity observed in the present study appeared predominantly in the right auditory cortex. Because our experiment used moving visual stimuli, this hemispheric asymmetry may simply reflect a predisposition for motion processing in the right auditory cortex. This possibility is supported by the finding that in hearing subjects, the right auditory cortex (specifically, the planum temporale) is specialized for processing auditory motion15. Thus, right auditory cortex in the deaf, devoid of its normal auditory input, nature neuroscience • volume 4 no 12 • december 2001




may come to serve motion processing in the visual modality. Remarkably, the reciprocal result has recently been reported in blind subjects. Here, responses to moving auditory stimuli are observed predominantly in the right visual cortex of the blind2, again suggesting a predisposition toward motion processing in the right hemisphere. Most importantly, our demonstration of cross-modal plasticity in deaf subjects, in conjunction with that observed in the blind, attests to the robust ability of the human brain to reorganize in response to the removal early in development of input from one sensory modality.

ACKNOWLEDGEMENTS Supported by an NSF grant to K.R.D. and an NRSA grant to E.M.F. We thank G. Boynton and M. Sereno for helpful discussions, G. Brown, L. Eyler Zorrilla, P. Goldin and S. Tapert for assistance with data analysis, and D. Cai for assistance with subject recruitment.


RECEIVED 11 JUNE; ACCEPTED 17 OCTOBER 2001 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15.

Sadato, N. et al. Nature 380, 526–528 (1996). Weeks, R. et al. J. Neurosci. 20, 2664–2672 (2000). Kujala, T. et al. Trends Neurosci 23, 115–120 (2000). Cox, R. W. Computers and Biomedical Research 29, 162–173 (1996). Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain (Thieme Medical, New York, 1988). Christman, S. Cerebral Asymmetries in Sensory and Perceptual Processing (Elsevier Science B.V., 1997). Westbury, C. F. et al. Cereb. Cortex 9, 392–405 (1999). Rademacher, J. et al. Neuroimage 13, 669–683 (2001). Martinez, A. et al. Nat. Neurosci. 2, 364–369 (1999). Gandhi, S. P. et al. Proc. Natl. Acad. Sci. USA 96, 3314–3319 (1999). Petitto, L.A. et al. Proc. Natl. Acad. Sci. USA 97, 13961–13966 (2000). Nishimura, H. et al. Nature 397, 116 (1999). Calvert, G. A. et al. Science 276, 593–596 (1997). Neville, H. J. Ann. NY Acad. Sci. 608, 71–91 (1990). Baumgart, F. et al. Nature 400, 724–726 (1999).

1173


articles


Distinct regulators control the expression of the mid-hindbrain organizer signal FGF8 Weilan Ye1, Maxime Bouchard2, Donna Stone1, Xiaodong Liu1, Francis Vella4, James Lee1, Harukazu Nakamura3, Siew-Lan Ang4, Meinrad Busslinger2 and Arnon Rosenthal5 1 Department of Molecular Biology, Genentech, 1 DNA Way, South San Francisco, California 94080, USA 2 Research Institute of Molecular Pathology, Vienna Biocenter, Dr. Bohr-Gasse 7, A-1030 Vienna, Austria 3 Department of Molecular Neurobiology, Institute of Development, Aging & Cancer, Tohoku University, Seiryo-machi 4-1, Aoba-ku, Sendai 980-8575 Japan 4 IGBMC, 1 rue Laurent Fries BP163, 67404 Illkirch Cedex CU de Strasbourg, France 5 Rinat Neuroscience, 3155 Porter Drive, Palo Alto, California 94304, USA

The first two authors contributed equally to this work Correspondence should be addressed to A.R. ([email protected])

Published online: 12 November 2001, DOI: 10.1038/nn761 Local expression of FGF8 at the mid/hindbrain boundary (MHB) governs the development of multiple neurons and support cells. Here we show that the paired-domain protein Pax2 is necessary and sufficient for the induction of FGF8 in part by regulating the expression of Pax5&8. A network of transcription and secreted factors, including En1, Otx2, Gbx2, Grg4 and Wnt1&4, that is established independently of Pax2, further refines the expression domain and level of FGF8 at the MHB through opposing effects on Pax2 activity. Our results indicate that the expression of local organizing factors is controlled by combinatorial interaction between inductive and modulatory factors.

During development, distinct classes of neurons and support cells appear in stereotypic locations and subsequently define the brain architecture and function. The identity and position of these cells are controlled in part by local organizing centers that emit secreted inductive factors. A well-characterized local brain organizer is found at the boundary between the midbrain (MB) and hindbrain (HB), the isthmus. This organizer is required for the development of all cell types and nuclei in the MB as well as for the development of the cerebellum in the anterior HB. Transplantation of the isthmic organizer to the diencephalon leads to the formation of an ectopic MB1, whereas transplantation to the HB results in the development of an ectopic cerebellum2. The isthmic organizer activity seems to be mediated by FGF8 (ref. 3). Despite the seminal role of FGF8 in MB and cerebellum development, the mechanism that delineates its induction and stable expression in a narrow band of isthmus cells is not fully understood. The isthmus develops within the expression domains of the paired domain protein Pax2 (ref. 4) and the homeodomain protein Engrailed-1 (En1)5. It is further demarcated by the juxtaposition of the homeodomain transcription factors Otx2, which is expressed in the MB6, and Gbx2, which is expressed in the HB7. FGF8 first appears in the anterior HB adjacent to the secreted proteins Wnt1 (ref. 4) and Wnt4 (in the chick) 8, which are expressed in the posterior MB. It has been suggested that the positioning and/or induction of FGF8 are regulated by interaction between Otx2+ and Gbx2+ cells. This is because ectopic expression of Gbx2 or Otx2 in the nature neuroscience • volume 4 no 12 • december 2001

MHB of the mouse9,10 or chick11 embryos leads either to a shift in the position of FGF8 expression or to its induction. In contrast, mutation of Gbx2 or Otx2 results abnormal positioning and eventual loss of FGF8 (refs. 12–14). However, the actual role of Otx2 and Gbx2 in the control of FGF8 expression and the spatial limits of their activities has not been defined. Furthermore, the role of other regulatory genes (such as Pax2, En1 and Wnt1) in the induction and stable expression of FGF8 at the MHB is poorly understood. To determine the pathways controlling the induction and expression pattern of FGF8 and to investigate the roles of Otx2, Gbx2, En1, Pax2, 5, 8 and Wnt in this process, we performed multiple gain- and loss-of-function studies in chick and mouse embryos. Here we provide evidence that Pax2 is necessary and sufficient for the induction of FGF8 as well as for the expression of Pax5 and Pax8 at the MHB. In contrast, Otx2 and Gbx2 are neither necessary nor sufficient for the induction of FGF8 and can only define its position within the Pax2 expression domain. We further found that Otx2 and Gbx2 act in part by having opposing influences on the expression of Grg4 (ref. 15), which is a transcriptional corepressor of Pax and En proteins16,17. Otx2 also has a non-cellautonomous role in the maintenance of FGF8 expression in part through the induction of Wnt1 (and Wnt4), which in turn act through En1 (ref. 18). These findings outline a general mechanism for the establishment of local organizing centers in the vertebrate brain, which involves combinatorial interactions between transcriptional activators, repressors and secreted proteins. 1175



articles

a

b

c

d

e

Fig. 1. Expression pattern of endogenous Otx2, Gbx2, Pax2, FGF8 and Wnt1 in the chick embryos. (a) Otx2 (red) and Gbx2 (blue) at stage 6 of chick embryogenesis. (b) Gbx2 (red) and Pax2 (blue) at stage 9. (c) Gbx2 (red) and FGF8 (blue) at stage 11. (d) Otx2 (red) and FGF8 (blue) at stage 20. (e) Pax2 (red) and Wnt1 (blue) at stage 16. (f) Wnt1 (blue) and FGF8 (red) at stage 15. (a) and (b) are whole-mount samples; (c–f) are flat-mount specimens. All are dorsal views with anterior to the left. Scale bar, 0.14 mm (a); 0.23 mm (b, e); 0.26 mm (c); 0.29 mm (d); 0.2 mm (f). ANR, anterior neural ridge; R1, rhombomere 1; MHB, mid/hindbrain boundary; FB, forebrain; MB, midbrain; HB, hindbrain.

appeared at the newly generated anterior but not posterior Otx2/Gbx2 border and only in the Otx2– cells (Fig. 2a). Likewise, scattered expression of Otx2 extending into the entire HB resulted in a posterior expansion of FGF8 expression in the adjacent Otx2– cells (Fig. 2b; also see ref. 11). By using Hoxa2 as a reference to define the R1/R2 boundary, we found that Otx2 can induce FGF8 expression in a non-cellautonomous manner only in the anterior two thirds of R1 (Fig. 2e). At the stage when the ectopic Otx2 protein is translated (around stages 13–16), this region coincides with the caudal limit of the Pax2 domain (Fig. 1e). Uniform expression of Otx2 slightly posterior to its endogenous domain resulted in a posterior shift of FGF8 expression as previously observed in transgenic mice10 or chick embryos ectopically expressing Otx2 (ref. 11, data not shown). In contrast, uniform expression of a large patch of Otx2 from the MB to R2 led to ablation of endogenous FGF8 expression and did not result in the induction of FGF8 at any other location (data not shown). These findings suggest that Otx2 stimulates FGF8 expression in non-cell-autonomous manner and only in a region that is demarcated by Pax2 expression. Otx2 also inhibits FGF8 expression cellautonomously (Fig. 2b).

f

Results Control of FGF8 expression by Otx2 Otx2, a vertebrate homologue of the Drosophila orthodenticle gene, is expressed in the forebrain and MB before FGF8 expression and delimits the anterior boundary of the isthmus as early as the headfold stage (Fig. 1a and d)6. Ectopic expression of Otx2 under the En1 promoter in the anterior HB of the mouse embryo relocates the MHB to the newly generated Otx2/Gbx2 border10. In addition, ectopic expression of Otx2 or Gbx2 in the chick embryo11 resulted in ectopic induction of FGF8. These findings led to the suggestion that Otx2+ cells co-define the MHB and thus the site of FGF8 induction. We have re-examined the role of Otx2 by expressing it at various locations along the anterior-posterior (A/P) axis of the neural tube of the chick embryo. Local uniform expression of a small patch of ectopic Otx2 straddling rhombomeres 1 (R1) and 2 (R2) caused the appearance of a second stripe of FGF8+ cells. This stripe

Fig. 2. Ectopic expression of Otx2 and Pax2 in the chick embryo. (a) Ectopic expression of mouse Otx2 (red) at the anterior HB leads to the formation of a second FGF8 (blue) expression domain at the anterior ectopic Otx2/Gbx2 boundary. Endogenous chick Otx2 is not stained. (b) Ectopic scattered expression of mouse Otx2 (red, endogenous chick Otx2 is also stained) in the HB leads to induction of FGF8 (blue) in the anterior R1. (c) Ectopic expression of Pax2 (red) leads to weak induction of FGF8 (blue) in the diencephalon. (d) Ectopic expression of Pax2 (red) from the midbrain to R4 leads to weak induction of tiny patches of FGF8 (blue, arrowhead) in the R4, but not in the midbrain or R1–R3. (e) Ectopic expression of scattered Otx2 (not stained) in the HB (same as in b) leads to induction of FGF8 (blue) in the anterior two thirds of R1 as judged by the expression domain of Hoxa2 (red). (f) Ectopic expression of scattered Otx2 (not stained, same as in b) together with Pax2 (not stained, same as in d) in the HB leads to the expansion of FGF8 (blue) beyond R3, as judged by Hoxa2 (red) expression. The upper side of the embryo is the electroporated side, whereas the lower side serves as control. All samples are flat-mounts shown in dorsal view with anterior to the left. Scale bar, 0.5 mm (a, f); 0.3 mm (b, c); 0.4 mm (d); 0.32 mm (e). Ec, ectopic; Ep, electroporation.

1176

Control of FGF8 expression by Gbx2 Gbx2, a homeobox gene related to Drosophila unplugged, is expressed in the HB7 as early as the head-fold stage, preceding

a

b

c

d

e

f




articles

Fig. 3. Ectopic expression of Gbx2 and Pax2 in the chick embryo. (a) Ectopic expression of Gbx2 (red) in the MB very close to the MHB leads to ectopic FGF8 (blue) expression at the posterior ectopic Otx2/Gbx2 boundary. (b) Ectopic expression of Gbx2 (not stained, similar to a) leads to the induction of FGF8 (blue) within the endogenous Pax2 (red) domain. (c) Ectopic expression of Gbx2 (red) in the midbrain outside of Pax2 (not stained) fails to induce ectopic FGF8 (blue). (d) Ectopic expression of Gbx2 (not stained) in the MHB slightly anterior to the endogenous Gbx2 leads to anterior shift of FGF8 (blue) and the elimination of Otx2 (red) within the ectopic Gbx2 domain. (e) Local ectopic expression of Pax2 (red) together with Gbx2 (not stained) in the midbrain leads to the formation of a second FGF8 (blue) expression domain within the midbrain, well beyond the endogenous Pax2 expression domain. Expression of the endogenous Pax2 in the MB is very weak at stage 20 shown here. (f) Ectopic expression of Gbx2 (red) together with Pax2 (not stained) in the midbrain leads to expansion of the FGF8 (blue) expression domain throughout the midbrain. All samples are flatmounts shown in dorsal view with anterior to the left. Top sides, electroporated sides; bottom sides, control sides. Scale bar, 0.34 mm.

the expression of FGF8 (Fig. 1a–c). In the mouse, ectopic expression of Gbx2 in the posterior MB under the Wnt1 promoter9 results in a relocation of the MHB to the newly generated Otx2/Gbx2 border. Likewise, ectopic expression of Gbx2 in the chick embryo11 leads to the expression of FGF8 in ectopic locations. These findings suggested that Gbx2 also defines the MHB and thus the site of FGF8 expression19. To determine the roles of Gbx2+ cells in the induction of FGF8, we next examined the consequence of expressing it at various locations along the A/P axis of the chick neural tube. Forced expression of Gbx2 in the MB in close proximity and anterior to the endogenous MHB led to a small patch of FGF8 expression at the ectopic Gbx2/Otx2 border (Fig. 3a). Ectopic FGF8 expression could only be achieved within the Pax2+ MB zone near the endogenous MHB (Fig. 3b). In contrast, ectopic expression of Gbx2 outside of the Pax2+ territory in the MB failed to induce FGF8 (Fig. 3c). In addition, expression of Gbx2 within and slightly anterior to the MHB resulted in the suppression of Otx2 (Fig. 3d) and in an anterior shift of the FGF8+ cells to the new Gbx2/Otx2 border (Fig. 3d) as observed in Gbx2 transgenic mouse9 and chick embryos11. Finally, homogeneous high-level expression of Gbx2 throughout the MB and anterior HB, which shifted the Gbx2/Otx2 boundary outside of the Pax2+ domain, led to the extinction of endogenous FGF8 expression and did not result in ectopic induction of FGF8 (data not shown). The endogenous Pax2 domain did not change in response to the new Gbx2/Otx2 boundary (Fig. 3b). Taken together, these findings support the idea that Gbx2 controls the expression of FGF8 only in a narrow region in the posterior MB defined by the presence of Pax2+ cells. Control of FGF8 expression by Pax2 Expression of the paired domain protein Pax2 also precedes FGF8 expression in the MHB and occupies both the posterior Otx2+ and anterior Gbx2+ domains (Fig. 1b and e)20. Pax2 expression is activated in the absence of Otx2 (ref. 13) or Gbx2 (ref. 12) and is eventually refined to a narrow band across the MHB, covering a region immediately anterior to the Wnt1 and the anterior part of R1 (Fig. 1e). After the onset of Pax2 expression, Pax5 and Pax8 (ref. 21) are also activated in the MHB region, and FGF8 is induced in a broad band in the anterior Gbx2+ domain, occupying a large portion of R1 (Fig. 1c). Our findings suggested that Pax2 rather than the Otx2/Gbx2 boundary might directly control the initial expression of FGF8. To nature neuroscience • volume 4 no 12 • december 2001

a

b

c

d

e

f

examine this possibility, Pax2 was ectopically expressed at various locations along the A/P axis in the chick embryo either alone or together with Otx2 or Gbx2. As reported22, electroporation of Pax2 into the neural tube of the chick embryo led to weak expression of FGF8 mRNA in the diencephalon, telencephalon (Fig. 2c) and the posterior HB (R4) (Fig. 2d), suggesting that Pax2 alone is capable of inducing FGF8. However, Pax2 did not induce FGF8 in the MB or anterior HB (R1 to R3) (Fig. 2c and d), indicating that additional regulators are required in these brain regions. We then examined whether the interaction between Pax2 and Otx2 would allow the induction of FGF8 in the MB and anterior HB. As shown above, FGF8 was not detected in the posterior R1 through R3 following ectopic expression of Otx2 (Fig. 2b and e) or Pax2 (Fig. 2d) alone. In contrast, scattered expression of Otx2 (similar to that shown in Fig. 2b) in conjunction with uniform expression of Pax2 in R1–R3 (similar to that shown in Fig. 2d) led to ectopic expression of FGF8 in Pax2+ Otx2– cells throughout R1–R3 (Fig. 2f and data not shown). Thus, induction or sustained expression of FGF8 in the anterior HB requires, in addition to Pax2, non-cell-autonomous signal from Otx2+ cells. We next determined whether interaction between Pax2 and Gbx2 is required for the induction or sustained expression of FGF8 in the MB and anterior HB. Although no ectopic expression of FGF8 was detected in the central and posterior MB following electroporation of Pax2 (Fig. 2c and d) and Gbx2 (Fig. 3a and b) alone, co-electroporation of Pax2 and Gbx2 led to the widespread expression of FGF8 in Pax2+ Gbx2+ MB cells (Fig. 3e and f). These findings indicate that Pax2 alone can induce FGF8 outside of the MHB region but that it requires the presence of Gbx2 and a non-cell-autonomous signal from the Otx2+cells for induction or stable expression of FGF8 in the MHB region. Pax2 is essential for FGF8 induction The gain-of-function experiments suggested that Pax2, Otx2 and Gbx2 might all be significant in the induction or stable expression of FGF8. To discern the exact roles and relationships between these genes in vivo, we analyzed mice deficient in Otx2, Gbx2 or 1177



articles

a

b

c

d

e

f

g

h

i

j

Pax2. Chimeric Otx2–/– mice, which do not express Otx2 in the neural tube13, still transiently expressed FGF8 at the MHB within the Pax2 domain until the 8-somite stage (data not shown and ref. 23). Likewise, Gbx2–/– mice displayed expanded and stable expression of FGF8 in the MHB region 12 (data not shown). Therefore, Otx2 or Gbx2 are not required for the initial induction of FGF8 at the MHB, but instead determine its precise position and long-term expression. We then analyzed C3H/He Pax2-deficient embryos, in which Pax2 was replaced by the β-galactosidase gene24. Although FGF8 expression is initiated in wild-type embryos at the 3-somite stage before the onset of Pax8 and independently of Pax5 (ref. 25), it is never induced in the MHB of Pax2–/– embryos at the 3–4 somite stage (Fig. 4a) or later developmental stages (Fig. 4b). Importantly, the MHB tissue is still present in these Pax2–/– embryos at the 10-somite stage, as evidenced by a normal complement of β-galactosidase-expressing cells (Fig. 4c) and by the presence of a welldefined Gbx2/Otx2 boundary (Fig. 4d–f). Thus, the correct temporal expression of Pax2, but not that of Otx2 or Gbx2, is essential for the induction of FGF8 in the MHB of the C3H/He embryos. 1178

Fig. 4. Absence of FGF8 expression in Pax2–/– mouse embryos. (a, b) Pax2–/– embryos on the C3H/He background do not express FGF8 in the MHB at the 4-somite (a) and 7-somite (b) stages in contrast to the control Pax2+/– embryos. (c) Pax2–/– C3H/He embryos still express β-galactosidase in the MHB at the 10-somite stage, indicating that the MHB tissue is still present at this time point. (d, e) Otx2 is expressed normally in Pax2–/– embryos at 0-somite (d) and 5-somite (e) stages. (f) Gbx2 is expressed normally in 3-somite Pax2–/– embryos. (g) En1 is expressed at a slightly reduced level in 8-somite Pax2–/–embryos. (h) Wnt1 is expressed normally in 7-somite Pax2–/– embryo. (i) Pax5 is expressed only at a very low level in Pax2–/– embryo at the 7-somite stage. (j) Pax8 is not expressed in Pax2–/– embryos at the 8-somite stage. Arrowhead, MHB. Som, somite.

We further examined the consequence of Pax2 deficiency for the expression of Otx2, Gbx2, Pax5, Pax8, En1 and Wnt1. The expression of Otx2 was unchanged in Pax2–/– embryos at either the 0- or 5-somite stages (Fig. 4d and e). Likewise, no change in the expression pattern of Gbx2 was detected at the 3-somite stage (Fig. 4f), supporting the idea that the regulation of Otx2 and Gbx2 is independent of Pax2. In addition, Pax2–/– embryos displayed normal expression of Wnt1 (Fig. 4h) and slightly reduced expression of En1 (Fig. 4g), arguing that Pax2 is not a general regulator of gene expression in the MHB region. In contrast, only a very low level of Pax5 mRNA (Fig. 4i) and no Pax8 expression (Fig. 4j) were detected in Pax2–/– embryos. These findings argue that Pax2 regulates Pax5 and Pax8 either directly or indirectly. Otx2 and Gbx2 have opposite effects on Grg4 The combined loss- and gain-of-function data obtained with Otx2 and Gbx2 suggested that these two proteins control the precise position (compare Fig. 1c with d)4 and stable expression rather than the initial induction of FGF8 at the MHB. As Otx2 prevents and Gbx2 stimulates FGF8 expression in a cell-autonomous manner, we first examined possible cell-intrinsic mechanisms by which these two proteins could determine the domain of FGF8 expression. Specifically, we investigated the possibility that Otx2 and Gbx2 control the position of FGF8 by regulating the expression of the transcriptional corepressor Grg4. Grg4 is a vertebrate homologue of the Drosophila Groucho gene26, which is expressed in the MB and very weakly in the dorsal HB and diencephalon15,17. Moreover, it seems to be excluded from the FGF8 expression domain in the anterior HB (Fig. 5a and b), it changes Pax5 from a transcriptional activator into a repressor16 and it suppresses the expression of En2 and FGF8 in the MHB region17. Consistent with the hypothesis that Otx2 and Gbx2 define the expression domain of FGF8 through their influence on Grg4, ectopic expression of Otx2 in the chick HB led to induction of Grg4 in R1 to R3 (Fig. 5a). Conversely, ectopic expression of Gbx2 in the MB resulted in suppression of Grg4 (Fig. 5b) and Otx2 (Fig. 3d) in this region. To further examine whether Grg4 negatively regulates Pax2 during MHB development, we generated a mutant Pax2-∆OP protein that fails to bind Grg4 due to deletion of its octapeptide (OP) interaction motif16. Ectopic expression of the mutant but not wild-type Pax2 protein led to ectopic expression of FGF8 in the midbrain (Fig. 5c and d). Thus, in the diencephalon and posterior HB, where Grg4 is not expressed, Pax2 can alone induce FGF8. In contrast, in the MHB, where Grg4 is present, the transcriptional activity of Pax2 depends on antagonistic effects of Otx2 and Gbx2 on Grg4 expression. Wnt mediates non-cell-autonomous function of Otx2 As Otx2 is important for maintaining high-level expression of FGF8 through non-cell-autonomous mechanism, we set out to nature neuroscience • volume 4 no 12 • december 2001



articles

Fig. 5. Regulation and activity of Grg4, En1 and Wnt1. (a) Ectopic mouse Otx2 (red) induces the expression of Grg4 (blue) in the HB. (b) Ectopic Gbx2 (red) suppresses Grg4 (blue) in the MB. (c) Wild-type Pax2 (red) fails to induce FGF8 (blue) in the MB. (d) The mutant Pax2∆OP protein (red), which does not bind Grg4 due to deletion of the octapeptide (OP) motif16, induces FGF8 (blue) in the MB and HB. (e) Ectopic Otx2 (red) induces the expression of Wnt1 (blue) in the HB. (f) Wnt1 (red) induces the expression of FGF8 (blue) in the anterior R1. (g) Ectopic expression of En1 (red) from the midbrain to R4 does not lead to efficient ectopic expression of FGF8 (blue). (h) Ectopic expression of En1 (not stained) together with Pax2 (not stained) in the MHB region leads to ectopic expression of FGF8 throughout the hindbrain (blue). Dorsal views are shown with the anterior end to the left. Scale bar, 0.34 mm (a, c, d); 0.36 mm (b, e, f).

determine the molecular nature of the Otx2-derived signal. One candidate we examined was the secreted protein Wnt1. The expression of Wnt1 was apparently not sufficient for the initial induction of FGF8 (compare Fig. 4h and b), but is essential for the stability of FGF8 expression and the maintenance of the MHB organizer18,27. In the chick embryo, Wnt1 and Wnt4 are induced concurrently with FGF8, in a small stripe of cells at the posterior edge of the Otx2+ boundary (Fig. 1e and f; data not shown). FGF8 is initially expressed in a wide domain in the anterior R1 (Fig. 1c) and later refined to a narrow band immediately next to the Otx2 (Fig. 1d) and Wnt1 expression domains (Fig. 1f). Wnt1 is never induced in Otx2–/– chimeric mouse embryos14, but is induced by forced expression of Otx2 in the chick HB (Fig. 5e and data not shown). More importantly, ectopically expressed Wnt1 can induce the expression of FGF8 in the anterior R1 (Fig. 5f) in a manner similar to Otx2 (Fig. 2b). Because genetic epistasis experiments have shown that Wnt1 maintains the MHB and the expression of FGF8 indirectly through the homeodomain protein En1 (ref. 18), we also examined the ability of En1 to control FGF8 expression. En1 was shown to induce FGF8 in the diencephalon28 and MB29. However, in the absence of both En1 and En2, FGF8 is still induced in the MHB region 5. En1 by itself was unable to efficiently induce FGF8 expression in the chick HB (Fig. 5g) but strengthened the induction of FGF8 by Pax2 in R1–3 (Fig. 5h), indicating that En1 has an auxiliary function in regulating the level of FGF8 expression.

DISCUSSION Local inductive centers that emit secreted proteins determine the identity and initial position of most cell types in the vertebrate brain. Here we provided evidence that the expression of FGF8, the key mediator of the MHB organizer, is controlled by combinatorial interactions between the transcription factors Pax2, En1, Grg4, Otx2 and Gbx2. Pax2 is the key inducer, whereas Otx2, Gbx2, Grg4, Wnt1 and En1 have modulatory roles. Similar combinatorial and sequential interactions between transcriptional repressors, activators and secreted proteins are likely to control the position of other local organizing centers in the vertebrate brain. The role of Otx2 and Gbx2 in the control of the MHB Transplantation experiments in the chick and mouse embryos revealed that the MHB possesses organizing activity, as it can induce an ectopic MB in the diencephalon or an ectopic cerebellum in the HB30. The molecule responsible for the patterning activity of the MHB seems to be FGF8. This is evident from the finding that FGF8 beads induce a new MHB, MB and cerebellum3, whereas the reduction of FGF8 activity31,32 leads to defects in MB and anterior HB development. nature neuroscience • volume 4 no 12 • december 2001

a

b

c

d

e

f

g

h

The mechanism controlling the formation of an FGF8-expressing MHB is not well understood. One possibility is that the MHB is formed following juxtaposition of different neural plate (planar) signals. This hypothesis was supported by the finding that the MHB can regenerate in vivo between the MB and HB following surgical removal of part of the MB33 and that an FGF8-producing MHB will develop following juxtaposition of R1 and MB tissue in explant culture34. Because the juxtaposition of the MB and HB is demarcated early by the expression of the transcription factors Otx2 and Gbx2, respectively, it has been further suggested that the mutually antagonistic action of these two proteins is central in the development of the MHB. Here we used chimeric mouse embryos, which are Otx2-deficient only in the neural tube23, to confirm that Otx2/Gbx2 boundary is not required for the initial induction of FGF8 (data not shown). Instead, our findings argue that the Otx2/Gbx2 boundary is involved in the positioning of FGF8 expression rather than in its induction. The role of Pax2 in the induction of the MHB organizer Although En1, En2, Wnt1, FGF8, Pax5, Pax8 and multiple other genes are expressed in overlapping or adjacent domains of the MHB region, their individual deletion in the C57BL/6 genetic backgrounds did not lead to ablation of FGF8 expression. We show here that Pax2 alone is essential either directly or indirectly for the induction of FGF8 as well as of Pax5 and Pax8 at the MHB in C3H/He mouse and is therefore the master regulator of FGF8. Although Pax2 is essential and sufficient for the induction of FGF8, stable expression and precise positioning of FGF8 at the MHB still requires the regulation of Pax2 activity through the opposing action of Gbx2 and Otx2 on the expression of the tran1179



articles

Fig. 6. Putative interactions between negative and positive regulators of FGF8 in the MHB. Pax2 induces FGF8 expression, whereas Otx2 and Gbx2 further refine the position and stable expression of FGF8 through their opposing effects on the expression of Grg4 and through the induction of Wnt proteins by Otx2. En, which may be induced by the mesoderm47 and sustained by Wnt1, supports the induction of FGF8 by Pax2. This model suggests that the induction and positioning of FGF8 expression in the MHB are regulated by integration of vertical (En inducer) and planar (Otx2, Gbx2, Wnt1) signals. Whether Pax2 expression is induced by vertical or planar signals remains to be determined. Arrowheads indicate positive regulation (not necessarily direct), and T-shaped bars denote negative regulation (not necessarily direct). Dashed lines indicate transient expression, and contiguous lines denote stable expression. Grg4 negatively regulates the activity of Pax and En proteins16. Di, diencephalon.

scriptional corepressor Grg4. Precise FGF8 expression also depends on the cooperative interaction of Pax2 with En1, whose expression in the HB may be controlled by the non-cellautonomous Wnt1 signal from the MB18 (Fig. 6). Involvement of non-neural tissues in FGF8 expression It has been suggested that vertical signals from underlying mesodermal tissue may be responsible, at least in part, for the induction of FGF8 in the MHB. For example, FGF8, which is secreted by the underlying cardiac mesendodermal cells3, may induce FGF8 in the MHB. However, FGF8 mRNA is transcribed in the absence of functional FGF8 protein in neural plate explant cultures31. Furthermore, FGF8 expression is abolished in Pax2–/– mice in which the cardiac mesoderm expression of FGF8 is normal (data not shown). Alternatively, it has been proposed that En1, which could be activated by FGF4 from the notochord, may induce FGF8 in the MHB29, possibly in conjunction with En2, which is also activated by vertical signals35. However, En1 does not seem to be required for the induction of FGF8 in vivo5. Finally, it is possible that Pax2 is activated in the MHB region in response to vertical signals and then, in turn, induces FGF8. However, we found that explants of chick embryos from the early-streak stage (stages 3–4) induce Pax2 in the presumptive MHB in the absence of both axial and paraxial mesodermal tissues (data not shown). Likewise, we found that Pax2 was activated in mouse explants that were separated from the head mesenchyme and axial mesoderm as soon as the neuroectoderm was induced (early E7; data not shown). These findings suggest that a subset of cells is committed to express Pax2 as soon as the neural ectoderm is induced. Whether these cells are committed to express Pax2 in response to vertical or planar signals remains to be determined. The maintenance and function of FGF8 in the MHB FGF8, Wnt, En1, En2, Pax2 and Pax5 are dependent on each other for stable expression and, when one of them is missing, the expression of the remaining genes is extinguished over time36. Conversely, these genes can induce and maintain each other’s expression at ectopic locations3,22,29,37,38. This feedback regulation involves direct interactions between Pax2 and the MHB-specific enhancer of Pax5 (ref. 21), En proteins and regulatory sequences on the FGF8 gene39 as well as Pax proteins and a regulatory element of the En2 gene5,40. This cross-regulatory network provides a mechanism to stabilize, maintain and sharpen the expression domain of FGF8. Although FGF8 is capable of inducing an ectopic MHB, MB and anterior HB, its mechanism of action is not understood. The finding that FGF8 induces an ectopic MB in the dien1180

cephalon but an ectopic cerebellum in the MB suggests that it either acts instructively in conjunction with other patterning signals or functions as a permissive or mitogenic signal for committed progenitor cells41. In addition, although a relay mechanism has not been entirely excluded, the finding that distinct genes and cell types are induced at stereotypic distances from a point source of FGF8 (ref. 42) argues for a morphogenetic function of FGF8. The specific function of the Pax, En and Wnt genes, apart from regulating FGF8 expression, is not fully understood either. Gene ablation studies suggested that Wnt1 is required for cell proliferation in the MB43,44. However, ectopic expression of Wnt1 in the chick embryo did not lead to abnormal proliferation, suggesting that Wnt1 is not rate limiting for this process. In contrast, En1 seems to be essential for the induction of axon guidance cues41 and possibly for cell survival in the MB and HB45. In summary, we have provided evidence that the precise expression domain of at least one local organizer in the vertebrate nervous system is determined in a stepwise process involving initial induction and subsequent refinement by a feedback mechanism between multiple transcription factors and secreted proteins (Fig. 6).

METHODS Electroporation of chick embryos. Electroporation was done as described previously46.The chick Pax2-∆OP mutant was generated by PCR (polymerase chain reaction)-based mutagenesis, which resulted in deletion of the entire octapeptide motif (YSINGILG; amino acids 185–192; ref. 22). All animal experiments were reviewed and approved by Genentech institutional animal use committee. In situ hybridization. Chick or mouse embryos were processed for in situ hybridization as described46. In most cases, neural tube of the chick embryos was cut open along the roof plate, and the mesoderm was partially removed to allow flat-mount preparation for photography. All chick figures shown in this paper are dorsal views of flat-mount specimens. Mutant mice. The Pax2 mutant24, Otx2 chimeric13 and Gbx2 mutant12 mice were generated, maintained and genotyped as described.

ACKNOWLEDGEMENTS We thank A. Joyner for sharing unpublished results and for the chick En1 and En2 cDNAs, G. Martin, J. Rubenstein and E. Miyashita for Gbx2 mutant mice, A. Simeone for the full-length mouse and partial chick Otx2 cDNAs, A. Leutz for the full-length chick Gbx2 cDNA, G. Martin for the chick FGF8 cDNA, A. Lumsden for the chick Hoxa2 cDNA, A. McMahon for the chick Wnt1 and Wnt4 cDNAs and S. Greenwood for reading the manuscript. This research was supported in part by Boehringer Ingelheim.



articles

RECEIVED 26 SEPTEMBER; ACCEPTED 21 OCTOBER 2001


25. 1. Martinez, S., Wassef, M. & Alvarado-Mallart, R. M. Induction of a mesencephalic phenotype in the 2-day-old chick prosencephalon is preceded by the early expression of the homeobox gene en. Neuron 6, 971–981 (1991). 2. Marin, F. & Puelles, L. Patterning of the embryonic avian midbrain after experimental inversions: a polarizing activity from the isthmus. Dev. Biol. (Orlando) 163, 19–37 (1994). 3. Crossley, P. H., Martinez, S. & Martin, G. R. Midbrain development induced by FGF8 in the chick embryo. Nature 380, 66–68 (1996). 4. Hidalgo-Sanchez, M., Millet, S., Simeone, A. & Alvarado-Mallart, R. M. Comparative analysis of Otx2, Gbx2, Pax2, Fgf8 and Wnt1 gene expressions during the formation of the chick midbrain/hindbrain domain. Mech. Dev. 81, 175–178 (1999). 5. Liu, A. & Joyner, A. L. EN and GBX2 play essential roles downstream of FGF8 in patterning the mouse mid/hindbrain region. Development 128, 181–191 (2001). 6. Boncinelli, E., Gulisano, M. & Broccoli, V. Emx and Otx homeobox genes in the developing mouse brain. J. Neurobiol. 24, 1356–1366 (1993). 7. Bouillet, P., Chazaud, C., Oulad-Abdelghani, M., Dollé, P. & Chambon, P. Sequence and expression pattern of the Stra7 (Gbx-2) homeobox-containing gene induced by retinoic acid in P19 embryonal carcinoma cells. Dev. Dyn. 204, 372–382 (1995). 8. Hollyday, M., McMahon, J. A. & McMahon, A. P. Wnt expression patterns in chick embryo nervous system. Mech. Dev. 52, 9–25 (1995). 9. Millet, S. et al. A role for Gbx2 in repression of Otx2 and positioning the mid/hindbrain organizer. Nature 401, 161–164 (1999). 10. Broccoli, V., Boncinelli, E. & Wurst, W. The caudal limit of Otx2 expression positions the isthmic organizer. Nature 401, 164–168 (1999). 11. Katahira, T. et al. Interaction between Otx2 and Gbx2 defines the organizing center for the optic tectum. Mech. Dev. 91, 43–52 (2000). 12. Wassarman, K. M. et al. Specification of the anterior hindbrain and establishment of a normal mid/hindbrain organizer is dependent on Gbx2 gene function. Development 124, 2923–2934 (1997). 13. Rhinn, M. et al. Sequential roles for Otx2 in visceral endoderm and neuroectoderm for forebrain and midbrain induction and specification. Development 125, 845–856 (1998). 14. Rhinn, M., Dierich, A., Le Meur, M. & Ang, S. Cell autonomous and non-cell autonomous functions of Otx2 in patterning the rostral brain. Development 126, 4295–4304 (1999). 15. Koop, K. E., MacDonald, L. M. & Lobe, C. G. Transcripts of Grg4, a murine groucho-related gene, are detected in adjacent tissues to other murine neurogenic gene homologues during embryonic development. Mech. Dev. 59, 73–87 (1996). 16. Eberhard, D., Jimenez, G., Heavey, B. & Busslinger, M. Transcriptional repression by Pax5 (BSAP) through interaction with corepressors of the Groucho family. EMBO J. 19, 2292–2303 (2000). 17. Sugiyama, S., Funahashi, J. & Nakamura, H. Antagonizing activity of chick Grg4 against tectum-organizing activity. Dev. Biol. 221, 168–180 (2000). 18. Danielian, P. S. & McMahon, A. P. Engrailed-1 as a target of the Wnt-1 signalling pathway in vertebrate midbrain development. Nature 383, 332–334 (1996). 19. Simeone, A. Positioning the isthmic organizer where Otx2 and Gbx2 meet. Trends Genet. 16, 237–240 (2000). 20. Dressler, G. R., Deutsch, U., Chowdhury, K., Nornes, H. O. & Gruss, P. Pax2, a new murine paired-box-containing gene and its expression in the developing excretory system. Development 109, 787–795 (1990). 21. Pfeffer, P. L., Bouchard, M. & Busslinger, M. Pax2 and homeodomain proteins cooperatively regulate a 435 bp enhancer of the mouse Pax5 gene at the midbrain–hindbrain boundary. Development 127, 1017–1028 (2000). 22. Okafuji, T., Funahashi, J. & Nakamura, H. Roles of Pax-2 in initiation of the chick tectal development. Brain Res. Dev. Brain Res. 116, 41–49 (1999). 23. Acampora, D. et al. Visceral endoderm-restricted translation of Otx1 mediates recovery of Otx2 requirements for specification of anterior neural plate and normal gastrulation. Development 125, 5091–5104 (1998). 24. Bouchard, M., Pfeffer, P. & Busslinger, M. Functional equivalence of the


26. 27.

28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42.

43. 44. 45. 46. 47.

transcription factors Pax2 and Pax5 in mouse development. Development 127, 3703–3713 (2000). Urbanek, P., Fetka, I., Meisler, M. H. & Busslinger, M. Cooperation of Pax2 and Pax5 in midbrain and cerebellum development. Proc. Natl. Acad. Sci. USA 94, 5703–5708 (1997). Mallo, M., Franco del Amo, F. & Gridley, T. Cloning and developmental expression of Grg, a mouse gene related to the groucho transcript of the Drosophila Enhancer of split complex. Mech. Dev. 42, 67–76 (1993). McMahon, A. P., Joyner, A. L., Bradley, A. & McMahon, J. A. The midbrain–hindbrain phenotype of Wnt-1-/Wnt-1- mice results from stepwise deletion of engrailed-expressing cells by 9.5 days postcoitum. Cell 69, 581–595 (1992). Araki, I. & Nakamura, H. Engrailed defines the position of dorsal dimesencephalic boundary by repressing diencephalic fate. Development 126, 5127–5135 (1999). Shamim, H. et al. Sequential roles for Fgf4, En1 and Fgf8 in specification and regionalisation of the midbrain. Development 126, 945–959 (1999). Nakamura, H., Nakano, K. E., Igawa, H. H., Takagi, S. & Fujisawa, H. Plasticity and rigidity of differentiation of brain vesicles studied in quailchick chimeras. Cell Differ. 19, 187–193 (1986). Ye, W., Shimamura, K., Rubenstein, J. L., Hynes, M. A. & Rosenthal, A. FGF and Shh signals control dopaminergic and serotonergic cell fate in the anterior neural plate. Cell 93, 755–766 (1998). Meyers, E. N., Lewandoski, M. & Martin, G. R. An Fgf8 mutant allelic series generated by Cre- and Flp-mediated recombination. Nat. Genet. 18, 136–141 (1998). Nieuwkoop, P. D. The successive steps in the pattern formation of the amphibian central nervous system. Dev. Growth Differ. 32, 149–154 (1989). Irving, C. & Mason, I. Regeneration of isthmic tissue is the result of a specific and direct interaction between rhombomere 1 and midbrain. Development 126, 3981–3989 (1999). Ang, S.-L. & Rossant, J. Anterior mesendoderm induces mouse Engrailed genes in explant cultures. Development 118, 139–149 (1993). Wurst, W. & Bally-Cuif, L. Neural plate patterning: upstream and downstream of the isthmic organizer. Nat. Rev. Neurosci. 2, 99–108 (2001). Funahashi, J. et al. Role of Pax-5 in the regulation of a mid-hindbrain organizer’s activity. Dev. Growth Differ. 41, 59–72 (1999). Liu, A., Losos, K. & Joyner, A. L. FGF8 can activate Gbx2 and transform regions of the rostral mouse brain into a hindbrain fate. Development 126, 4827–4838 (1999). Gemel, J., Jacobsen, C. & MacArthur, C. A. Fibroblast growth factor-8 expression is regulated by intronic engrailed and Pbx1-binding sites. J. Biol. Chem. 274, 6020–6026 (1999). Li Song, D. & Joyner, A. L. Two Pax2/5/8-binding sites in Engrailed2 are required for proper initiation of endogenous mid-hindbrain expression. Mech. Dev. 90, 155–165 (2000). Lee, S. M., Danielian, P. S., Fritzsch, B. & McMahon, A. P. Evidence that FGF8 signalling from the midbrain–hindbrain junction regulates growth and polarity in the developing midbrain. Development 124, 959–969 (1997). Martinez, S., Crossley, P. H., Cobos, I., Rubenstein, J. L. & Martin, G. R. FGF8 induces formation of an ectopic isthmic organizer and isthmocerebellar development via a repressive effect on Otx2 expression. Development 126, 1189–1200 (1999). McMahon, A. P. & Bradley, A. The Wnt-1 (int-1) proto-oncogene is required for development of a large region of the mouse brain. Cell 62, 1073–1085 (1990). Thomas, K. R. & Capecchi, M. R. Targeted disruption of the murine int-1 proto-oncogene resulting in severe abnormalities in midbrain and cerebellar development. Nature 346, 847–850 (1990). Wurst, W., Auerbach, A. B. & Joyner, A. L. Multiple developmental defects in Engrailed-1 mutant mice: an early mid-hindbrain deletion and patterning defects in forelimbs and sternum. Development 120, 2065–2075 (1994). Hynes, M. et al. The seven-transmembrane receptor smoothened cellautonomously induces multiple ventral cell types. Nat. Neurosci. 3, 41–46 (2000). Ang, S.-L., Conlon, R. A., Jin, O. & Rossant, J. Positive and negative signals from mesoderm regulate the expression of mouse Otx2 in ectoderm explants. Development 120, 2979–2989 (1994).

1181


articles


A Gr receptor is required for response to the sugar trehalose in taste neurons of Drosophila Anupama Dahanukar, Kara Foster, Wynand M. van der Goes van Naters and John R. Carlson Department of Molecular, Cellular, and Developmental Biology, Yale University, PO Box 208103, New Haven, Connecticut 06520-8103, USA The first three authors contributed equally to this work Correspondence should be addressed to J.R.C. ([email protected])

Published online: 12 November 2001, DOI: 10.1038/nn765 We recently identified from the Drosophila genome database a large family of G protein–coupled receptor genes, the Gr genes, and predicted that they encode taste receptors on the basis of their structure and specificity of expression. The expression of Gr genes in gustatory neurons has subsequently been confirmed and 56 family members have been reported. Here we provide functional evidence that one Gr gene, Gr5a, encodes a taste receptor required for response to the sugar trehalose. In two different mutants that carry deletions in Gr5a, electrophysiological and behavioral responses to trehalose were diminished but the response to sucrose was unaffected. Transgenic rescue experiments showed that Gr5a confers response to trehalose. The results correlate a particular taste ligand with a Gr receptor and indicate a role for G protein–mediated signaling in the transduction of sweet taste in Drosophila.

The ability of the gustatory system to detect and discriminate different taste stimuli is a universal feature of animals. In the adult fly, gustatory receptor neurons are housed in hair-like structures called sensilla1–4. These are located on the external and internal mouth parts, the tarsal segments of the legs and the anterior margins of the wings. The primary gustatory organ in the adult is the labellum, which is located at the tip of the proboscis and bears approximately 66 taste hairs1. Most labellar taste hairs are innervated by a single mechanosensory neuron and four bipolar chemosensory neurons, which have been classified as a sugar cell, a water cell and two salt cells based on their electrophysiological responses to various stimuli4. The axons of the chemosensory neurons project to the subesophageal ganglion in the brain, where gustatory information is processed1,5,6. Stimulation of these neurons by chemical cues elicits a variety of behaviors, such as proboscis extension, ingestion and proboscis retraction4. Perception of sweetness is a critical taste modality for terrestrial animals such as Drosophila that ingest sweet substances for nutrition. G protein–coupled receptors (GPCRs) have been implicated in the reception of monosodium glutamate (umami)7 and bitter taste8–10 in mammals. Recently, two vertebrate GPCRs, T1R2 and T1R3, have been shown to mediate responses to several sweet-tasting molecules, including sucrose11–14. We recently identified from the Drosophila genome database a large family of GPCR genes, the Gr genes, and predicted that they encode taste receptors on the basis of their structure and specificity of expression15. Of the first 19 Gr genes identified, 18 were found to be expressed in the labellum. Gr gene expression was not detected in a variety of other tissues or in the proboscis of pox-neuro flies, mutants in which chemosensory taste neurons 1182

are eliminated16. The expression of Gr genes in gustatory neurons has subsequently been confirmed by other researchers17,18, and 56 family members have so far been reported. However, the functional significance of Gr genes has not been demonstrated. Genetic studies in Drosophila have identified a locus that mediates response to the disaccharide trehalose19. Behavioral tests have shown that the Tre locus alters the taste sensitivity to trehalose without affecting the response to other sugars. The Tre locus has been mapped to cytogenetic region 5A using various wild-type strains and chromosomal rearrangement stocks that show differences in trehalose response19,20. A GPCR gene, CG3171, in region 5A has recently been identified; it has been named Tre1 and proposed to encode the taste receptor for trehalose21. Tre1 shows no homology to the Gr genes; rather, it has 20–22% sequence identity to vertebrate melatonin receptors. The two genes in the Drosophila genome to which Tre1 is most similar both have been annotated in the fly sequence database as probable hormone receptors (http://flybase.bio.indiana.edu version 7/01). We found that a member of the Gr family, Gr5a, is tightly linked to the Tre locus. To determine whether Gr5a plays a role in trehalose reception, we studied the response to trehalose in mutant flies that have deletions in Gr5a. We find that in flies mutant for Gr5a, the electrophysiological as well as behavioral response to trehalose is diminished. Notably, this defect is rescued by supplying a wild-type copy of Gr5a on a transgene, but not by supplying a mutant copy of Gr5a. Rescue is not dependent on the presence of a wild-type copy of Tre1. Our results thus indicate that Gr5a, a member of the Gr gene family, is required for the reception of trehalose and validate the functional role of a Gr gene as a gustatory receptor. nature neuroscience • volume 4 no 12 • december 2001


articles


Fig.1. Molecular structure of the Gr5a locus. Both Tre1 (open boxes) and Gr5a (filled boxes) transcription units are indicated. The inverted triangle depicts the insertion site of the P element in strain 496. Sequences deleted in ∆5 and ∆19 are indicated below.

RESULTS Gr5a is tightly linked to the Tre locus To test the hypothesis that Gr genes encode taste receptors, we asked whether any Gr genes reside at loci implicated in taste perception. The trehalose response locus19, whose alleles confer different levels of response to the disaccharide trehalose, has been mapped to cytogenetic region 5A on the X chromosome20. One Gr gene, Gr5a, is also located in region 5A and is thus tightly linked to this locus. Gr5a is a member of a small subfamily of eight Gr genes18. We determined the intron–exon structure of Gr5a (Fig. 1) by 5´ and 3´ RACE and RT-PCR experiments. The experiments also showed expression of Gr5a in the proboscis and the legs (not shown), both of which contain sensilla that respond to trehalose (ref. 21 and data shown below). The 5´ end of Gr5a lies less than 900 base pairs from CG3171, which has previously been reported to encode the trehalose receptor and has been named Tre1 (ref. 21 and Fig. 1). Gr5a mutants show sugar-specific physiological defects To investigate the possible role of Gr5a in trehalose reception, we used two strains of flies, each with a deletion in the Gr5aTre1 genomic region. The two deletion lines, ∆EP(X)-19 and ∆EP(X)-5, henceforth referred to as ∆19 and ∆5, were generated by imprecise excision of a P element that lies in the region between the two genes in strain EP(X)0496, henceforth referred to as 496. We determined by sequence analysis that in ∆19, the proximal endpoint of the deletion lies within the second exon of Gr5a (3´ to the codon specifying Ala 62), and in ∆5, the proximal endpoint lies in the first intron (which lies 3´ to the codon specifying Arg 36) (CG15779 in Gadfly, Release 1, http://hedgehog.lbl.gov:8002/cgi-bin/annot/query/). Thus, in each deletion

Fig. 2. Gr5a mutants show sugar-specific physiological and behavioral defects. (a) Neural responses to trehalose are diminished in the deletion lines. Shown here are concentration–response plots of mean impulse rates in recordings from labellar sensilla (n = 12–13). Error bars indicate s.e.m. The response to trehalose in 496 differs significantly from the responses in ∆19 and ∆5 (ANOVA with Scheffé post hoc tests; p < 0.001), whereas the strains do not differ significantly in their response to sucrose. (b) The behavioral response to trehalose is defective in the deletion lines. Shown are dose–sensitivity curves determined using the two-choice preference test. The tested sucrose concentrations correspond to 1, 2 and 3 mM. The dashed line indicates PI50; error bars indicate s.e.m. There is no significant difference in the sucrose response of 496, ∆5 and ∆19. For comparison of the trehalose response between the different strains, PI50 values were calculated for each experiment (n = 10 experiments; 3 concentrations per experiment) and used for statistical analysis. There is no significant difference between ∆19 and ∆5, but both are significantly different from 496 (p < 0.001). Arcsine-transformed data were analyzed using ANOVA and Sidak post hoc tests. nature neuroscience • volume 4 no 12 • december 2001

strain, the translation initiation codon of Gr5a is removed (Fig. 1). Flies that are either homozygous or hemizygous for these deletions are viable. We next asked whether the deletion mutants lacking Gr5a were defective in trehalose response, using two independent assays. First, we measured electrophysiological response to sugars in individual labellar taste hairs of the large (L) or medium (M) class22, using the tip-recording method23. L and M sensilla each have four chemosensory neurons, classified according to the stimuli that elicit a response from them: a sugar cell, two salt cells and a water cell19,24,25. The sugar cell responds to a number of different sugars, including disaccharides as sucrose and trehalose26 , which is present in yeast (an important food source of Drosophila). Individual sensilla were stimulated by placing an electrode containing the sugar over the tip of the sensillum. Action potentials are elicited in both the sugar and the water cell, but the activities of the two cells can be distinguished by their different amplitudes. Dose–response curves for sucrose and trehalose in the various strains are shown in Fig. 2a. Sucrose response was the same for all strains across a broad range of concentrations. In the same cells, however, the trehalose responses of ∆5 and ∆19 were drastically reduced compared to that of the parental strain 496. Gr5a mutants show sugar-specific behavioral defects We next tested the behavioral response to trehalose using the twochoice preference test19. This protocol compares the consumption of two sugars offered simultaneously to populations of flies.

a

b

1183



articles

Fig. 3. Gr5a rescues the physiological defect of Tre mutants. (a) Genomic rescue constructs. The wild-type construct (T+G+) contains both the Tre1 and the Gr5a transcription units. Asterisks indicate the position of stop codon mutations in either Tre1 (T–G+) or Gr5a (T+G–). (b) Sample traces of physiological recordings 200–700 ms after stimulus onset. (c) Mean physiological responses. Shown are results for the indicated concentrations, which give a response closest to half-maximum in Fig. 2a (7 ≤ n ≤ 15). Error bars indicate s.e.m.s. Trehalose was also tested at 10–1.5 M (31.6 mM) and 10–0.5 M (316 mM) and sucrose was also tested at 10–1 M (100 mM) and 10–0.5 M (316 mM), with similar results. For trehalose, Gr5a+ lines differ significantly from Gr5a– lines (ANOVA with Sidak post hoc tests, p< 0.001). Mean responses to trehalose in Canton S wild-type and 496 parental lines are also indicated. Mean responses to sucrose do not differ significantly among the lines. ‘–’ denotes the ∆19 or ∆5 line in the absence of a transgene.

a

b

In the first control experiment, we tested the preference for different concentrations of sucrose versus a standard concentration of 2 mM sucrose (Fig. 2b, left panel). Flies of all strains tested preferred 3 mM (10–2.5 M) sucrose to 2 mM sucrose, with a preference index (PI) of 0.9, and they preferred 2 mM sucrose to 1 mM sucrose (PI = 0.2). Consistent with the physiological data, there was no difference in the response to sucrose between strain 496 and the strains that are mutant for Gr5a. To investigate trehalose response, we tested several concentrations of trehalose against 2 mM sucrose. The PI50 value, the concentration of trehalose at which flies consumed as much trehalose as sucrose, was 11 mM for 496, but 77 mM for ∆5 and 76 mM for ∆19 (Fig. 2b, right). Thus, in this behavioral protocol both mutants were severely defective in their response to trehalose. Transformation rescue of trehalose response with Gr5a To determine whether loss of Gr5a expression is responsible for the reduced response to trehalose in the deletion mutants, we engineered a 10-kb genomic DNA rescue construct that included both the Gr5a and the Tre1 coding regions (Fig. 3a). To assess the contributions of each gene separately, we also generated derivatives of this 10-kb construct that included a stop codon near the N-terminus of either Gr5a or Tre1. We generated transgenic flies containing these constructs and crossed them into ∆5 and ∆19 genetic backgrounds. The resulting flies were tested for rescue of the trehalose response defect using the physiological and behavioral assays described above. The physiological response of the sugar cell to trehalose was rescued by the construct with wild-type copies of both Tre1 and Gr5a (T +G + in Fig. 3) in both the ∆19 and ∆5 backgrounds (Fig. 3c). Although rescue occurred in both the Tre1+Gr5a+ and the Tre1-Gr5a+ transgenic flies, it did not occur in the Tre1+Gr5atransgenic flies (Fig. 3b and c), indicating that rescue required a wild-type Gr5a but not a wild-type Tre1. None of the transgenes affected response to sucrose (Figs. 3b and c). We then tested the behavioral phenotype of the transgenic flies. We measured the preference for trehalose at a 10 –1.5 M (31.6 mM) concentration, which gave the maximal difference in PI between the parental and deletion strains (Fig. 2b). Rescue again occurred, and the pattern of rescue was consistent with the physiological data: it was dependent on Gr5a and not on Tre1 in both the ∆19 and ∆5 flies (Fig. 4).

DISCUSSION We have shown here that Gr5a, a member of a large family of candidate taste receptors, corresponds to the Tre locus, a genetic locus that affects response to the sugar trehalose. The most com1184

c

pelling evidence that Gr5a is responsible for the trehalose sensitivity phenotype is that deletion mutations affecting Tre function can be rescued by a transgenic construct containing a wild-type copy of Gr5a, but not by a construct containing a mutant copy of Gr5a. Rescue has been shown both by measurements of single-cell electrophysiology and by behavioral assays. The equivalence of Gr5a and the Tre locus is also consistent with mapping data indicating that all reported Tre deletions remove portions of Gr5a (Fig. 1 and ref. 21). The simplest interpretation of our results is that Gr5a encodes a taste receptor for trehalose. The absence of a functional Gr5a affects the physiological response of a taste neuron to trehalose, but not the response of the same neuron to anothnature neuroscience • volume 4 no 12 • december 2001


articles


Fig. 4. Gr5a rescues the behavioral defect of Tre mutants. Shown are the results of a two-choice test that measures preference between 10–1.5 M (31.6 mM) trehalose and 2 mM sucrose. Here 8 ≤ n ≤10, except that n = 5 for ∆19;T+G– and n = 4 for ∆5;T+G–. Error bars indicate 95% confidence intervals. ‘–’ denotes the ∆19 or ∆5 line in the absence of a transgene.

er disaccharide sugar, sucrose. Thus, the specificity of Gr5a function is consistent with that expected of a taste receptor, and inconsistent with that expected of a GPCR playing a general role in taste neuron development or function. Trehalose reception could also be mediated in part by additional receptors; however, the severely reduced trehalose response in flies lacking Gr5a and the rescue data (for example, Fig. 4) suggest that the response to trehalose is mediated primarily by Gr5a. A previous report identified the product of Tre1 as the trehalose receptor21. That report used a heat shock–inducible Tre1 cDNA construct to rescue the phenotype of a deletion mutant similar to ∆5. Although our results do not formally exclude a role for Tre1 in trehalose response, they do not support it. In considering the differing conclusions of these two studies, we note that although Tre1 mutants were shown to be abnormal in both a two-choice preference test and a proboscis extension test, rescue was described only in the two-choice preference test and only for a single pair of concentrations (2 mM sucrose and 80 mM trehalose)21. Moreover, although some limited data are reported to indicate a physiological phenotype for a Tre1 mutant, rescue of the peripheral physiological defect by the hs-Tre1 transgene has not been shown. We note that Tre1 is expressed in embryonic EST collections, suggesting that it is expressed in early Drosophila development. Tre1 is also expressed ubiquitously in adult tissue, according to a paper published while the current manuscript was under review27. This study also reports that polymorphisms in the sequence of Gr5a, but not Tre1, correlate with the trehalose phenotype. The sugar cells in all wild-type L- or M-class sensilla from which we have taken recordings respond to both sucrose and trehalose. In our recordings from Gr5a deletion mutants, we found no sensilla of these types whose sugar cells respond at wild-type levels to trehalose. Though not exhaustive, our data suggest that Gr5a is expressed in the sugar cells of all the L and M sensilla on the labellum. These two sensillar classes together constitute ∼30 of the ∼66 sensilla on the labellum; thus, Gr5a seems to be expressed in at least 30 neurons. Although this number has not been confirmed by in situ hybridization17, 30 is a larger number than is seen for the several members of the Gr family whose expression has been detected by in situ hybridization or reportergene expression17,18. For these receptors, the number of cells exhibiting expression ranges from 4 to 22 in the labellum17,18. Because response of all tested Gr5a mutant sugar cells was abnormal to trehalose but normal to sucrose, it seems likely that each of these cells expresses at least two receptor genes: Gr5a, which mediates response to trehalose, and another receptor that mediates response to sucrose. Colocalization of receptors within a single sugar cell might constrain the ability of the animal to discriminate among sugars through differential activation of distinct taste neurons. It remains possible, however, that discriminature neuroscience • volume 4 no 12 • december 2001

nation might be achieved by insulation of different signaling pathways, as probably happens within individual chemosensory neurons of Caenorhabditis elegans28. Although studies in mammals have implicated G protein–mediated signaling in the transduction of sweet taste, the mechanism in invertebrates is largely unknown29. There is evidence that sweet taste in larger flies is mediated at least in part through a cGMP second messenger30; however, there is also a report of a channel that is activated directly by sucrose without the mediation of second messengers or G proteins31. Our results support a role for G protein–mediated transduction of the disaccharide trehalose in Drosophila, as is found for sweet taste in mammals. In summary, the simplest interpretation of our data is that a member of the Gr gene family encodes a taste receptor required for response to the sugar trehalose, as indicated by both electrophysiological and behavioral analysis of mutant and transgenic flies. The association of a particular ligand with a particular Gr taste receptor now allows for a variety of studies, including detailed functional studies of the receptor and of the mechanism by which it transduces gustatory information. It will be interesting to determine whether the other genes in the Gr5a subfamily encode sweet taste receptors for other sugars.

METHODS Constructs and flies. A 10-kb genomic fragment that includes both the Tre1 and Gr5a coding regions was amplified by Expand Hi-Fi PCR (Boehringer Ingelheim, Germany) using genomic DNA from Canton S flies as template and primers 5´-GAGAGGTTGCGTTAAGCCAC and 5´-CGCAGCTGTGTAGCATAGTG. Amber mutations were engineered by introducing a linker (CTAGCTAGCTAG) in either the Not1 site in Tre1 or the HindIII site in Gr5a (see Fig. 3a). Transgenic flies were generated by standard means. At least two independent lines for each transgene were tested in both assays with similar results. Transgenic flies were heterozygous for the transgene in each case. Strains 496, ∆19 and ∆5 were obtained from K. Isono (Tohoku University, Japan). Electrophysiology. Action potentials were recorded23 from single labellar sensilla of male flies aged 5–15 days. Sugars were obtained from Aldrich (D-(+)-trehalose, >99%) or from Sigma (sucrose, >99.5%; St. Louis, Missouri). Sugars were dissolved in water, with tricholine citrate (Sigma) added as electrolyte at 0.03 M. Tricholine citrate does not elicit a response from the salt cells, nor does it appreciably affect the sugar cell responses26. Neural response was quantified as impulse rate by counting the number of impulses generated in the 500-ms period beginning 200 ms after onset of stimulation. Behavioral assay. The two-choice preference test was carried out as described elsewhere with minor modifications19,20. For each trial, 50 flies aged 3–5 d were fed on standard molasses culture medium for a day, then starved on water-saturated tissues for 21 h. Flies were fed on sugars in 60-well plates for 2 h at 25 °C in the dark and then killed by freezing at –20˚C. Indigo Carmine dye (Spectra Colors, Kearny, New Jersey) was present in the 2 mM sucrose standard at a concentration of 0.125 mg per ml. Acid red 106 (Sigma) was present at a concentration of 0.5 mg per ml in the wells containing variable sugar concentrations. Ten independent trials were carried out for each concentration of the sugar, and results were scored blind. Only those trials in which at least 20% of the flies had 1185


articles

fed were included for statistical analysis. Preference index (PI) values were calculated according to the formula (NR + 0.5NP)(NR + NB + NP)–1, where NR, NB and NP are the number of flies with red, blue and purple abdomens, respectively19,20. In preliminary experiments, males and females were tested separately and found not to differ; in all subsequent experiments males were used exclusively.


ACKNOWLEDGMENTS We thank K. Isono for providing strains and members of the Carlson laboratory for comments on the manuscript. The research was supported by an NRSA to A.D. and NIH grants and a McKnight Investigator Award to J.R.C.

RECEIVED 16 AUGUST; ACCEPTED 24 OCTOBER 2001 1. Nayak, S. & Singh, R. N. Sensilla on the tarsal segments and mouthparts of adult Drosophila melanogaster Meigen (Diptera: Drosophilidae). Int. J. Insect Morphol. & Embryol. 12, 273–291 (1983). 2. Stocker, R. The organization of the chemosensory system in Drosophila melanogaster: a review. Cell Tissue Res. 275, 3–26 (1994). 3. Singh, R. N. Neurobiology of the gustatory systems of Drosophila and some terrestrial insects. Microsc. Res. Tech. 39, 547–563 (1997). 4. Dethier, V. The Hungry Fly (Harvard Press, Cambridge, Massachusetts, 1976). 5. Stocker, R. F. & Schorderet, M. Cobalt filling of sensory projections from internal and external mouthparts in Drosophila. Cell Tissue Res. 216, 513–523 (1981). 6. Shanbag, S. R. & Singh, R. N. Functional implications of the projections of neurons from individual labellar sensillum of Drosophila melanogaster as revealed by neuronal marker horseradish peroxidase. Cell Tissue Res. 267, 273–282 (1992). 7. Chaudhari, N., Landin, A. & Roper, S. A metabotropic glutamate receptor variant functions as a taste receptor. Nature Neurosci. 3, 113–119 (2000). 8. Adler, E. et al. A novel family of mammalian taste receptors. Cell 100, 693–702 (2000). 9. Matsunami, H., Montmayeur, J. & Buck, L. A family of candidate taste receptors in human and mouse. Nature 404, 601–604 (2000). 10. Chandrashekar, J. et al. T2Rs function as bitter taste receptors. Cell 100, 703–711 (2000). 11. Nelson, G. et al. Mammalian sweet taste receptors. Cell 106, 381–390 (2001). 12. Montmayeur, J. P., Liberles, S. D., Matsunami, H. & Buck, L. B. A candidate taste receptor gene near a sweet taste locus. Nature Neurosci. 4, 492–498 (2001).

1186

13. Max, M. et al. Tas1r3, encoding a new candidate taste receptor, is allelic to the sweet responsiveness locus Sac. Nature Genetics 28, 58–63 (2001). 14. Sainz, E., Korley, J. N., Battey, J. F. & Sullivan, S. L. Identification of a novel member of the T1R family of putative taste receptors. J. Neurochemistry 77, 896–903 (2001). 15. Clyne, P., Warr, C. & Carlson, J. Candidate taste receptors in Drosophila. Science 287, 1830–1834 (2000). 16. Dambly-Chaudiere, C. et al. The paired box gene pox neuro: a determinant of poly-innervated sense organs in Drosophila. Cell 69, 159–172 (1992). 17. Scott, K. et al. A chemosensory gene family encoding candidate gustatory and olfactory receptors in Drosophila. Cell 104, 661–673 (2001). 18. Dunipace, L., Meister, S., McNealy, C. & Amrein, H. Spatially restricted expression of candidate taste receptors in the Drosophila gustatory system. Curr. Biol. 11, 822–835 (2001). 19. Tanimura, T., Isono, K., Takamura, T. & Shimada, I. Genetic dimorphism in the taste sensitivity to trehalose in Drosophila melanogaster. J. Comp. Physiol. 147, 433–437 (1982). 20. Tanimura, T., Isono, K. & Yamamoto, M. Taste sensitivity to trehalose and its alteration by gene dosage in Drosophila melanogaster. Genetics 119, 399–406 (1988). 21. Ishimoto, H., Matsumoto, A. & Tanimura, T. Molecular identification of a taste receptor gene for trehalose in Drosophila. Science 289, 116–119 (2000). 22. Ray, K., Hartenstein, V. & Rodrigues, V. Development of the taste bristles on the labellum of Drosophila melanogaster. Dev. Biol. 155, 26–37 (1993). 23. Hodgson, E., Lettvin, J. & Roeder, K. Physiology of a primary chemoreceptor unit. Science 122, 417–418 (1955). 24. Rodrigues, V. & Siddiqi, O. A gustatory mutant of Drosophila defective in pyranose receptors. Mol. Gen. Genet. 181, 406–408 (1981). 25. Tanimura, T. & Shimada, I. Multiple receptor proteins for sweet taste in Drosophila discriminated by papain treatment. J. Comp. Physiol. 141, 265–269 (1981). 26. Wieczorek, H. & Wolff, G. The labellar sugar receptor of Drosophila. J. Comp. Physiol. A 164, 825–834 (1989). 27. Ueno, K. et al. Trehalose sensitivity in Drosophila correlates with mutations in and expression of the gustatory receptor gene Gr5a. Curr. Biol. 11, 1451–1455 (2001). 28. L’Etoile, N. & Bargmann, C. Olfaction and odor discrimination are mediated by the C. elegans guanylyl cyclase ODR-1. Neuron 25, 575–586 (2000). 29. Glendinning, J., Chaudhari, N. & Kinnamon, S. Transduction and molecular biology. in Neurobiology of Taste and Smell II (eds. Finger, T. & Restrepo, D.) 315–351 (Academic Press, New York, 2000). 30. Amakawa, T., Ozaki, M. & Kawata, K. Effects of cyclic GMP on the sugar taste receptor cell of the fly Phormia regina. J. Insect Physiol. 36, 281–286 (1990). 31. Murikami, M. & Kijima, H. Transduction ion channels directly gated by sugars on the insect taste cell. J. Gen. Physiol. 115, 455–466 (2000).



articles


Synapsin dispersion and reclustering during synaptic activity Ping Chi1,2, Paul Greengard2 and Timothy A. Ryan1 1 Department of Biochemistry, Weill Medical College of Cornell University, 1300 York Avenue, New York, New York 10021, USA 2 Laboratory of Molecular and Cellular Neuroscience, The Rockefeller University, 1230 York Avenue, New York, New York 10021, USA

Correspondence and requests for materials should be addressed to T.A.R. ([email protected])

Published online: 29 October 2001, DOI: 10.1038/nn756 Presynaptic modulation of synaptic transmission provides an important basis for control of synaptic function. The synapsins, a family of highly conserved proteins associated with synaptic vesicles, have long been implicated in the regulation of neurotransmitter release. However, direct physiological measurements of the molecular mechanisms have been lacking. Here we show that in living hippocampal terminals, green fluorescent protein (GFP)-labeled synapsin Ia dissociates from synaptic vesicles, disperses into axons during action potential (AP) firing, and reclusters to synapses after the cessation of synaptic activity. Using various mutated forms of synapsin Ia that prevent phosphorylation at specific sites, we performed simultaneous FM 4-64 measurements of vesicle pool mobilization along with synapsin dispersion kinetics. These studies indicate that the rate of synapsin dispersion is controlled by phosphorylation, which in turn controls the kinetics of vesicle pool turnover. Thus synapsin acts as a phosphorylation-state-dependent regulator of synaptic vesicle mobilization, and hence, neurotransmitter release.

Synaptic vesicle availability and mobilization are crucial elements in the regulation of synaptic transmission and synaptic plasticity. Synapsins, a family of highly conserved neuronal phosphoproteins that are specifically associated with synaptic vesicles1, have been implicated in the regulation of neurotransmitter release by controlling the number of vesicles available for exocytosis. Synapsins exist in all organisms with a nervous system, and are encoded by three distinct genes, synapsin I, II and III, in most vertebrates2–6. Synapsins are the most abundant synaptic vesicle proteins, with synapsin I alone accounting for 6% of total vesicle protein1,7. They are present in nearly all presynaptic nerve terminals, but different neurons have a distinct repertoire of different synapsins1,6,8–10. The high abundance, the specific association with synaptic vesicles, the highly conserved features, as well as the widespread distribution at nerve terminals, all signify synapsins as important and evolutionarily conserved regulatory proteins in synaptic transmission. Biochemical studies have revealed multiple regulatory roles of synapsins. In vitro studies have shown that synapsins can interact with lipid and protein components of synaptic vesicles, as well as various cytoskeletal proteins, such as actin, spectrin and microtubules, in a phosphorylation-dependent manner1,11–24. These studies suggest that synapsins would move dynamically in response to physiological stimuli, and have led to the following hypothesis: binding of synapsins to synaptic vesicles prevents neurotransmitter release, and during synaptic activity, synapsins are phosphorylated, dissociate from synaptic vesicles and allow vesicles to mobilize and fuse with the plasma membrane1,20,23,24. Given the abundance and conservation of the synapsin family of proteins, it is interesting that mice with genetic perturbations to specifically remove synapsin genes are viable and have nature neuroscience • volume 4 no 12 • december 2001

relatively limited behavioral phenotypes25–27. Although there are clear differences in synaptic physiology and presynaptic ultrastructure in the mutant animals, the lack of lethality and the modest effect on a number of learning protocols in the mutants have cast some doubt on the original hypothesis of synapsin function. However, as the primary role of synapsins that has been proposed is a regulatory one, it would be difficult to predict how loss of a layer of regulation would be manifested. To more fully determine the mechanisms by which synapsins regulate synaptic function, we GFP-labeled synapsin Ia to examine the dynamic behavior of synapsin in response to action potential (AP) firing. We combined this approach with kinetic measurements of vesicle-pool turnover monitored by FM 4-64 to examine various mutants at the calcium-dependent phosphorylation sites of synapsin Ia in both wild-type and synapsin double knockout backgrounds. Here we demonstrate an activity-dependent dissociation of synapsin from synaptic vesicles and subsequent dispersion into the axon. The rate of the dispersion is regulated by calcium-dependent phosphorylation, which in turn controls the efficiency of vesicle pool turnover. These data thus provide a direct demonstration of the mechanism by which synapsins regulate the availability of synaptic vesicles for neurotransmitter release.

RESULTS Synapsin dissociates from vesicles during activity A variety of in vitro biochemical studies imply that synapsins dynamically associate with and dissociate from synaptic vesicles in response to synaptic activity1,20,21,24,28–30. To study synapsin dynamics in vivo, we performed immunolocalization of synapsin and an integral synaptic vesicle protein, synaptophysin. Double immunostaining of synapsin and synaptophysin in hippocampal cell cultures at rest showed a typical 1187


Rest

Anti-synapsin Anti-synaptophysin

Fig. 1. Synapsin disperses from synaptic vesicles during activity. (a, d, g) Co-immunolocalization of synaptophysin (red; a) and synapsin (green; d) in hippocampal cell cultures fixed at rest show punctate staining for both markers (g). (b, e, h) In a parallel culture, fixed immediately after a train of a 900-AP stimulation at 10 Hz, synapsin staining (e, h) is much more diffuse than synaptophysin staining (b, h) indicating that synapsins dissociate from synaptic vesicles and redistribute into axons (arrows) in response to electrical stimulation. (c, f, i) In a different specimen fixed 10 min after stimulation, immunostaining of synapsin returns to a punctate pattern colocalized with synaptophysin (c, i), indicating a dynamic relocalization of synapsins from axons to synapses post-stimulation (f, i). Scale bar, 5 µm.

Stimulated

Recovery

a

b

c

d

e

f

g

h

i

Overlay

colocalized punctate staining pattern at the presynaptic nerve terminals (Fig. 1a, d and g). In parallel cultures fixed immediately following a train of 900 AP, the immunostaining of synaptophysin remained largely punctate, indicating that synaptic vesicles remained at nerve terminals (Fig. 1b and h). However, the immunostaining of synapsins became much more diffuse during stimulation (Fig. 1e and h), demonstrating a physical separation of synapsins from synaptophysin and suggesting that a substantial fraction of synapsins have dissociated from synaptic vesicles and redistributed into axons. The immunolocalization of synapsins returned to a punctate pattern in cultures fixed 10 minutes after stimulation (Fig. 1f and i), and it colocalized well with the immunostaining of synaptophysin (Fig. 1c and i). These observations demonstrate a dissociation of synapsin from synaptic vesicles and a dynamic movement in response to synaptic activity at nerve terminals.

a

c

Stimulus

Rest

∆F/F0 (GFP-Syn Ia)

2500

0

0.2

∆F/F0 (GFP-Syn Ia)

b

10 Hz 0.0

τ = 12.9 s

–0.2

Bouton –0.4

10 Hz

2.0 1.5

Axon 1.0

τ = 13.0 s

0.5 0.0 0

100

200

Time (s)

0

e 10 Hz

0.2 0.0

Ftotal Fbouton

–0.2 –0.4 –0.6 0

40

Time (s)

1188

Real-time measurements of synapsin dispersion To measure the real-time dynamic movement of synapsin Ia at living hippocampal nerve terminals, we transfected neuronal cell cultures with a plasmid encoding GFP-labeled synapsin Ia. To test whether GFP-synapsin Ia behaves similarly to endogenous synapsins, we examined the localization of GFP-labeled synapsin Ia before, during and after synaptic activity. GFP-synapsin Ia, when expressed transiently in cultured rat hippocampal neurons, was targeted to nerve terminals (Fig. 2a), and appeared to have the same localization as native synapsins at rest (Fig. 1d). During a train of 900 AP stimulation, GFP-synapsin Ia fluorescence decreased at the nerve terminal and increased in the inter-bouton (axonal) region in a manner similar to that of endogenous synapsins (Fig. 2a and b). Ten minutes after stimulation, GFP-synapsin Ia reclustered within the nerve terminals, and appeared virtually indistin-

d

1100

∆ F/F0(GFP-Syn Ia)


articles

80

120

300

Fig. 2. Kinetics of GFP-synapsin Ia dispersion and recovery. (a) GFP-synapsin Ia fluorescence at synapses (green arrows) decreases upon stimulation. (b) GFP-synapsin Ia fluorescence in axons increases upon stimulation (red arrow). Same image as in (a) with different color scale to emphasize the fluorescence in the axon. The color scales show fluorescence intensity in arbitrary fluorescence units. Scale bar, 2 µm. (c, d) Time courses of fluorescence intensity at synapses (c) and in axons (d), during and following a train of 900 AP at 10 Hz (black bar). The decay of fluorescence intensity during stimulation, averaged over 43 boutons expressing GFP-synapsin Ia, is fit by a single exponential with τ = 12.9 s (c). During the same period of AP stimulation, GFP-synapsin Ia redistributes into axons with a similar rate constant, τ = 13.0 s (d). GFP-synapsin Ia reclusters within the terminal (c), and decreases in the axonal region (d) in approximately 4 min. (e) Time-lapse measurement of the total integrated fluorescence in the field of view and the fluorescence of GFP-synapsin Ia at synapses during a train of 900 AP field stimulation at 10 Hz (black bar). The total fluorescence remains relatively constant during stimulation, whereas the fluorescence intensity of GFP-synapsin Ia at synapses (n = 43) decreases. nature neuroscience • volume 4 no 12 • december 2001


articles

∆F/F0 (GFP-VAMPaxon)

FM 4-64 fluorescence

τGFP-Syn Ia (s)

FM 4-64 fluorescence


∆F/F0 (GFP-VAMPaxon)

guishable from GFP-synapsin Ia in a prestimulated culture (data not shown). GFP, when expressed alone, uniformly labeled the cell body, axons and dendrites, and showed no activity-dependent concentration change (data not shown). Quantitative analyses of the dynamic movement indicated that the concentration of GFP-synapsin Ia decreased by approximately 45% and reached a steady state at synaptic boutons during AP firing (Fig. 2a and c), coincident with a steep rise of GFP-synapsin Ia concentration in axonal regions (Fig. 2b and d); the total fluorescence in the whole imaging field remained relatively constant (Fig. 2e). Following stimulation, GFPsynapsin Ia reclustered within nerve terminals within approximately four minutes (Fig. 2c), with a gradual decrease of synapsin concentration in axons over the same time course (Fig. 2d). Stimulation in the absence of external calcium failed to disperse GFP-synapsin Ia (data not shown). These results indicate that this imaging approach provides a high-fidelity real-time measurement of the dynamic movement of synapsins in response to synaptic activity.

∆F/F0 (GFP-Syn Iaaxon)


FM 4-64 destaining

Fig. 3 Comparison of the dynamics of GFP-synapsin Ia disa b 10 Hz persion and recycling vesicle turnover. (a) Colocalization 1.00 of GFP-synapsin Ia-expressing boutons (green) with FM 464- loaded terminals (red), giving rise to yellow puncta, Non-transfected τ = 20.7 s 0.75 indicating that GFP-synapsin Ia-positive terminals functionGFP-Syn Ia + ally recycle vesicles. In addition, in the same field, there are τ = 21.0 s 0.50 boutons that do not express GFP-synapsin Ia, labeled by FM 4-64 only (red), which are used as internal controls in 0.25 FM 4-64 destaining experiments. Scale bar, 5 µm. 0.00 (b) Overexpression of GFP-synapsin Ia (wild type) does not affect the kinetics of vesicle pool turnover as assayed 30 120 60 0 90 Time (s) using FM 4-64 destaining. Destaining kinetics of FM 4-64d c labeled synaptic vesicle pools during a train of 900 action 50 1.2 10 Hz potentials at 10 Hz (black bar) in GFP-synapsin Ia-positive Acid (pH 4.0) (n = 40) and GFP-synapsin Ia-negative (n = 50) boutons are 0.8 40 shown. Average time constants of FM 4-64 destaining in 0.4 Acid (pH 4.0) 30 GFP-synapsin Ia-positive and negative boutons were τ = 21.0 s and τ = 20.7 s, respectively. Similar results were 0.0 20 obtained in five of five experiments. (c) Frequency depen5 Hz 10 Hz –0.4 dence of GFP-synapsin Ia dispersion and FM 4-64 destain20 Hz 10 ing measured in GFP-synapsin Ia expressing nerve –0.8 terminals. GFP-synapsin Ia dispersion kinetics and FM 4-64 0 0 200 300 400 500 0 20 40 60 80 100 destaining kinetics at different frequencies are well correTime (s) τ FM4-64 (s) lated over a 5–20 Hz stimulus frequency range (r = 0.99). Time constants of GFP-synapsin Ia dispersion and FM 4-64 10 Hz f e turnover at different frequencies were averaged over 1.2 1.2 10 Hz 1.2 1.2 13–33 boutons pooled from 3 experiments. GFP-synapsin Ia disperses with statistically faster kinetics than FM 4-64 0.9 0.9 0.9 0.9 destaining at all frequencies tested (p < 0.002). (d) GFPFM 4-64 VAMP dispersion into the axon during AP firing is 0.6 0.6 0.6 0.6 FM 4-64 τ = 26.9 s restricted to the plasma membrane. Before stimulation, a τ = 25.5 s GFP-SYNIAaxon GFP-VAMP axon = 15.9 s τ large fraction of GFP-VAMP in the axon is on the plasma 0.3 0.3 0.3 τ = 25.4 s 0.3 membrane, and is quenchable by brief application of sur0.0 0.0 face impermeant acid (pH 4.0). During stimulation (1,200 0.0 0.0 action potentials at 10 Hz), the concentration of GFP–0.3 –0.3 VAMP increases at the axon. This additional stimulation–30 0 30 60 90 120 0 30 60 90 120 induced fluorescence remained quenchable with Time (s) Time (s) impermeant acid, indicating that the dispersion consisted entirely of surface GFP-VAMP and not vesicles simply dispersing into the axon. Axon regions analyzed, n = 51. (e) Comparison of GFP-synapsin Ia fluorescence (right y-axis) in the axon regions (n = 39) and simultaneous measurement of FM 4-64 turnover (left y-axis) at the GFP-synapsin Ia expressing nerve terminals (n = 40) during a train of 900 AP stimulation (black bar). The kinetics of the rise of GFP-synapsin Ia fluorescence in the axon (τ = 15.9 s) is much faster (p < 0.001) than that of FM 4-64 turnover (τ = 26.9 s), suggesting a physical separation of GFP-synapsin Ia from synaptic vesicles during activity. (f) Comparison of GFP-VAMP fluorescence (right y-axis) along the axon regions (n = 51) and simultaneous measurement of FM 4-64 turnover (left y-axis) at the GFP-VAMPpositive nerve terminals (n = 14) during a train of 900 AP stimulation (black bar). The kinetics of the rise of GFP-VAMP fluorescence along the axon (τ = 25.4 s) is very similar to that of FM 4-64 turnover (τ = 25.5 s). Similar results were obtained in three of three experiments.

Synapsin Ia disperses faster than vesicle pool turnover To examine the physiological relevance of the dynamic movement of synapsins at nerve terminals, we used FM 4-64 (ref. 31), a red-shifted variant of FM 1-43, to simultaneously monitor synapsin dispersion and synaptic vesicle turnover in GFPsynapsin Ia-expressing synapses. Loading of FM 4-64 by AP stimulation resulted in labeling all GFP-synapsin Ia-expressing synapses (Fig. 3a, yellow puncta), indicating that these synapses functionally recycle synaptic vesicles. The extent of FM 4-64 loading was similar in synapses either expressing or not expressing GFP-synapsin Ia (data not shown). Subsequent AP stimulation of FM 4-64-loaded nerve terminals resulted in dye destaining as synaptic vesicles fused with the plasma membrane and released the dye. GFP-synapsin Ia-expressing and non-expressing synapses in the same culture exhibited similar rates of FM 4-64 destaining (Fig. 3b), indicating that transient overexpression of GFP-synapsin Ia (wild-type) did not affect synaptic vesicle turnover at synapses. The kinetics of GFP-synapsin Ia dispersion at 10 Hz stimulation (τ = 12.9 s, Fig. 2c) was significantly faster than synaptic vesicle 1189


0.2

b

10 Hz

10

τ = 20.0 s

–0.2 –0.4 –0.6

Number of events

1.0 0.0

Non-transfected boutons < τ > = 23.1 s

c

10 Hz

FM 4-64 destaining


a

∆F/F0 (GFP-Syn Ia, S2/3-A)

articles

Non-transfected τ = 24.0 s GFP-Syn Ia-S2/3A + τ = 31.7 s

5 0

5 0

0

100

200 300 Time (s)

400

0

20 40 Time (s)

GFP-Syn Ia-S2/3A + <τ > = 31.4 s

10

60

80

0

10

20

30

τ FM4-64 (s)

40

50

Fig. 4. Synapsin Ia is a phosphorylation-dependent negative regulator of vesicle pool turnover. (a) Mutations of serine to alanine at CaM kinase II sites 2 and 3 (S2/3A) slow the rate of GFP-synapsin Ia dispersion when expressed in rat hippocampal cell culture. Time course of fluorescence intensity at synapses, averaged over 20 boutons expressing GFP-synapsin Ia-S2/3A, during (dark bar) and following a train of 900 AP at 10 Hz. The dispersion kinetics of the GFP-synapsin Ia mutant during stimulation is much slower (τ = 20.0s) than wild type (Fig. 2c, τ = 12.9 s; p < 0.001). The time course of reclustering is similar to wild-type GFP-synapsin Ia (Fig. 2c). (b) Synaptic vesicle pool turnover monitored by FM 4-64 destaining is significantly slowed by the mutant form of GFP-synapsin Ia as compared to non-transfected boutons. A typical example is shown in this panel on a semi-log plot. The average time constant for FM 4-64 destaining during 1,200 AP (dark bar) in GFP-synapsin Ia-S2/3A-positive boutons (τ = 31.7 s, n = 21) is significantly larger than that in GFP-synapsin Ia-negative boutons (τ = 24.0 s, n = 35; p < 0.001). (c) Frequency distribution of FM 4-64 destaining time constants (τFM 4-64) measured at individual GFP-synapsin Ia-S2/3A expressing and non-expressing boutons.

turnover assayed by FM 4-64 (τ = 21.0 s, Fig. 3b, p < 0.001). Stimulation at higher (20 Hz) or lower (5 Hz) frequency also revealed that kinetics of GFP-synapsin Ia dispersion were significantly faster than kinetics of FM 4-64 turnover (Fig. 3c, p < 0.002), and that the kinetics of synapsin Ia dispersion and vesicle pool turnover were well correlated in this frequency range (Fig. 3c). The difference in time scale between vesicle pool turnover and GFP-synapsin Ia dispersion provides additional evidence for a dynamic dissociation of synapsin Ia from synaptic vesicles during synaptic activity. To further test the physical separation of synapsin from synaptic vesicles during stimulation, we specifically compared the increase of GFP-synapsin Ia fluorescence in axons with simultaneously monitored FM 4-64-labeled vesicle turnover. The kinetics of GFP-synapsin Ia appearing in axons (τ = 15.9 s) was much faster than that of FM 4-64 turnover (Fig. 3e, τ = 26.9 s, p < 0.001), indicating a physical separation of GFPsynapsin Ia from synaptic vesicles that precedes the fusion of synaptic vesicles with plasma membrane. We performed the same measurements for an integral-membrane protein of synaptic vesicles, VAMP, labeled with GFP on the lumenal domain along with FM 4-64 detaining kinetics. We previously showed that during stimulation, a small portion of the pool of VAMP disperses onto the axonal surface following stimulation32. Similarly, the GFP-VAMP signal increased to a small degree along the axonal region during stimulation. The kinetics of the appearance of GFP-VAMP in the axon, however, matched that of vesicle pool turnover very well (Fig. 3f), and was much slower than GFP-synapsin Ia dispersion (p < 0.001). Furthermore, the VAMP axonal signals measured in these experiments during stimulation were fully quenched by transient application of an impermeant acidic buffer (pH 4.0; Fig. 3d), in agreement with previous studies32. As similar applications of impermeant acidic buffer did not quench intracellular GFP (such as GFP-synapsin Ia fluorescence, data not shown), we conclude that the elevation in VAMP-GFP fluorescence along the axon during stimulation is confined to the plasma membrane and does not correspond to vesicles breaking away from presynaptic clusters. These findings, along with the differential immunolocalization of synaptophysin and synapsins in stimulated nerve terminals (Fig. 1b, e and h), 1190

strongly indicate that the measured synapsin Ia dispersion corresponds to dissociation of synapsin from synaptic vesicles and redistribution into the axon during activity in a step that precedes fusion of vesicles with the plasma membrane. Phosphorylation regulates synapsin movement Previous studies indicate that phosphorylation at site 1, a calcium-calmodulin-dependent kinase I/IV (CaM kinase I/IV) and protein kinase A (PKA) site, and at sites 2 and 3, CaM kinase II sites33, leads to dissociation of synapsin Ia from synaptic vesicles14,21,29. Therefore, we specifically investigated the physiological functions of CaM kinase phosphorylation sites in synapsin Ia by directly mutating each of them from serine to alanine, an amino acid of similar size that cannot be phosphorylated. When both CaM kinase II sites were mutated to alanine (S2/3A), the kinetics of dissociation and dispersion of GFP-synapsin Ia-S2/3A from boutons during AP stimulation was slowed significantly (p < 0.001), with a τ of 20.0 s (Fig. 4a) as compared to a τ of 12.9 s observed with wild-type GFPsynapsin Ia (Fig. 2c). The kinetics of synaptic vesicle turnover as monitored by FM 4-64 was also significantly slowed in boutons expressing GFP-synapsin Ia-S2/3A (τ = 31.7 s) as compared to non-expressing boutons (τ = 24.0 s; Fig. 4b and c, p < 0.001). Similar observations were obtained from 7 other experiments with S2/3A; average τ dispersion was 20.9 s and there was an approximately 22% slowing of FM 4-64 turnover (Fig. 5a and b, p < 0.0001). These results are consistent with studies that show decreased and increased synaptic transmission in squid giant synapse injected, respectively, with dephosphorylated synapsin Ia and CaM K II34,35. The time course of reclustering of GFPsynapsin Ia-S2/3A to nerve terminals was not significantly different from wild-type when averaged over eight experiments (data not shown). To determine whether the different CaM kinase phosphorylation sites have differential effects on synaptic transmission, we measured the rate of synapsin dispersion and FM 4-64 destaining in rat hippocampal cultures transfected with various permutations of serine to alanine at the CaM kinase sites: S1A, S2A, S3A, S2/3A, and with all of sites 1, 2 and 3 mutated, S1/2/3A. All of these GFP-synapsin mutants except for S2A showed significantnature neuroscience • volume 4 no 12 • december 2001


ly decreased rates of dispersion and correspondingly decreased rates of FM 4-64 destaining compared to the wild type (Fig. 5a and b, p < 0.03 for each comparison). Furthermore, there was a strong correlation between how fast GFP-synapsin Ia and mutants dissociated from vesicles during AP firing and how fast the whole recycling vesicle pool could be mobilized to fuse with the plasma membrane; that is, as dispersion kinetics of the GFP-synapsin Ia mutant were slower, the effect was bigger on slowing synaptic vesicle turnover as monitored by FM 4-64 (Fig. 5c). These results strongly suggest that synapsin Ia is a negative regulator of neurotransmitter release, and that its function is controlled by calcium-dependent phosphorylation. All three identified synapsins, I, II and III, can form homodimers36. Heterodimerization of synapsin I and II and of synapsin II and III (but not of synapsin I and III) have also been demonstrated by glutathione S-transferase (GST)-pull down or coimmunoprecipitation assays36. It seemed possible that some of the effects of GFP-synapsin Ia mutants observed in our transfection experiments in rat neuronal cultures might have been confounded by the presence of endogenous synapsins. To further determine the physiological function of synapsin Ia, we examined the CaM kinase phosphorylation site mutants in hippocampal cultures from synapsin I/II–/– mice in which interactions between transfected mutants and wild-type endogenous synapsins are minimal. Wild-type GFP-synapsin Ia dispersed in synapsin I/II (Fig. 6a) and wild-type mouse (data not shown) neuronal cultures with kinetics similar to those in rat cultures (Fig. 5a). The kinetics of vesicle pool turnover monitored by FM 1-43 was approximately 10% faster in synapsin I/II –/– synapses (unpublished data), consistent with the hypothesis that synapsin acts as a negative regulator of pool turnover. GFPsynapsin Ia-S2/3A and S1/2/3A mutants showed significantly slowed rates of dispersion and correspondingly decreased rates of FM 4-64 turnover in synapsin I/II–/– background (Fig. 6a and b, p < 0.03 for each comparison). The differential effect of various mutant synapsins on pool turnover kinetics does not seem to arise from differential expression levels or targeting efficiency of these mutant constructs, as the average fluoresnature neuroscience • volume 4 no 12 • december 2001

b 30 25 20 15 10 5 0 wt S1

A 2A 3A 2/3A 1/2/3 S S S S

A

Percent slowing FM 4-64 destaining

c Percent slowing FM 4-64 destaining

Fig. 5. Dispersion kinetics of GFPWild-type (rat) synapsin Ia mutants regulate the effi- a 25 ciency of vesicle pool turnover in wild-type rat neurons. (a) Effect of 20 different CaM kinase site mutants on the dispersion kinetics of GFP15 synapsin Ia. Number of experiments for each specific mutation is as fol10 lows: wild type (wt), 6; site 1 (S1A), 6; site 2 (S2A), 3; site 3 (S3A), 5 3; sites 2 and 3 (S2/3A), 8; sites 1, 2 and 3 (S1/2/3A), 12. Except for S2A, 0 A all GFP-synapsin Ia mutants show A A A 3A 2/3 wt S1 S2 S3 S2/ S1/ statistically significant difference from wild-type GFP-synapsin Ia (p < 0.02). (b) Effect of various mutants on FM 4-64 destaining. The effect is calculated as the percentage increase of the time constant for FM 4-64 destaining of GFP-synapsin Ia-mutant-positive boutons compared to nontransfected boutons in the same experiment to reduce the effect of variability in different neuronal cultures. Typically, data from 15–30 transfected boutons and 40–50 non-transfected boutons were obtained in each experiment. Except for S2A, all other GFP-synapsin Ia mutants show statistically significant difference from wild-type GFP-synapsin Ia (p < 0.03). (c) Positive correlation between the dispersion kinetics and the effect on FM 4-64 destaining for various GFP synapsin Ia mutations.

τdispersionGFP-Syn Ia mutants (s)


articles

30

20

wt S1A S2A S3A S2/3A S1/2/3A

10

0 15

20

25

τdispersion GFP-Syn Ia mutants (s)

cence intensity of GFP-synapsin mutants at individual boutons was similar (data not shown). The effect of either S1/2/3A or S2/3A mutations in GFPsynapsin Ia (Fig. 6a and b) was more pronounced when expressed in synapsin I/II–/– neuronal cultures than in wildtype mouse (data not shown) or rat cultures (Fig. 5a and b, p < 0.0001 for each comparison). This presumably reflects an interaction between GFP-synapsin Ia mutant and either endogenous wild-type synapsin I or endogenous wild-type synapsin II, when expressed in wild-type background. In contrast, the S1A mutation had a similar effect on FM 4-64 turnover in synapsin I/II –/– (Fig. 6a and b) and wild-type (Fig. 5a and b) neuronal cultures. This observation suggests that site 1 directly modulates binding affinities of synapsin Ia to synaptic vesicles, but not through interactions with endogenous synapsins, consistent with the observation that site 1 phosphorylation regulates the phospholipid binding of synapsins to synaptic vesicles21. Importantly, in the rat as well as in the synapsin I/II–/− mouse, there was a strong correlation between how fast GFP-synapsin Ia and its mutants dispersed and how fast FM 4-64 was released (Figs. 5c and 6c). All of these observations not only point to synapsin Ia as a negative regulator of synaptic vesicle turnover, but also demonstrate two different levels of regulation exerted by different CaM kinase phosphorylation sites. The CaM kinase II sites seem to modulate the binding affinity of synapsin Ia to synaptic vesicles through interaction among synapsins, whereas the CaM kinase I/IV site (also a PKA site) seems to regulate the binding of synapsin Ia to vesicles directly.

DISCUSSION The studies presented here have allowed us to visualize the dynamic movement of a regulator of neurotransmission, synapsin Ia, during synaptic activity in living nerve terminals. This dynamic movement was regulated by the phosphorylation state of various CaM kinase sites of synapsin Ia, revealing the importance of the activities of CaM kinase I/IV and II on synapsin dynamics during well-defined periods of synaptic activity. Further, the correlation between the rate of dispersion of synapsin Ia mutants and the kinetics of vesicle pool turnover strongly suggest that synapsin Ia regulates neurotransmitter release via its dynamic interaction with synaptic vesicles. Our observation of activity-dependent synapsin dispersion using immunohistochemistry corroborates a previous report37. However, in that study, it was concluded that dispersed synapsin remained associated with vesicle membranes. Here, using realtime measurements and detailed comparisons of synapsin dispersion, vesicle pool turnover and the redistribution of an integral membrane protein of synaptic vesicles (VAMP), we 1191


articles

Fig. 6. Dispersion kinetics of GFP-synapsin Ia mutants regulate the efficiency of vesicle pool turnover in synapsin I/II–/– neurons. 60 40 60 (a) Dispersion kinetics of GFP-synapsin Ia S2/3A S1/2/3A mutated at CaM kinase phosphorylation 45 40 30 sites. The number of experiments for each mutation is as follows: wt, 2; S1A, 3; S2/3A, 30 20 20 4; S1/2/3A, 7. Abbreviations are as in Fig. 5. S1A All GFP-synapsin Ia mutants tested showed wt 15 S1A significantly slower dispersion kinetics than 10 0 wt S2/3A 0 wild-type GFP-synapsin Ia (p < 0.001). S1/2/3A (b) Effect of different mutations on FM 4-64 0 –20 –15 turnover. All mutants show statistically sig20 30 40 S1A S2/3A S1/2/3A wt τdispersion GFP-Syn Ia mutants (s) nificant b a c difference from wild-type GFP-synapsin Ia (p < 0.03). (c) Positive correlation between the dispersion kinetics demonstrated that a large fraction of synapsin must dissociate and the effect on FM 4-64 destaining kinetics for various GFP synapsin Ia from vesicles during activity. mutations when expressed in the absence of endogenous synapsin I and II.

mice


–/–


τdispersion GFP-Syn Ia mutants (s)


Synapsin I/II

The phenotypes of synapsin I-, II- and I/II-deficient mice characterized by postsynaptic electrophysiological measurements25–27 are considered mild. Further detailed analyses of presynaptic vesicle recycling of synapsin I-deficient mice by FM 1-43 only reveal reduced vesicle pool turnover with brief trains of AP stimulation and reduced total recycling pool size, whereas other parameters including endocytosis and repriming are unaltered 38. However, it is possible that the phenotypic changes in these mice are alleviated by functional redundancy of synapsin III4 and by long-term compensatory changes in the levels of other nerve terminal proteins as demonstrated in synapsin I/II deficient mice26. Our studies demonstrate a link between dissociation of synapsin Ia and the ability of vesicles to mobilize and fuse with the plasma membrane during AP firing. However, several phenomena remain to be explained. First, it is not known if the dispersion and reclustering of synapsins following phosphorylation are driven by free diffusion or by an active transport process. Second, we do not understand why there is still dispersion in a fully mutated S1/2/3A form of synapsin when expressed in a synapsin I/II –/– background. This result suggests that in addition to the CaM kinase sites, an unknown activity-dependent modification of synapsin modulates its dissociation from vesicles. This could potentially be achieved through other phosphorylation sites, or through calciumdependent binding of ATP 39,40. Third, the finding that the concentration of GFP-synapsin Ia decreases only by approximately 45% before reaching steady state during AP firing at nerve terminals (Fig. 2c) suggests that a portion of synapsins do not dissociate from vesicles, as has been suggested by in vitro experiments21. Analysis of fluorescence intensity distribution of a freely diffusing cytoplasmic volume marker (cytoplasmically expressed GFP) suggests that if all GFP-synapsin dissociated, the extent of dispersion would be greater than approximately 75–80% (data not shown). What prevents the remaining GFP-synapsin Ia from dispersing? Whether this is correlated with a non-recycling pool of vesicles or whether there is something else in addition to calcium-dependent phosphorylation that is required for synapsin dispersion remains to be determined. Although previous knockout analysis clearly demonstrated that synapsin I and II alone cannot account for the clustering of vesicles at active zones, the studies presented here indicate that synapsins control the availability of vesicles through their ability to dissociate from synaptic vesicles in a phosphorylation-dependent manner. Thus, synapsins provide a phosphorylation-mediated layer of control over presynaptic 1192

function by their ability to control vesicle availability at the nerve terminal during repetitive AP firing.

METHODS cDNA subcloning and site-directed mutagenesis. EGFP-synapsin Ia fusion protein was generated by subcloning the rat synapsin Ia cDNA into the pEGFP-C1 vector (Clontech, Palo Alto, California). Site-directed mutagenesis was done using the Stratagene (La Jolla, California) QuikChange site-directed mutagenesis kit. Hippocampal cell culture and transfection. Synapsin I/II–/– mice were generated by homologous recombination41. Hippocampal CA3–CA1 regions were dissected from 3–4-day-old Sprague–Dawley rats, and 0–1-day wild type and synapsin I/II–/– mice. The dissected hippocampal regions were then dissociated, plated and cultured as described42. Transfections of GFP-synapsin Ia and its phosphorylation-site mutants as well as GFP-VAMP were done using calcium phosphate precipitation as described43, and performed on 7–8-day-old neuronal cultures. All animal experiments and use were approved by the Institutional Animal Care and Use Committee of the Weill Medical College of Cornell University. Immunocytochemistry. Individual dishes of 2–3-week-old rat hippocampal cultures were processed for immunofluorescence of both synaptophysin and synapsins after fixation in 4% paraformaldehyde (Electron Microscopy Science, Washington, Pennsylvania), 1× PBS, and 0.041% sucrose for 15 min. Dishes were separated into groups and subjected to 1 of 3 conditions immediately before fixation: control, 900 AP field stimulation, or 900 AP field stimulation followed by 10-min rest period. Cells were then permeabilized in the same fixative plus 0.25% Triton for 15 min, blocked in 10% BSA for an hour at 37°C and incubated overnight with both monoclonal anti-synaptophysin antibody (clone SVP-38, Sigma, St. Louis, Missouri) and affinity-purified anti-synapsin Ia, IIa and IIIa10 polyclonal antibody (G-304). Cells were then incubated with an Alexa 488-labeled anti-mouse IgG and an Alexa 546 anti-rabbit IgG secondary antibody (Molecular Probes, Eugene, Oregon) for an hour, and mounted for observation. Image analysis was performed using a blinded procedure without knowledge of the experimental treatment. Experimental conditions. Experiments were performed on 5–10-day post-transfected cultures. Coverslips were mounted in a perfusion chamber equipped with field stimulation electrodes on the stage of a custom-built laser-scanning confocal microscope as described42. Unless otherwise noted, cells were perfused at room temperature (∼24°C) in a saline solution consisting of 119 mM NaCl, 2.5 mM KCl, 2 mM CaCl2, 2 mM MgCl2, 5 mM HEPES (pH 7.4), 30 mM glucose, 10 µM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) and 50 µM D, L -2amino-5-phosphonovaleric acid (APV). Acidic-solution with final pH nature neuroscience • volume 4 no 12 • december 2001


articles


of 4.0 was prepared by replacing HEPES in the standard saline with MES (pK a 6.1), all other components in the saline remaining unchanged. Synaptic vesicle pools were labeled by field-stimulating cultures for 30 s at 20 Hz in the presence of FM 4-64 in normal saline. An additional 60 s of dye exposure was allowed to ensure complete labeling of all recycling vesicles. The cultures were subsequently rinsed in dye-free solution for 10 min before dye destaining. Unless otherwise stated, all reagents were obtained from Sigma. Optical measurements, microscopy and analysis. Laser-scanning fluorescence images were acquired as described32. Quantitative measurements of fluorescence intensity at individual boutons and neighboring axonal regions were obtained by averaging a 4 × 4 area of pixel intensities. FM 4-64 destaining data were normalized to the total loss of fluorescence during the train of AP, determined by subtracting the average of the final three time points from that of the first five time points before stimulation for each individual bouton. Time constants for FM 4-64 destaining were obtained by fitting the destaining curves to single exponential decays. The fluorescence change of GFP-synapsin Ia and its mutants either at synaptic boutons or in axons were normalized to the starting fluorescence (F0) at individual selected region. The time constants for dispersion of GFP-synapsin Ia wild type and mutants were determined similarly by fitting the dispersion curve to single exponential decay, or in the case of minimal or very slow dispersion by directly taking the time point at 63% of the normalized fluorescence decrease.

ACKNOWLEDGEMENTS We would like to thank R. Scheller for providing the GFP-VAMP construct, W. Yan for technical assistance and members of the Ryan and Greengard labs for discussions. This work was supported by grants from the NIH to T.A.R. (NS24692 and GM61925-01) and P.G. (MH39327). T.A.R. is an Alfred P. Sloan Research Fellow.

RECIEVED 3 OCTOBER; ACCEPTED 12 OCTOBER 2001

15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30.

1. Greengard, P., Valtorta, F., Czernik, A. J. & Benfenati, F. Synaptic vesicle phosphoproteins and regulation of synaptic function. Science 259, 780–785 (1993). 2. Hilfiker, S. et al. Two sites of action for synapsin domain E in regulating neurotransmitter release. Nat. Neurosci. 1, 29–35 (1998). 3. Hosaka, M. & Sudhof, T. C. Synapsin III, a novel synapsin with an unusual regulation by Ca2+. J. Biol. Chem. 273, 13371–13374 (1998). 4. Kao, H. T. et al. A third member of the synapsin gene family. Proc. Natl. Acad. Sci. USA 95, 4667–4672 (1998). 5. Kao, H. T. et al. Molecular evolution of the synapsin gene family. J. Exp. Zool. 285, 360–377 (1999). 6. Sudhof, T. C. et al. Synapsins: mosaics of shared and individual domains in a family of synaptic vesicle phosphoproteins. Science 245, 1474–1480 (1989). 7. Huttner, W. B., Schiebler, W., Greengard, P. & De Camilli, P. Synapsin I (protein I), a nerve terminal-specific phosphoprotein. III. Its association with synaptic vesicles studied in a highly purified synaptic vesicle preparation. J. Cell Biol. 96, 1374–1388 (1983). 8. De Camilli, P., Harris, S. M. Jr., Huttner, W. B. & Greengard, P. Synapsin I (Protein I), a nerve terminal-specific phosphoprotein. II. Its specific association with synaptic vesicles demonstrated by immunocytochemistry in agarose-embedded synaptosomes. J. Cell Biol. 96, 1355–1373 (1983). 9. Mandell, J. W. et al. Synapsins in the vertebrate retina: absence from ribbon synapses and heterogeneous distribution among conventional synapses. Neuron 5, 19–33 (1990). 10. Mandell, J. W., Czernik, A. J., De Camilli, P., Greengard, P. & TownesAnderson, E. Differential expression of synapsins I and II among rat retinal synapses. J. Neurosci. 12, 1736–1749 (1992). 11. Bahler, M. & Greengard, P. Synapsin I bundles F-actin in a phosphorylationdependent manner. Nature 326, 704–707 (1987). 12. Bahler, M., Benfenati, F., Valtorta, F., Czernik, A. J. & Greengard, P. Characterization of synapsin I fragments produced by cysteine-specific cleavage: a study of their interactions with F-actin. J. Cell Biol. 108, 1841–1849 (1989). 13. Benfenati, F., Valtorta, F., Bahler, M. & Greengard, P. Synapsin I, a neuronspecific phosphoprotein interacting with small synaptic vesicles and F-actin. Cell Biol. Int. Rep. 13, 1007–1021 (1989). 14. Benfenati, F., Bahler, M., Jahn, R. & Greengard, P. Interactions of synapsin I


31. 32. 33.

34.

35.

36. 37. 38. 39. 40. 41. 42. 43.

with small synaptic vesicles: distinct sites in synapsin I bind to vesicle phospholipids and vesicle proteins. J. Cell Biol. 108, 1863–1872 (1989). Benfenati, F., Greengard, P., Brunner, J. & Bahler, M. Electrostatic and hydrophobic interactions of synapsin I and synapsin I fragments with phospholipid bilayers. J. Cell Biol. 108, 1851–1862 (1989). Benfenati, F., Valtorta, F., Chieregatti, E. & Greengard, P. Interaction of free and synaptic vesicle-bound synapsin I with F-actin. Neuron 8, 377–386 (1992). Benfenati, F. et al. Interactions of synapsin I with phospholipids: possible role in synaptic vesicle clustering and in the maintenance of bilayer structures. J. Cell Biol. 123, 1845–1855 (1993). Ceccaldi, P. E. et al. Dephosphorylated synapsin I anchors synaptic vesicles to actin cytoskeleton: an analysis by videomicroscopy. J. Cell Biol. 128, 905–912 (1995). Chilcote, T. J., Siow, Y. L., Schaeffer, E., Greengard, P. & Thiel, G. Synapsin IIa bundles actin filaments. J. Neurochem. 63, 1568–1571 (1994). Greengard, P., Browning, M. D., McGuinness, T. L. & Llinas, R. Synapsin I, a phosphoprotein associated with synaptic vesicles: possible role in regulation of neurotransmitter release. Adv. Exp. Med. Biol. 221, 135–153 (1987). Hosaka, M., Hammer, R. E. & Sudhof, T. C. A phospho-switch controls the dynamic association of synapsins with synaptic vesicles. Neuron 24, 377–387 (1999). Thiel, G., Sudhof, T. C. & Greengard, P. Synapsin II. Mapping of a domain in the NH2-terminal region which binds to small synaptic vesicles. J. Biol. Chem. 265, 16527–16533 (1990). Schiebler, W., Jahn, R., Doucet, J. P., Rothlein, J. & Greengard, P. Characterization of synapsin I binding to small synaptic vesicles. J. Biol. Chem. 261, 8383–8390 (1986). De Camilli, P., Benfenati, F., Valtorta, F. & Greengard, P. The synapsins. Annu. Rev. Cell Biol. 6, 433–460 (1990). Li, L. et al. Impairment of synaptic vesicle clustering and of synaptic transmission, and increased seizure propensity, in synapsin I-deficient mice. Proc. Natl. Acad. Sci. USA 92, 9235–9239 (1995). Rosahl, T. W. et al. Essential functions of synapsins I and II in synaptic vesicle regulation. Nature 375, 488–493 (1995). Rosahl, T. W. et al. Short-term synaptic plasticity is altered in mice lacking synapsin I. Cell 75, 661–670 (1993). Stefani, G. et al. Kinetic analysis of the phosphorylation-dependent interactions of synapsin I with rat brain synaptic vesicles. J. Physiol. (Lond.) 504, 501–515 (1997). Sihra, T. S., Wang, J. K., Gorelick, F. S. & Greengard, P. Translocation of synapsin I in response to depolarization of isolated nerve terminals. Proc. Natl. Acad. Sci. USA 86, 8108–8112 (1989). Torri, T. F., Bossi, M., Fesce, R., Greengard, P. & Valtorta, F. Synapsin I partially dissociates from synaptic vesicles during exocytosis induced by electrical stimulation. Neuron 9, 1143–1153 (1992). Henkel, A. W. & Betz, W. J. Staurosporine blocks evoked release of FM1-43 but not acetylcholine from frog motor nerve terminals. J. Neurosci. 15, 8246–8258 (1995). Sankaranarayanan, S. & Ryan, T. A. Real-time measurements of vesicleSNARE recycling in synapses of the central nervous system. Nat. Cell Biol. 2, 197–204 (2000). Czernik, A. J., Pang, D. T. & Greengard, P. Amino acid sequences surrounding the cAMP-dependent and calcium/calmodulin-dependent phosphorylation sites in rat and bovine synapsin I. Proc. Natl. Acad. Sci. USA 84, 7518–7522 (1987). Llinas, R., McGuinness, T. L., Leonard, C. S., Sugimori, M. & Greengard, P. Intraterminal injection of synapsin I or calcium/calmodulin-dependent protein kinase II alters neurotransmitter release at the squid giant synapse. Proc. Natl. Acad. Sci. USA 82, 3035–3039 (1985). Llinas, R., Gruner, J. A., Sugimori, M., McGuinness, T. L. & Greengard, P. Regulation by synapsin I and Ca2+-calmodulin-dependent protein kinase II of the transmitter release in squid giant synapse. J. Physiol. (Lond.) 436, 257–282 (1991). Hosaka, M. & Sudhof, T. C. Homo- and heterodimerization of synapsins. J. Biol. Chem. 274, 16747–16753 (1999). Tanaka, H. et al. Molecular modification of N-cadherin in response to synaptic activity. Neuron 25, 93–107 (2000). Ryan, T. A., Li, L., Chin, L. S., Greengard, P. & Smith, S. J. Synaptic vesicle recycling in synapsin I knock-out mice. J. Cell Biol. 134, 1219–1227 (1996). Esser, L. et al. Synapsin I is structurally similar to ATP-utilizing enzymes. EMBO J. 17, 977–984 (1998). Hosaka, M. & Sudhof, T. C. Synapsins I and II are ATP-binding proteins with differential Ca2+ regulation. J. Biol. Chem. 273, 1425–1429 (1998). Ferreira, A. et al. Distinct roles of synapsin I and synapsin II during neuronal development. Mol. Med. 4, 22–28 (1998). Ryan, T. A. Inhibitors of myosin light chain kinase block synaptic vesicle pool mobilization during action potential firing. J. Neurosci. 19, 1317–1323 (1999). Threadgill, R., Bobb, K. & Ghosh, A. Regulation of dendritic growth and remodeling by Rho, Rac, and Cdc42. Neuron 19, 625–634 (1997).

1193


articles


TNFα contributes to the death of NGF-dependent neurons during development Victoria Barker1, Gayle Middleton1, Fleur Davey2 and Alun M. Davies1,3 1 Department of Preclinical Veterinary Sciences, Royal (Dick) School of Veterinary Studies, Summerhall Square, Edinburgh EH9 1QH, Scotland 2 School of Biomedical Sciences, Bute Medical Buildings, University of St. Andrews, St. Andrews, Fife KY16 9AT, Scotland 3 Rinat Neuroscience Corporation, 3155 Porter Drive, Palo Alto, California 94304, USA

Correspondence should be addressed to A.M.D. ([email protected])

Published online: 29 October 2001, DOI: 10.1038/nn755 Many sympathetic and sensory neurons depend on a supply of nerve growth factor (NGF) from their targets during development, and neurons that fail to obtain sufficient NGF die by apoptosis. Here we show that tumor necrosis factor alpha (TNFα) is involved in bringing about the death of NGFdeprived neurons. Function-blocking antibodies against either TNFα or TNF receptor 1 (TNFR1) rescued many sympathetic and sensory neurons following NGF deprivation in vitro. Fewer sympathetic and sensory neurons died during the phase of naturally occurring neuronal death in TNF-deficient embryos, and neurons from these embryos survived in culture better than wild-type neurons. These neurons coexpress TNFα and TNFR1 during this stage of development, suggesting that TNFα acts by an autocrine loop.

Neurons are generated in excess in the developing vertebrate peripheral nervous system, and the superfluous neurons are lost during a phase of programmed cell death that occurs shortly after they innervate their targets. NGF is the founder of a family of structurally related secreted proteins termed neurotrophins that promote and regulate the survival of many kinds of neurons in the peripheral nervous system during this stage of development. NGF is required for the survival of sympathetic neurons and a subset of nociceptive sensory neurons. These neurons are lost in developing rodents treated with function-blocking anti-NGF antibodies1 and in mice that are homozygous for null mutations in the NGF gene2 or the TrkA gene, which encodes the NGF receptor tyrosine kinase3. NGF is synthesized in the peripheral target tissues of NGF-dependent neurons in proportion to their innervation density4,5, and administration of exogenous NGF prevents naturally occurring cell death within populations of NGF-dependent neurons during development1. Because the binding of NGF to TrkA generates survival signals within the cell that prevent caspase activation and subsequent apoptosis6, it is assumed that neurons die following NGF deprivation because of the withdrawal of these survival signals. However, we show here that this death is due in part to the action of TNFα, a proinflammatory cytokine that induces apoptosis in some cell types. TNFα exerts its effects by binding to the receptors TNFR1 and TNFR2. The cytotoxic effects of TNFα are mediated via TNFR1, which has a cytoplasmic death domain that interacts with the adapter protein TRADD following ligand binding. TRADD in turn interacts with another adapter protein FADD, which recruits and activates pro-caspase 8 with the resultant activation of the cell death machinery7. TNFR2 lacks a death domain but synergistically enhances TNFR1-induced cytotoxicity8. In the nervous system, TNFα contributes to neuronal death associated 1194

with ischemia9, HIV-1 infection10 and axotomy11, and can exert a neuroprotective effect against glutamate excitotoxicity12,13. It is not known, however, if TNFα is involved in bringing about the death of neurons that fail to obtain adequate trophic support from their innervation targets during normal embryonic development.

RESULTS TNFα and TNFR1 antibodies rescue neurons To investigate if TNFα is involved in promoting apoptosis of embryonic neurons deprived of trophic support, we studied the effects of function-blocking anti-TNFα and anti-TNFR1 antibodies on the survival of NGF-dependent neurons following NGF deprivation. We established low-density, dissociated cultures of sympathetic neurons from the superior cervical ganglia (SCG) and sensory neurons from the trigeminal ganglia of mouse embryos at E16, when most of these neurons have become dependent on NGF for survival in vitro and naturally occurring neuronal death is occurring in vivo14–16. After 12 hours incubation with NGF, the cultures were washed extensively to remove this neurotrophin and were either resupplemented with NGF, or grown without NGF and treated with anti-TNFα or anti-TNFR1 antibodies or TNFα. Whereas about 80% of the sympathetic and sensory neurons survived for at least another 48 hours after resupplementation with NGF, over 80% died within 48 hours of NGF deprivation (Fig. 1). Treatment of NGF-deprived neurons with either anti-TNFα or anti-TNFR1 antibodies rescued a quarter to a third of the neurons that would have otherwise died (Fig. 1). Dose–response analysis revealed that this effect of the antibodies reached a plateau at a concentration of 2 ng/ml, and higher concentrations of antibodies singularly and in combination did not rescue additional neurons (data not shown). These results were not due to non-specific actions of antibodies on nature neuroscience • volume 4 no 12 • december 2001



articles

Fig. 1. Survival of E16 SCG and trigeminal neurons cultured with NGF, TNFα and functionblocking TNFα or TNFR1 antibodies after NGF deprivation. The neurons were grown for 12 h after plating in medium containing 2 ng/ml NGF. After washing, growth factors and antibodies were added to a concentration of 2 ng/ml. The numbers of surviving neurons were counted after a further 48-h incubation and expressed as a percentage of the number counted after NGF deprivation. Control cultures received no reagents following NGF deprivation. The means and standard errors of six separate experiments are shown.

neuronal survival because there was no significant difference between the number of neurons surviving in control cultures and cultures treated with antibodies against an intracellular antigen (βIII tubulin) even at concentrations as high as 50 ng/ml. Anti-TNFα and anti-TNFR1 rescued a similar proportion of sympathetic and sensory neurons that were plated in medium containing these antibodies from the outset (data not shown). These results suggest that TNFα synthesized by cells in dissociated cultures of embryonic sympathetic and sensory ganglia is involved in bringing about neuronal death following NGF deprivation, and that this action of TNFα is mediated at least in part by TNFR1. TNFα treatment also reduced the survival of sympathetic and sensory neurons grown either with or without NGF (Fig. 2). Dose–response analysis revealed that as little as 3.2 pg/ml TNFα caused significant reductions in the survival of neurons grown in either the presence or absence of NGF (p < 0.005, t-tests). This cytotoxic effect of TNFα reached a plateau at a concentration of 2 ng/ml, and no additional death was observed at a 25-fold higher concentration. At these concentrations of TNFα, less than 2% of the neurons were remaining in NGF-free cultures (compared to approximately 15% survival in NGF-free cultures without TNFα) and there was an approximately 20% drop in

survival of neurons grown with TNFα plus NGF relative to neurons grown with NGF alone. These results suggest that endogenously produced TNFα is not functioning to maximum effect in promoting neuronal death and that only a proportion of the neurons are susceptible to the cytotoxic actions of TNFα at this stage of development. Also, TNFα receptors are active in sympathetic neurons, as TNFα treatment promotes nuclear translocation of the RelA subunit of NFκB (data not shown), although we do not know if this signaling pathway is involved in mediating the cytotoxic action of TNFα on these neurons. Enhanced survival of TNF-deficient neurons in culture To provide additional evidence for the involvement of TNFα in bringing about neuronal death following NGF deprivation, we compared the survival of embryonic SCG and trigeminal neurons cultured from wild-type mice and mice that are homozygous for a null mutation in the TNFα gene. After 48 hours incubation, there were approximately twice as many sympathetic and sensory neurons surviving in cultures established from E16 TNFα–/– mice compared with neurons from wild-type embryos in the same litters (Fig. 3). These results further support the idea that endogenously produced TNFα is involved in bringing about the death of embryonic sympathetic and sensory neurons in the absence of NGF.

Fig. 2. E16 SCG and trigeminal neuron dose responses to NGF and TNFα. The neurons were grown for 12 h after plating in medium containing 2 ng/ml NGF. After washing, the cultures were supplemented with NGF, TNFα or NGF plus TNFα over a range of concentrations. The numbers of surviving neurons were counted after a further 48-h incubation and expressed as a percentage of the number counted after NGF deprivation. Control cultures received no reagents following NGF deprivation. The means and standard errors of six separate experiments are shown. nature neuroscience • volume 4 no 12 • december 2001

1195


articles


Fig. 3. In vitro survival of SCG and trigeminal neurons from TNF-deficient and wild-type embryos. Cultures were established from E16 TNF–/– and TNF+/+ embryos. After 48 h incubation in defined, serumfree medium, the number of surviving neurons was counted, and is expressed here as a percentage of the number counted 6 h after plating. The means and standard errors of three separate experiments are shown.

Reduced neuronal death in TNF-deficient embryos To ascertain the physiological relevance of our in vitro observations, we estimated the number of surviving neurons and the extent of neuronal death in the SCG and trigeminal ganglia of wild-type and TNFα–/– mice at two stages during the period of naturally occurring neuronal death by counting the total number of neurons and the number of pyknotic neurons in serial sections (Fig. 4). At E16, the SCG and trigeminal ganglia of TNFα–/– embryos contained only a quarter as many dying neurons as in wild-type embryos, and there were significantly more neurons

in the SCG (52% increase, p < 0.0001, t-test) and trigeminal ganglia (23% increase, p < 0.0001, t-test). By the first postnatal day (P1), the SCG and trigeminal ganglia of TNFα–/– embryos contained less than half as many dying neurons as in wild-type embryos, and neurons were still significantly more abundant in the SCG (14% increase, p < 0.001, t-test) and trigeminal ganglia (26% increase, p < 0.0001, t-test) of these TNFα-deficient embryos compared with wild-type embryos. These results demonstrate that TNFα is required to stimulate the death of a proportion of neurons in the developing SCG and trigeminal ganglia during the period of naturally occurring neuronal death. Analysis of neuronal death in the nodose ganglia also revealed significantly reduced neuronal death and increased numbers of neurons in TNFα-deficient mice at these stages of development (data not shown), suggesting that TNFα is also involved in promoting the death of sensory neurons that depend on BDNF and NT4 for survival during the period of naturally occurring neuronal death. Developing neurons express TNFα and TNFR1 To further substantiate the role of TNFα in embryonic sensory and sympathetic neurons and investigate its mode of action, we used immunocytochemistry to identify which cells express TNFα and TNFR1 in dissociated cultures. Because the

Fig. 4. Number of pyknotic neurons and total numbers of neurons in the SCG and trigeminal of E16 and P1 TNFα+/+ and TNFα–/– mice. The means and standard errors of the data obtained from both sets of ganglia from 3 E16 and 4 P1 TNFα+/+ mice and from 4 E16 and 5 P1 TNFα–/– mice are shown.

1196




articles

a

b

Fig. 5. Expression of TNFα and TNFR1 by cultured SCG neurons. E16 SCG neurons from wild-type embryos were stained for TNFα (a) and TNFR1 (b); E16 SCG neurons (arrowheads) from TNFα–/– embryos did not stain for TNFα (c), and no neuronal staining was observed in wild-type neurons (arrows) if primary antibodies were omitted (d). The few non-neuronal cells in these cultures showed a low level of TNFα immunoreactivity (arrowhead, a). Scale bar, 20 µM.

c

d

these cells and Fas-L is induced following neurotrophic factor deprivation. Treatment with Fas-L causes the death of many neurons even in the presence of optimal concentrations of neurotrophic factors. Thus, Fas signaling is triggered by neurotrophic factor withdrawal and this in turn brings about the subsequent demise of many of the neurons. Similarly, TNFα-mediated neuronal death following brain injury and ischemia in adult rats is associated with rapid upregulation in the expression of TNFα21,22. Immunocytochemical analysis of TNFα and TNFR1 expression on cultured embryonic sympathetic neurons suggests that virtually all of these neurons express this ligand/receptor combination. This observation, together with our demonstration that inhibition of neuronal death by function-blocking anti-TNFα and anti-TNFR1 antibodies was observed in low dissociated density cultures that contained only a very small percentage of non-neuronal cells, suggest that TNFα contributes to the death of NGF-deprived neurons by an autocrine mechanism. However, the low level of TNFα immunoreactivity observed in non-neuronal cells raises the possibility that the neurons may additionally obtain some TNFα from other cells. In summary, our findings add a twist to the molecular mechanisms that bring about the death of developing NGF-dependent neurons that fail to obtain adequate NGF to support their survival. The demise of these neurons seems to be caused in part by the cytotoxic actions of TNFα that they produce themselves. In future work, it will be important to ascertain the extent to which TNFα, FasL and possibly other cytotoxic cytokines are involved in bringing about naturally occurring neuronal death in these and other populations of neurons in the developing vertebrate nervous system.

serum-free medium used in our cultures is not conducive to the survival of non-neuronal cells, most cells remaining in E16 cultures after 48 hours of incubation were neurons (>90%). Virtually all of these neurons were immunoreactive for TNFα protein and for TNFR1 (Fig. 5). The small number of nonneuronal cells in these cultures displayed only a low level of immunoreactivity for TNFα and TNFR1. Neurons in cultures established from TNFα–/– embryos showed no immunoreactivity for TNFα and no staining was observed in cultures of wild-type neurons if the primary antibodies were omitted, demonstrating specific staining for TNFα and its receptor. These findings indicate that most if not all neurons synthesize TNFα and express the TNFR1 receptor during the stage of NGF dependence, and suggest that TNFα exerts its actions by an autocrine mechanism.

DISCUSSION We have used several complementary experimental approaches to demonstrate that TNFα is involved in bringing about the death of sympathetic and sensory neurons during the phase of programmed cell death in the developing peripheral nervous system. Function-blocking antibodies to either TNFα or TNFR1 but not control antibodies to an unrelated protein prevented the death of a significant proportion of these neurons following NGF deprivation in vitro. TNFα treatment promoted the death of cultured sympathetic and sensory neurons in a dose-dependent manner. The sympathetic and sensory ganglia of TNFα-deficient mice contained substantially fewer dying neurons and significantly greater numbers of neurons than the ganglia of wild-type littermates during the phase of programmed cell death in vivo. Sympathetic and sensory neurons from TNFα-deficient embryos survived better in culture than neurons from wild-type embryos. There are some parallels between our findings and recent work on Fas, a member of the TNF receptor family, and its ligand Fas-L, which have a well-established role in triggering apoptosis in lymphocytes17. Blocking interaction between Fas and Fas-L reduces cell death induced by neurotrophic factor deprivation in phaeochromocytoma cells, cerebellar granule cells and spinal motoneurons18–20. Fas is constitutively expressed in nature neuroscience • volume 4 no 12 • december 2001

METHODS Neuron culture and survival assays. Dissociated cultures were established from the SCG and trigeminal ganglia of E16 CD1 embryos. The dissected ganglia were trypsinised and dissociated by trituration and were grown in defined, serum-free medium on a poly-ornithine/laminin substratum23 in the 11-mm diameter wells of 4-well dishes (Greiner, Stonehouse, UK) at a density of 200–400 neurons per well. After an initial 12-h incubation period with 2 ng/ml NGF, the neurons were washed extensively to remove this factor and were subsequently incubated in medium with either no added factors (control), NGF (gift of A. Rosenthal, Rinat Neuroscience Corporation, Palo Alto, California), TNFα (Calbiochem, Nottingham, UK), NGF plus TNFα, function-blocking antiTNFα antibody (Upstate Biotechnology, Buckingham, UK), function-blocking anti-TNFR1 antibody (Oncogene Research Products, Nottingham, UK) or anti-βIII tubulin antibody (Promega, Southampton, UK). After 48 h, the numbers of surviving neurons in each well were counted and expressed as a percentage of the number of attached neurons after the 12-h wash. Very similar results were obtained if the 1197



articles

cultures were additionally treated with anti-NGF antibody to functionally inactivate any residual NGF after washing. To further investigate the role of TNFα in promoting neuronal death, we established cultures from mouse embryos that have a null mutation in the TNFα gene. TNFα +/– mice (Jackson Laboratories, Maine) were crossed to produce TNF α –/– , TNF α +/– and TNF α +/+ embryos. After 16 days gestation, the pregnant females were killed and the embryos were genotyped by a PCR-based method. The SCG and trigeminal ganglia from wild-type and homozygous embryos were separately pooled, and low-density, dissociated cultures were established. The neurons were grown in defined medium for 48 h and the number surviving after this time is expressed as a percentage of the number of attached neurons counted 6 h after plating. All animal work was approved by our institutional animal use committee and by the Home Office. Immunocytochemistry. Immunocytochemistry was used to visualize expression of TNFα and TNFR1 in cultured neurons. The cultures were fixed with methanol at –20°C followed by 1% H2O2 to quench endogenous peroxidase activity and permeabilization in 0.5% Triton X-100 in PBS. After incubating at 4°C overnight with 1:200 dilutions of either anti-mouse TNFα goat polyclonal antibody (L-19, Santa Cruz Biotechnology, Wembley, UK) or anti-mouse TNFR1 goat polyclonal antibody (Calbiochem), bound primary antibody was detected using the Vector Elite ABC kit (Vector Labs, Orton Southgate, UK) according to the manufacturer’s instructions. The stained neurons were viewed with a Nikon Diaphot microscope (Nikon, Kingston on Thames, UK). Quantification of neurons in the SCG and trigeminal ganglia. The heads from E16 and P1 TNF–/– and wild-type embryos were fixed for 1 h in Carnoy’s fluid (60% ethanol, 30% chloroform, and 10% glacial acetic acid). Following dehydration through a graded alcohol series, the tissue was paraffin-wax-embedded. Serial sections of the heads were cut at 8 µm and were mounted onto polylysine-coated slides (BDH) or Gold Seal Ultrastick Slides (Erie Scientific, Loughborough, UK). To identify all neurons in these sections, the sections were stained for β-tubulin class III. Sections were cleared in xylene and rehydrated before quenching in 3% hydrogen peroxide in methanol for 20 min. Non-specific antibody binding was blocked in 10% horse serum, 0.5% Triton X-100 in PBS before incubation with mouse anti-βIII tubulin antibody (Promega) diluted 1:10,000 in PBS overnight at 4°C. The cells were then labeled using biotinylated secondary antibody (1:200), avidin, biotinylated horseradish peroxidase macromolecular complex (Vectastain ABC Kit, Vector Labs). The substrate used for the reaction was 1 mg/ml diaminobenzidine tetrachloride (FastDAB, Sigma, St. Louis, Missouri). The sections were then counterstained with Gill’s hemotoxylin before dehydration and mounting. Neuronal number was quantified using a digital stereology system that employs a combination of the optical dissector and volume fraction/Cavalieri methods (Kinetics Imaging, Bromborough, UK). For quantification of the number of pyknotic nuclei in the ganglia, the sections were stained with cresyl fast violet acetate. Pyknotic nuclei were then counted using the digital stereology system. All sections were coded prior to estimating the number of neurons and pyknotic nuclei to avoid any observer bias.

1198

ACKNOWLEDGEMENTS This work was supported by grants from the Wellcome Trust and European Commission.

RECEIVED 18 AUGUST; ACCEPTED 21 SEPTEMBER 2001 1. Levi-Montalcini, R. The nerve growth factor 35 years later. Science 237, 1154–1162 (1987). 2. Crowley, C. et al. Mice lacking nerve growth factor display perinatal loss of sensory and sympathetic neurons yet develop basal forebrain cholinergic neurons. Cell 76, 1001–1011 (1994). 3. Smeyne, R.J. et al. Severe sensory and sympathetic neuropathies in mice carrying a disrupted Trk/NGF receptor gene. Nature 368, 246–249 (1994). 4. Korsching, S. & Thoenen, H. Nerve growth factor in sympathetic ganglia and corresponding target organs of the rat: correlation with density of sympathetic innervation. Proc. Natl. Acad. Sci. USA 80, 3513–3516 (1983). 5. Harper, S. & Davies, A. M. NGF mRNA expression in developing cutaneous epithelium related to innervation density. Development 110, 515–519 (1990). 6. Kaplan, D. R. & Miller, F. D. Neurotrophin signal transduction in the nervous system. Curr. Opin. Neurobiol. 10, 381–391 (2000). 7. Ashkenazi, A. & Dixit, V. M. Death receptors: signaling and modulation. Science 281, 1305–1308 (1998). 8. Grell, M. et al. Induction of cell death by tumour necrosis factor (TNF) receptor 2, CD40 and CD30: a role for TNF-R1 activation by endogenous membrane-anchored TNF. EMBO J. 18, 3034–3043 (1999). 9. Dawson, D. A., Martin, D. & Hallenbeck, J. M. Inhibition of tumor necrosis factor-alpha reduces focal cerebral ischemic injury in the spontaneously hypertensive rat. Neurosci. Lett. 218, 41–44 (1996). 10. New, D. R., Maggirwar, S. B., Epstein, L. G., Dewhurst, S. & Gelbard, H. A. HIV-1 Tat induces neuronal death via tumor necrosis factor-α and activation of non-N-methyl-D-aspartate receptors by a NFκB-independent mechanism. J. Biol. Chem. 273, 17852–17858 (1998). 11. Terrado, J. et al. Soluble TNF receptors partially protect injured motoneurons in the postnatal CNS. Eur. J. Neurosci. 12, 3443–3447 (2000). 12. Houzen, H., Kikuchi, S., Kanno, M., Shinpo, K. & Tashiro, K. Tumor necrosis factor enhancement of transient outward potassium currents in cultured rat cortical neurons. J. Neurosci. Res. 50, 990–999 (1997). 13. Carlson, N. G., Bacchi, A., Rogers, S. W. & Gahring, L. C. Nicotine blocks TNF-α-mediated neuroprotection to NMDA by an α-bungarotoxin-sensitive pathway. J. Neurobiol. 35, 29–36 (1998). 14. Wyatt, S. & Davies, A. M. Regulation of nerve growth factor receptor gene expression in sympathetic neurons during development. J. Cell Biol. 130, 1435–1446 (1995). 15. Davies, A. M. & Lumsden, A. G. S. Relation of target encounter and neuronal death to nerve growth factor responsiveness in the developing mouse trigeminal ganglion. J. Comp. Neurol. 223, 124–137 (1984). 16. Francis, N. et al. NT-3, like NGF, is required for survival of sympathetic neurons, but not their precursors. Dev. Biol. 210, 411–427 (1999). 17. Nagata, S. Apoptosis by death factor. Cell 88, 355–365 (1997). 18. Le-Niculescu, H. et al. Withdrawal of survival factors results in activation of the JNK pathway in neuronal cells leading to Fas ligand induction and cell death. Mol. Cell. Biol. 19, 751–63 (1999). 19. Brunet, A. et al. Akt promotes cell survival by phosphorylating and inhibiting a Forkhead transcription factor. Cell 96, 857–868 (1999). 20. Raoul, C., Henderson, C. E. & Pettmann, B. Programmed cell death of embryonic motoneurons triggered through the Fas death receptor. J. Cell Biol. 147, 1049–1062 (1999). 21. Knoblach, S. M., Fan, L. & Faden, A. I. Early neuronal expression of tumor necrosis factor-α after experimental brain injury contributes to neurological impairment. J. Neuroimmunol. 95, 115–125 (1999). 22. Botchkina, G. I., Meistrell, M. E. III, Botchkina, I. L. & Tracey, K. J. Expression of TNF and TNF receptors (p55 and p75) in the rat brain after focal cerebral ischemia. Mol. Med. 3, 765–781 (1997). 23. Davies, A. M., Lee, K. F. & Jaenisch, R. p75-deficient trigeminal sensory neurons have an altered response to NGF but not to other neurotrophins. Neuron 11, 565–574 (1993).



articles


Wallerian degeneration of injured axons and synapses is delayed by a Ube4b/Nmnat chimeric gene Till G.A. Mack1, Michael Reiner2, Bogdan Beirowski2, Weiqian Mi1, Monica Emanuelli3, Diana Wagner1, Derek Thomson4, Tom Gillingwater4, Felipe Court4, Laura Conforti5, F. Shama Fernando6, Andrea Tarlton7, Christian Andressen2, Klaus Addicks2, Giulio Magni3, Richard R. Ribchester4, V. Hugh Perry8 and Michael P. Coleman1,6 1 Center for Molecular Medicine (ZMMK) and Institute for Genetics, University of Cologne, Zuelpicher Strasse 47, D-50674 Cologne, Germany 2 Department of Anatomy I, University of Cologne, Joseph-Stelzmann Strasse 9, D-50931 Cologne, Germany 3 Institute of Biochemistry, University of Ancona, Via Ranieri, 60131 Ancona, Italy 4 Department of Neuroscience, University of Edinburgh, 1 George Square, Edinburgh, EH8 9JZ, UK 5 Molecular Neurobiology Laboratory, Mario Negri Pharmaceutical Research Institute, Via Eritrea, 62, 20157 Milan, Italy 6 Department of Pharmacology, University of Oxford, Mansfield Road, Oxford, OX1 3QT, UK 7 Department of Biochemistry, University of Oxford, South Parks Road, Oxford, OX1 3QU, UK 8 School of Biological Sciences, University of Southampton, Biomedical Sciences Building, Southampton, SO16 7PX, UK

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

Published online: 19 November 2001, DOI: 10.1038/nn770 Axons and their synapses distal to an injury undergo rapid Wallerian degeneration, but axons in the C57BL/WldS mouse are protected. The degenerative and protective mechanisms are unknown. We identified the protective gene, which encodes an N-terminal fragment of ubiquitination factor E4B (Ube4b) fused to nicotinamide mononucleotide adenylyltransferase (Nmnat), and showed that it confers a dose-dependent block of Wallerian degeneration. Transected distal axons survived for two weeks, and neuromuscular junctions were also protected. Surprisingly, the Wld protein was located predominantly in the nucleus, indicating an indirect protective mechanism. Nmnat enzyme activity, but not NAD+ content, was increased fourfold in WldS tissues. Thus, axon protection is likely to be mediated by altered ubiquitination or pyridine nucleotide metabolism.

The distal segment of an injured nerve normally undergoes Wallerian degeneration within 24–48 hours1. Axon death in diverse neurodegenerative diseases follows the same final pathway. The earliest observable events, disruption of the cytoskeleton and blebbing of the axolemma, occur within the axon itself, whereas later stages also involve the reaction of other cell types, such as Schwann cells and macrophages. It is not known how Wallerian degeneration is initiated, but the mechanism is clearly distinct from neuronal cell body degeneration2,3. Remarkably, central and peripheral nervous system axons in the slow Wallerian degeneration mutant mouse, C57BL/WldS, survive several weeks after transection4–6. The neuroprotective phenotype is dominant5 and intrinsic to the axon2,7,8. Thus, an unknown protective factor should exist in WldS axons even before nerve transection, as the protected distal segment of axon is isolated from sites of protein translation. The existence of the putative regulatory molecule suggests that Wallerian degeneration is not a passive process, as previously thought, but an active one that removes damaged axons8. How this degenerative process is prevented in healthy axons, and thus the nature of the process itself, should follow from the identification of the Wld gene. nature neuroscience • volume 4 no 12 • december 2001

Wallerian degeneration has a prominent causative role in a spectrum of human neuropathologies. Axon loss occurs not only in traumatic disorders such as spinal cord injury9, but is an early event in numerous neurological disorders of diverse etiology, such as amyotrophic lateral sclerosis10, multiple sclerosis11 and toxic neuropathy12. The WldS mutation protects also from vincristine toxicity13 and its potential for protection of axons in diverse neurological diseases is an area of considerable current interest. Protection of neuronal cell bodies often fails to prevent neurological disease14,15, so it is also important to find ways to protect axons. Thus, the Wld gene may open a new avenue for therapeutic strategies. An 85-kb tandem triplication16 has been characterized within the WldS region on distal mouse chromosome 4 (ref. 17). This led to the discovery of a chimeric gene containing the 5´ end of Ube4b and a gene of previously unknown function, D4Cole1e18. Both parent proteins are also expressed in WldS mice. The human homolog of D4Cole1e (F.S. Fernando, unpublished data) is identical to human NMNAT19, a key enzyme in the synthetic pathway of NAD+ (ref. 20). Although the chimeric gene is the most plausible candidate for Wld, the triplication also directly affects retinol binding protein 7 (Rbp7) and could exert a position effect 1199


articles


a

c

Fig. 1. Preserved axon ultrastructure in distal sciatic nerve 5 days after transection. Electron micrographs (3,400×) of transverse thin sections of lesioned sciatic nerve 2–4 mm distal to the lesion site after 5 days. (a) Wild-type, (b) 4836 homozygote, (c) 4830 homozygote (d) WldS homozygote. Insets, 20,000× magnifications of axons marked by asterisk (top inset) and ‘#’ (bottom inset) from main images.

b

β-actin promoter. Four transgenic lines (4830, 4836, 4839 and 4858) expressed the protein from multi-copy integrations. The transgene expression level increased in the order 4839 < 4830 ≈ 4858 < WldS ≈ 4836 (Fig. 3). Line 4858, whose expression level and phenotype was similar to line 4830, is not discussed further. Like spontaneous WldS mice, transgenic mice up to at least one year old showed no unusual overt phenotype.

d

on nearby genes. Indeed, mutation of the human homolog of one neighboring gene, kinesin family 1b (Kif1b)21, causes axon loss in Charcot-Marie-Tooth disease type 2A22. To test the hypothesis that the Ube4b/Nmnat chimeric gene confers the slow Wallerian degeneration phenotype, we expressed it in transgenic mice. The WldS phenotype was reproduced fully in one line and partially in other lines according to transgene expression level, thus proving that the chimeric gene is the Wld gene. We detected Wld protein in nuclei of neurons but neither in axons nor in Schwann cells, implicating the existence of downstream factor(s) that mediate the protective effect. Further, we report that the Wld protein showed Nmnat enzyme activity and acted in a strongly dose-dependent manner.

RESULTS

Structural preservation of transected axons First we tested the structural preservation of axons following unilateral sciatic nerve transection, analyzing a nerve segment 2–4 mm distal to the lesion after 3–5 days. Electron microscopy revealed that almost all axons in line 4836 homozygotes, separated from their cell bodies for five days, contained fully preserved cytoskeleton (Fig. 1 and Supplementary Fig. 1, available on the Nature Neuroscience web site), thus successfully reproducing the WldS phenotype. In contrast, all myelinated and unmyelinated wild-type axons showed clear signs of degeneration. Line 4830 axons were partially protected in accordance with the lower transgene expression level in this line (Fig. 3a) and line 4839 homozygotes also showed partial protection after

b

a

c

e

Generation of transgenic mice To test the role of the Ube4b/Nmnat chimeric gene, we generated transgenic mice expressing the Ube4b/Nmnat cDNA from a Fig. 2. Physiological and morphological evidence for preservation of axons and neuromuscular junctions. Intracellular recordings (a, b), vital staining (c, d) and immunostaining (e, f) in isolated transgenic muscles. (a) Homozygous 4836 FDB, 3 days after axotomy; (b) hemizygous 4836 FDB after 2 days. (c, d) Lumbrical muscles from 4836 hemizygote, 3 days after axotomy (c); 4836 homozygote, 5 days after axotomy (d) vitally stained with FM1-43, which indicates intact synaptic transmission mechanisms, together with TRITC-α-bungarotoxin staining of acetylcholine receptors. (e, f) Immunostaining for neurofilament/SV2 (FITC) and TRITC-α-bungarotoxin. (e) 4836 homozygous lumbrical muscle, 5 days after axotomy, showing occupied endplates and some axon swelling. (f) Confocal stereo pair of 5-day axotomized 4836 homozygous lumbrical muscle showing almost fully occupied endplates (left and right), partially occupied (second right) and junction with only a slender axonal filament tipped by a ‘retraction bulb’ (second left). Scale bar, 20 µm (c–e); 50 µm (f).

1200

d

f




articles

Fig. 3. Dose-dependence of axon protection by the Wld gene. (a) Top, western blot showing the extent of 200-kD neurofilament protein degradation in distal sciatic nerve 3 days after transection (5 days for 4836 and WldS homozygotes). Complete preservation of NF-200 in uncut contralateral nerve (lane 1) indicated degradation occurred only in vivo. Middle, western blot showing the expression level of the Wld protein in brain homogenates (detected by N70 antiserum). Bottom, same western blot probed with control monoclonal antibody β-tub 2.1 against β-tubulin. (b) Percentage of intact myelinated axons 2–4 mm distal to a sciatic nerve lesion. Preserved axons were counted using light microscopy in nerve segments immediately proximal to those used in (a). Counts of 93–99% in uncut contralateral nerves confirmed that observed degeneration occurred in vivo. n indicates total number of nerves counted. Mean ± s.e.m. of the middle three scores when n is odd; mean ± s.e.m. of the middle four when n is even. (c) Axon preservation as a function of the expression level of Wld protein. Axon preservation after 3 (black circle), 5 (white square) and 14 (cross) days as in (b); means ± s.e.m. where n > 1. Wld protein expression level is the signal quantified from (a), standardized against β-tubulin and expressed as a percentage of the expression in WldS homozygotes.

three days (Fig. 3a). Presence of the WldS phenotype in independent lines confirms that it is caused by the transgene rather than any integration effect. Thus, Ube4b/Nmnat is the Wld gene, and it protects both sensory and motor axons in Wld S and transgenic mutants. No alteration to Rbp7, Kif1b or any other gene is required to reproduce the phenotype of WldS mice. Functionally competent motor axons and synapses Next we tested whether axons were functionally as well as structurally preserved. Nerve–muscle preparations of flexor digitorum brevis (FDB) were isolated and stimulated 2–5 days after lesion. Axotomized wild-type muscles showed no response at these time points. In axotomized transgenics, however, as in the WldS mutant23, conduction of action potentials and synaptic transmission at neuromuscular junctions persisted for at least three days (Fig. 2). In a homozygous 4836 mouse, 80% (12/15) of muscle fibers responded to nerve stimulation three days after sciatic nerve section (for example, Fig. 2a). In another 4836 homozygote, functional innervation even after 5 days was indicated by 73% (22/30) of FDB muscle fibers showing spontaneous miniature endplate potentials (MEPP) and some responding to nerve stimulation (data not shown). Weaker, and sometimes more variable, responses indicating low quantal content were observed both in line 4836 hemizygotes (Fig. 2b) and homozygous 4830 mice. In the latter, 50% (12/24) and 30% (8/25) of fibers from two mice responded to stimulation two days after lesion. These weaker responses are in accord with the lower expression levels of the Wld protein (see below). Axotomized 4836 nerve terminals also recycled synaptic vesicles, visualized by activity-dependent staining with the styryl dye FM1-43, both upon incubating lumbrical muscles in a depolarizing solution (Fig. 2c and d) and upon stimulation of the distal stump of axotomized tibial nerve. Preservation of pre- and postsynaptic structures was confirmed using antibodies against NF165/SV2 and TRITC-α-bungarotoxin respectively (Fig. 2e and f). In two line 4836 homozygotes, 87–93% of endplates (n = 154 and 164) were occupied or partially occupied (8%) by nerve terminals 5 days after axotomy, similar to WldS mice23,24. Fewer than 50% (n = 187 and 210) were occupied in two 4836 hemizygotes after 3 days, and fewer still in 4830 homozygotes, again reflecting the lower transgene expression. nature neuroscience • volume 4 no 12 • december 2001

a

b

c

Thus, motor nerve conduction, synaptic transmission, vesicle recycling and motor nerve terminal morphology were preserved in a dose-dependent manner, with line 4836 showing a level of protection similar to WldS. Axons evidently persisted longer than functional motor nerve terminals, supporting the hypothesis that synaptic degeneration, at least in WldS mice, differs from that of axons and neuronal cell bodies25. Protection depends on Wld protein expression level The above experiments suggested that the degree of protection depends on the expression level of Wld protein. This was surprising because heterozygous WldS axons at short survival times degenerate only slightly faster than those of homozygotes5. We quantified Wld protein expression in each mutant strain to determine the level required to preserve cytoskeletal protein and axon structure for 3–5 days (Fig. 3). At lower expression levels, both measures of axon preservation indicated a strong dose-dependence. Axon counts at 3 days differed significantly between hemizygous 4836 and hemizygous 4830 mice (p < 0.001). Significant differences between homozygotes and hemizygotes of a single line (p < 0.01 for 4830; p < 0.001 for 4839) indicated that this is not a line-specific silencing effect26. Line 4839 hemizygotes were even indistinguishable from wild-type mice in neurofilament western blotting (NF-200 band, Fig. 3a), light microscopy (Fig. 3b) and electron microscopy (data not shown), despite expressing a small amount of the Wld protein (Fig. 3a). Thus, it is necessary for expression of Wld to reach a threshold level to exert a significant protective effect. To investigate whether the protective effect of Wld protein plateaus at higher expression levels, and to test axon protection in 1201


articles

Fig. 4. Axon protection 10–14 days after transection. (a) Western blot showing the extent of 200-kD neurofilament protein degradation in distal sciatic nerve 10–14 days after transection. Lane 1, 4836 homozygote after 14 days; lane 2, WldS homozygote after 14 days; lane 3, 4836 hemizygote after 14 days; lane 4, 4836 hemizygote after 10 days; lane 5, C57BL/6J after 12 days; lane 6, C57BL/6J unlesioned. (b–e) Light microscopic images (scale bar, 10 µm) of distal sciatic nerves following transection at a more proximal site 10–14 days earlier. (b) 4836 homozygote after 14 days (corresponding intact axon count, 73%), (c) WldS homozygote after 14 days (73%), (d) 4836 hemizygote after 10 days (35%), (e) C57BL/6J after 12 days (0%). Electron microscopy (data not shown) indicated that cytoskeleton and myelin was preserved in 4836 and WldS homozygotes as in Fig. 1.


a

b

d

c

e

transgenic mice under still more stringent conditions, we extended these studies to post-lesion times of 10–14 days. Fourteen days after transection, protection of axons in line 4836 homozygotes (69–73%) was as strong as that in WldS homozygotes (73–78%; Figs. 3c and 4), again indicating full reproduction of the WldS phenotype. Protection in line 4836 hemizygotes, however, was considerably weaker (35% after 10 days and only 7% after 14 days). Thus, the extent of axon protection differed far more between 4836 homo- and hemizygotes after 14 days than after 3–5 days, indicating that dose-dependence still operates at higher Wld expression levels. It follows that if Wld protein expression could be raised still further, an even stronger protective effect than in WldS and line 4836 mice could be achieved. Wld is a predominantly nuclear protein To determine whether the Wld protein could itself be the protective factor in axons, we determined its intracellular locaFig. 5. The intracellular location of the Wld protein. (a) Presence of the Wld protein in neuron nuclei of isocortex and (b) its absence from C57BL/6J control tissue. Red, anti-N70 antibody, which detects Wld and Ube4b. Green, anti-MAP2 antibody markedly outlining neuronal cell bodies. (c) Absence of Wld protein (red) in astrocytes (arrows, main picture). Cytoplasmic staining in ependymal cells (central channel) of WldS thoracic spinal cord could be Ube4b. Inset, absence in astrocytes and endothelial cells (arrow) of WldS isocortex. Green, antiGFAP. Blue, Hoechst Dye nuclear counterstain. (d–f) Confocal images of triangularis sterni muscle preparations immunostained with antiN70 antibody (green) and motor endplates counterstained with TRITC α-bungarotoxin (red). (d) 4836 homozygote, (e) WldS, (f) C57BL/6J. (g–i) Motor neurons in thoracic spinal cord of both WldS (g) and 4836 (h) expressed the Wld protein (red) in their nuclei, whereas those of C57BL/6J (i) did not. Cytoplasmic signals may be Ube4b or low level Wld protein. Counterstain, neurofilament (green). (j–l) Red channel (Wld protein plus Ube4b) images corresponding to (g–i). Scale bars, 5 µm (a, b), 10 µm (c), 50 µm (d–l).

1202

tion. An antiserum labeling the N-terminal 70 amino acids of Wld protein and Ube4b detected mainly a punctate nuclear pattern in WldS and transgenic neurons (Fig. 5a, c, g and h) and this signal was absent in wild-type neurons (Fig. 5b and i). There was no detectable signal either in motor axon terminals (Fig. 5d–f), which can be preserved by the Wld gene (Fig. 2e and f), or in sciatic nerve axons (Supplementary Fig. 2, available on the Nature Neuroscience web site). Swollen endbulbs of 24-hour transected central and peripheral nervous system axons, which accumulate other axonal proteins, also lacked Wld protein signal (data not shown). In addition, the antibody weakly labeled the normal Ube4b protein in cytoplasm of wild-type mice, so additional labeling of any small amount of cytoplasmic Wld protein might not be distinguishable. Although the WldS phenotype is intrinsic to neurons2,7,8, Wld protein was also detected as a punctate nuclear stain in other cell types in both WldS and transgenic mice (Fig. 5d and e). There was no sign of Wld protein in glial cells (Fig. 5c–e), although previous RT-PCR (reverse transcription-polymerase chain reaction) data indicate there may be a small amount in Schwann cells18. It is highly unlikely that expression in other cell types is required

a

d

c

b

e

f

g

h

i

j

k

l



articles


Fig. 6. NAD+ metabolism in WldS brain. (a) Increased Nmnat activity and (b) unaltered NAD+ content in WldS brain. Bar charts show Nmnat specific activity and NAD+ content in homogenates of fresh brain from homozygous WldS and C57BL/6J mice. Means and standard error are shown (n = 3).

a

b

to protect axons, but it is possible that the Wld protein confers as yet unknown properties on other cell types. For example, the delayed response of WldS muscles to denervation could have an intrinsic component27. We conclude that the Wld protein is predominantly located in the nucleus; thus, other factor(s) are likely to mediate the protective effect on axons. The Wld protein has Nmnat enzyme activity Sequence alignment with human NMNAT indicated that nucleotides 282–1140 of Wld span the entire Nmnat open reading frame. To detect any intrinsic Nmnat enzyme activity, we expressed protein recombinantly and measured the enzyme activity of the bacterial lysate. The observed specific activity for Nmnat (0.96 U/mg) was comparable with that of a bacterial lysate containing recombinant human NMNAT (1.74 U/mg)19. To determine whether there is a corresponding increase in Nmnat activity in WldS mice, we studied Nmnat activity in brain homogenates and found a fourfold increase in Wld S brain compared to C57BL/6J (p < 0.05; Fig. 6a). The total content of NAD+, however, was not significantly altered (p = 0.2; Fig. 6b). Therefore, the Wld protein confers an increase in Nmnat activity without altering the steady-state level of NAD+.

DISCUSSION We showed that the Ube4b/Nmnat chimeric gene is necessary and sufficient to protect injured axons for two weeks. We conclude that the Ube4b/Nmnat chimeric gene is the Wld gene, which encodes a unique neuroprotective factor for axons. It is important now to determine whether protection requires Ube4b sequences, Nmnat sequences, or both. The yeast homolog of Ube4b is required to multi-ubiquitinate proteins30 and a direct link between ubiquitination and axon degeneration comes from the Uch-l1 mutation in gracile axonal dystrophy 31. However, the Wld protein contains only 70 of 1,173 amino acids from Ube4b, and these are absent from the yeast homolog. They are therefore unlikely to confer multi-ubiquitination activity but may have a related role. The protective mechanism may be linked to the nuclear location of the Wld protein and perhaps to the non-homogeneous intranuclear distribution. Possibilities include sequestering of ubiquitination factors by protein–protein interactions and ubiquitination within the nucleus altering transcription factor stability or RNA processing, leading to an axon effect mediated by unknown proteins. Regulated nuclear transport of other ubiquitination factors can control ubiquitin-mediated degradation of nuclear substrates32. However, any Wld protein in the axon below the detection level of immunostaining could still have a direct protective role. Nmnat is a nuclear protein and the only known mammalian enzyme catalyzing the reaction NMN + ATP → NAD+ + PPi (ref. 20), a reaction generally assumed not to be at equilibrium because of the constitutive action of pyrophosphatases. Thus, the increase in Nmnat activity in WldS should increase NAD+ synthesis, and the maintenance of normal steady-state levels suggests that the putative additional NAD+ is metabolized. The product of a compensatory reaction could itself be involved in nature neuroscience • volume 4 no 12 • december 2001

axon protection. For example, poly-ADP ribosylation uses NAD+, influencing protein activity and cellular NAD+ and ATP content, especially in response to stress33,34. Mild activation of PARP without NAD+ depletion can be neuroprotective35. Another metabolite, NADPH, is a coenzyme for nitric oxide synthase, an enzyme linked to axon damage36, and synthesis of the signaling molecule cyclic ADP ribose from NAD+ regulates calcium release from intracellular stores37, potentially influencing calcium activated proteases in Wallerian degeneration. Both spontaneous and transgenic WldS mice could be used to investigate the function of each parent gene. We already show that overexpression of Nmnat activity causes no overt phenotype, and report an in vivo mutation of a mammalian E4 ubiquitination factor. Despite the critical role played by the ubiquitin-proteasome pathway in neurological disease and many other processes, we know remarkably little about the function of such proteins. Identification of the Wld gene facilitates studies to determine whether it protects axons in clinically relevant situations. The Wld S mutation is already known to protect neuronal processes in vitro from the toxic effects of vincristine13, indicating that traumatic and toxic axon death share a common final pathway. Current studies indicate that distal axon loss in myelin protein zero knockout mutants38 is rescued by WldS (M. Samsam and R. Martini, unpublished data), and that WldS protects in a mouse model of motoneuron disease (A. Ferri and A.C. Kato, unpublished data). Studies of WldS in diverse neurological diseases is facilitated by our identification of the Wld gene and by recent protocols for tracking the inheritance of WldS in crosses with neurological disease mutants39. Models of common and complex neurological disorders such as multiple sclerosis and diabetic neuropathy can now be investigated through the development of viral vectors for Wld and generation of transgenic WldS rats. Mutational analysis in human neurological disorders with a homologous chromosomal location, such as hereditary Parkinsonism 40 , also becomes possible. Delayed Wallerian degeneration also alters the glial response to injury, as in a mouse model of spinal cord injury, where it delays inflammatory cell and astrocytic responses9,41. It may be that this information could be used to optimize tissue destruction and repair processes. We conclude that the Wld gene is a chimera of Ube4b and Nmnat encoding a predominantly nuclear protein in neurons, and we propose that other factors may mediate the protective effect on the axon. Axon protection is strongly dose-dependent and pyridine nucleotide metabolism is altered in the WldS mouse. These findings open the way to a molecular understanding of Wallerian degeneration and to much-needed neuroprotective strategies that target not only the cell body, but also the axons and synaptic terminals. 1203


articles


METHODS Generation of transgenic mice. Ube4b/Nmnat cDNA was RT-PCR amplified from WldS brain using Platinum Pfx polymerase (Life Technologies, Karlsruhe, Germany) and cloned downstream of the β-actin promoter in pHβAPr-1 (ref. 42; Supplementary Fig. 3). Wld protein was thus expressed in neurons, where it has an intrinsic effect2,7,8, and other cell types, where WldS mice also express it (Fig. 5). A 6-kb fragment containing promoter, cDNA and polyadenylation signal was released using NdeI and EcoRI and gel-purified using QIAquick extraction (Qiagen, Hilden, Germany). Pronuclear injection into CBA X C57 F1 single-cell embryos (G. Kollias, Vari, Greece) resulted in nine founders from 62 pups. Genotyping of transgenic mice. DNA was prepared from a 5-mm tail biopsy using the Nucleon HT kit (Amersham Pharmacia, Freiburg, Germany). The 1.1-kb transgene coding region was detected using alkaline Southern blotting of a BamHI/HindIII double digest on Hybond N+ (Amersham Pharmacia) and hybridization with a corresponding 32P-labeled probe. Sciatic nerve lesion. Six- to eleven-week-old mice were anesthetized intraperitoneally with Ketanest (100 mg/kg; Bayer, Leverkusen, Germany) and Rompun (5 mg/kg; Parke Davis/Pfizer, Karlsruhe, Germany). Right sciatic nerves (upper thigh) were transected and the wounds were closed with single sutures. Two to fourteen days later, mice were killed, the swollen first 2 mm of the distal nerve was discarded, the next 2 mm was used for light and electron microscopy, and a segment 4–10 mm distal to the lesion site was used in western blotting. Further distal nerves and muscles were used for electrophysiology. Light and electron microscopy. Nerve segments were fixed for 1–3 days in fresh half-strength Karnovsky’s fixation (4% paraformaldehyde, 2% glutaraldehyde in 0.1 M sodium cacodylate, pH 7.3; ref. 43), extensively buffer-rinsed and osmicated for 4 h with 1% OsO4 in 0.1 M cacodylate. Samples were taken through a graded ethanol series including a uranylic acetate en bloc staining step overnight in 70% ethanol. Before infiltration with Araldite Cy212 epoxy resin (Serva, Heidelberg, Germany), propylene oxide was used as intermedium. Tissue blocks were cured for 60 h at 60°C. Semithin (0.5 µm) and thin (60 nm) cross-sections were taken on a Reichert Ultracut UCT ultramicrotome. Semithin sections for light microscopy were stained with methylene blue and thin sections, with 1% aqueous uranylic acetate (20 min), and sections were counterstained with Reynold’s lead citrate (7 min)44. Thin sections were mounted on 150 mesh Formvar coated copper grids and examined with a Zeiss EM 902 electron microscope at 80 kV acceleration voltage. Neuromuscular junction electrophysiology and morphology. FDB and lumbrical muscles and contralateral controls were removed in Cologne and placed in cold physiological saline (137 mM Na+, 4 mM K+, 2 mM Ca 2+ , 1 mM Mg 2+ , 147 mM Cl – , 5 mM glucose, 5 mM HEPES, pH 7.2–7.4, equilibrated with air or 100% oxygen). Electrophysiological experiments were done later the same day in Edinburgh, following transfer to a medium containing similar concentrations of Na+, K+, Ca2+ and Mg2+, plus 23 mM HCO3–, 2 mM H2PO4–, equilibrated with 95% O2/5% CO2. MEPPs and evoked synaptic responses to tibial nerve stimulation (EPPs) were recorded from FDB using an intracellular glass microelectrode and analyzed using WinWCP software45 (J. Dempster, University of Strathclyde). Recycled synaptic vesicles of motor nerve terminals were stained in lumbrical muscles using FM1-43 (Molecular Probes, Leiden, Netherlands) with 20 Hz nerve stimulation or depolarizing physiological solutions, and acetylcholine receptors subsequently stained with TRITCα-bungarotoxin (Molecular Probes)46,47. Endplates and terminals were examined in a Nikon fluorescence microscope (Kingston-upon-Thames, UK) using respectively a standard rhodamine filter cube and a customized cube with a 435 nm excitation filter, 455 nm dichroic mirror and a 10 nm bandpass 515 nm emission filter46. Conventional immunocytochemical and fluorescent bungarotoxin probes were used for structural analysis. Muscle preparations, fixed for 60 min in 0.1 M PBS, 4% paraformaldehyde, were incubated in TRITC-α1204

bungarotoxin (5 µg/ml; 30 min) followed by overnight primary antibody (1:200 monoclonal anti-165 kDa neurofilament plus anti-synaptic vesicle antigen SV2, Developmental Studies Hybridoma Bank, University of Iowa; or affinity purified rabbit polyclonal N70; see below). Secondary antibodies (1:200 dilution; 4 h) were FITC-conjugated sheep-anti-mouse IgG or anti-rabbit IgG (Diagnostics Scotland). Preparations were mounted in Vektashield (Vector Labs, Burlingame, California) and viewed in a fluorescence microscope using a Leitz 50× water immersion objective lens (NA 1.00; Wetzlar, Germany). Most images were captured with a Hamamatsu C5810 chilled color CCD camera and acquired using Openlab software (Improvision, Coventry, UK). Confocal images were obtained using a BioRad Radiance 2000 system (Hemel Hempstead, UK). Western blotting. We analyzed Wld protein expression level in mouse brains homogenized in two volumes of 20 mM HEPES (pH 7.5), 0.2 M CaCl2, 0.2 M MgSO4, 1 ml/20 g tissue protease inhibitor cocktail (Sigma, Taufkirchen, Germany) and 1 mg/ml DNase (Sigma). Cytoskeletal protein preservation was determined in lesioned sciatic nerves homogenized in 20 volumes of this buffer. Proteins were separated using standard SDS-PAGE and semi-dry blotted onto nitrocellulose. Loading and transfer were checked using Ponceau S (Sigma) and Coomassie Blue. Primary antibodies were applied (overnight, 4°C) followed by horseradish peroxidase-coupled secondary antibody (1 h, room temperature; goat-anti-mouse 1:3,000, goat-anti-rabbit 1:5,000; Dianova, Hamburg, Germany) and detection using enhanced chemiluminescence (Amersham Pharmacia). Chimeric protein expression was quantified using affinity-purified N70 antibody (below) and β-tub 2.1 (Sigma) control and Quantity One software (BioRad). Cytoskeletal protein degradation was analyzed using phosphate-independent monoclonal N52 (1:2,000; Sigma) against heavy neurofilament protein. Morphological quantification of axon preservation. We counted 5001500 myelinated axons in randomly chosen fields 2–4 mm distal to a sciatic nerve lesion in transverse semithin sections on a Zeiss Axiophot microscope (Göttingen, Germany) coupled to a digital camera. Survival criteria were normal myelin sheaths, uniform axoplasm and intact mitochondria, supported by electron microscopy spotchecks. Scoring was documented with Meta Imaging software (Universal Imaging Corporation, Downingtown, Pennsylvania). Standard errors of the mean and ttests were calculated using SPSS for Windows 10.0. Cloning and expression of recombinant proteins. Constructs were generated to express in bacteria the N-terminal 70 amino acids (N70) of the chimeric protein and full-length chimeric protein. Inserts were PCRamplified from transgene construct using Pfx polymerase (Life Technologies). Primers for N70 were5´-GACTAGCTAGCATGGAGGAGCTGA GCGCTGAC-3´ and 5´-ATCCGCTCGAGCTAGTCTGCTGCACCTATG GGGGA-3´. For full-length chimeric cDNA, the second primer was replaced by 5´CGCCTCGAGTCACAGAGTGGAATGGTTGTGC-3´. Products were ligated into NheI/XhoI double-digested pET-28b (+) vector (Novagen, Schwalbach, Germany) and transformed into XL-10 Gold (Stratagene, Amsterdam, Netherlands). Plasmids isolated using the Plasmid Mini Kit (Qiagen) were retransformed into BL21 (DE3) (Novagen). Protein expression was induced with 1 mM IPTG for 3 h at OD600 = 0.8. For analysis of Nmnat activity, cells were lysed by sonication. Generation of polyclonal antisera. The N70 bacterial pellet was resuspended in native binding buffer (20 mM sodium phosphate, 500 mM sodium chloride pH 7.8,100 µg/ml egg white lysozyme), sonicated on ice (6 × 15 s with 15-s intervals) and centrifuged (3,000 g, 15 min). N70 was purified using a ProBond column (Invitrogen, Groningen, Netherlands) and concentrated using a YM-3 Centricon centrifugal filter (Millipore, Bedford, Massachusetts). Antisera were raised by intradermal immunization of two SPF-rabbits by Eurogentec (Seraing, Belgium) with boosts at days 14, 28 and 56, and a final bleed at 80 days. For affinity purification 500 µg N70 protein bound to ProBond resin (2 ml wet volume) was blocked with 5% (w/v) dried skimmed milk powder plus 1% (w/v) BSA in native binding buffer. Resin was incubated with crude antiserum (2 ml in 8 ml binding buffer), and washed with 10 nature neuroscience • volume 4 no 12 • december 2001


articles


bed volumes binding buffer. Specific antibodies, binding only Wld protein and Ube4b, were eluted with 100 mM ethanolamine (pH 11.5), neutralized with 1.5 M Tris and dialyzed against PBS. Immunocytochemistry of spinal cord, nerves and brain. Perfusionfixed tissues were fixed for a further four hours in 4% paraformaldehyde, 0.1 M PBS before embedding in paraffin. Paraffin sections on poly-L-lysine coated slides were dewaxed, washed in PBS, rinsed in citrate buffer overnight (60°C) and incubated with 0.1% Triton X-100 (Sigma). After blocking (5% BSA in PBS), the following primary antibodies were applied (overnight, 4°C): affinity purified N70 (above), plus monoclonal anti-MAP2 (Sigma), mouse polyclonal anti-GFAP (Sigma) or monoclonal anti-neurofilament (Biogenex, San Ramon, California). Secondary antibodies (1 h, room temperature) were either Cy3-labeled goat-anti-rabbit (Dako, Hamburg, Germany) or Texas Red-labeled goat-anti-rabbit (Molecular Probes) together with Cy2labeled goat-anti-mouse (both Dako), diluted according to the manufacturers’ instructions. Determination of NMNAT activity. Mice were killed, and whole brains were immediately removed and cut into two equal hemispheres for NMNAT assay and NAD+ determination, respectively. Tissue was suspended in 3 volumes of 100 mM Tris-HCl, pH 7.4, 0.5 mM EDTA, 1 mM MgCl2, 1 mM DTT, 1 mM PMSF and homogenized on ice (3 × 3 s with 10-s intervals) using an Ultra-Turrax homogenizer at medium speed. NMNAT activity was determined using a reaction mixture of 40 mM Tris-HCl pH 7.5, 5 mM nicotinamide mononucleotide (NMN), 3.4 mM ATP, 18 mM MgCl2, 10 mM NaF, and brain homogenate (final volume of 0.5 ml). The reaction was started by the addition of NMN at 37°C and stopped after 10–40 min by adding 100 µl of assay mixture to 50 µl of ice-cold 1.2 M HClO4. After 10 min, at 0°C the mixture was centrifuged and 130 µl of supernatant was neutralized by addition of 35 µl 0.8-M K2CO3. NMNAT activity was calculated after HPLC identification and quantification of NAD+ produced48. One unit of enzyme activity catalyses the synthesis of 1 µmol of NAD+ per minute at 37°C. Determination of NAD+ content. A brain hemisphere in 2 volumes 7.5% ice-cold HClO4 was homogenized using an Ultra-Turrax (4 × 4 s, with 20 s intervals). After 15 min at 0°C, the suspension was centrifuged at 16,000 g for 1 min. The pH was adjusted to 6.0 using 0.8 M K2CO3. After centrifugation, the extract was analyzed by HPLC48. Animal experimentation was approved by Stadt Köln Veterinäramt, licence K13,11//00. GenBank accession numbers. The GenBank accession number for NMNAT is AF312734; for Wld, AF260924. Note: Supplementary figures are available on the Nature Neuroscience web site (http://neuroscience.nature.com/web_specials).

ACKNOWLEDGEMENTS We thank E. Janssen, C. Hoffmann (Department of Anatomy I, University of Cologne), F. Carnevali, F. Pierella (University of Ancona), S. Fearn and M. Botham (University of Southampton) for technical assistance, T. Vogt for supplying pHβAPr-1 plasmid and R. Martini for critically reading the manuscript. This work was supported by the Federal Ministry of Education and Research (FKZ: 01 KS 9502) and Center for Molecular Medicine, University of Cologne (ZMMK) (T.G.A.M., W.M., D.W., M.P.C.), a Wellcome Trust Biomedical Collaboration Grant (R.R.R., M.P.C.) and Consiglio Nazionale delle Ricerche Target Project “Biotechnology” (M.E., G.M.).

RECEIVED 27 SEPTEMBER; ACCEPTED 31 OCTOBER 2001 1. Waller, A. Experiments on the section of glossopharyngeal and hypoglossal nerves of the frog and observations of the alternatives produced thereby in the structure of their primitive fibres. Phil. Trans. R. Soc. Lond. 140, 423–429 (1850).


2. Deckwerth, T. L. & Johnson, E. M. Jr. Neurites can remain viable after destruction of the neuronal soma by programmed cell death (apoptosis). Dev. Biol. 165, 63–72 (1994). 3. Finn, J. T. et al. Evidence that Wallerian degeneration and localized axon degeneration induced by local neurotrophin deprivation do not involve caspases. J. Neurosci. 20, 1333–1341 (2000). 4. Lunn, E. R., Perry, V. H., Brown, M. C., Rosen, H. & Gordon, S. Absence of Wallerian degeneration does not hinder regeneration in peripheral nerve. Eur. J. Neurosci. 1, 27–33 (1989). 5. Perry, V. H., Lunn, E. R., Brown, M. C., Cahusac, S. & Gordon, S. Evidence that the rate of Wallerian degeneration is controlled by a single autosomal dominant gene. Eur. J. Neurosci. 2, 408–413 (1990). 6. Perry, V. H., Brown, M. C. & Lunn, E. R. Very slow retrograde and Wallerian degeneration in the CNS of C57BL/Ola mice. Eur. J. Neurosci. 3, 102–105 (1991). 7. Glass, J. D., Brushart, T. M., George, E. B. & Griffin, J. W. Prolonged survival of transected nerve fibres in C57BL/Ola mice is an intrinsic characteristic of the axon. J. Neurocytol. 22, 311–321 (1993). 8. Buckmaster, E. A., Perry, V. H. & Brown, M. C. The rate of Wallerian degeneration in cultured neurons from wild type and C57BL/WldS mice depends on time in culture and may be extended in the presence of elevated K+ levels. Eur. J. Neurosci. 7, 1596–1602 (1995). 9. Zhang, Z., Fujiki, M., Guth, L. & Steward, O. Genetic influences on cellular reactions to spinal cord injury: a wound-healing response present in normal mice is impaired in mice carrying a mutation (WldS) that causes delayed Wallerian degeneration. J. Comp. Neurol. 371, 485–495 (1996). 10. Dal Canto, M. C. & Gurney, M. E. Neuropathological changes in two lines of mice carrying a transgene for mutant human Cu, Zn SOD, and in mice overexpressing wild type human SOD: a model of familial amyotrophic lateral sclerosis (FALS). Brain Res. 676, 25–40 (1995). 11. Perry, V.H. & Anthony, D.C. Axon damage and repair in multiple sclerosis. Phil. Trans. R. Soc. Lond. B Biol. Sci. 354, 1641–1647 (1999). 12. Glynn, P. Neural development and neurodegeneration: two faces of neuropathy target esterase. Prog. Neurobiol. 61, 61–74 (2000). 13. Wang, M. S., Wu, Y., Culver, D. G. & Glass, J. D. The gene for slow Wallerian degeneration (WldS) is also protective against vincristine neuropathy. Neurobiol. Dis. 8, 155–161 (2001). 14. Sagot, Y. et al. Bcl-2 overexpression prevents motoneuron cell body loss but not axonal degeneration in a mouse model of a neurodegenerative disease. J. Neurosci. 15, 7727–7733 (1995). 15. Houseweart, M. K. & Cleveland, D. W. Bcl-2 overexpression does not protect neurons from mutant neurofilament-mediated motor neuron degeneration. J. Neurosci. 19, 6446–6456 (1999). 16. Coleman, M. P. et al. An 85-kb tandem triplication in the slow Wallerian degeneration (WldS) mouse. Proc. Natl. Acad. Sci. USA 95, 9985–9990 (1998). 17. Lyon, M. F., Ogunkolade, B. W., Brown, M. C., Atherton, D. J. & Perry, V. H. A gene affecting Wallerian nerve degeneration maps distally on mouse chromosome 4. Proc. Natl. Acad. Sci. USA 90, 9717–9720 (1993). 18. Conforti, L. et al. A Ufd2/D4Cole1e chimeric protein and overexpression of Rbp7 in the slow Wallerian degeneration (WldS) mouse. Proc. Natl. Acad. Sci. USA 97, 11377–11382 (2000). 19. Emanuelli, M. et al. Human NMN adenylyltransferase: molecular cloning, chromosomal localization, tissue mRNA levels, bacterial expression, and enzymatic properties. J. Biol. Chem. 276, 406–412 (2001). 20. Magni, G., Amici, A., Emanuelli, M., Raffaelli, N. & Ruggieri, S. Enzymology of NAD+ synthesis. Adv. Enzymol. Relat. Areas Mol. Biol. 73, 135–182 (1999). 21. Conforti, L. et al. The major brain isoform of Kif1b lacks the putative mitochondria-binding domain. Mamm. Genome 10, 617–622 (1999). 22. Zhao, C. et al. Charcot-Marie-Tooth disease type 2a caused by mutation in a microtubule motor kif1bbeta. Cell 105, 587–597 (2001). 23. Ribchester, R. R. et al. Persistence of neuromuscular junctions after axotomy in mice with slow Wallerian degeneration (C57BL/WldS). Eur. J. Neurosci. 7, 1641–1650 (1995). 24. Gillingwater, T. H. & Ribchester, R. R. Age-dependent synapse withdrawal at axotomised neuromuscular junctions in WldS mutant mice. J. Physiol. (Lond.) 523P, 53P (2000). 25. Gillingwater, T. H. & Ribchester, R. R. Compartmental neurodegeneration and synaptic plasticity in the WldS mutant mouse. J. Physiol. (Lond.) 534, 627–639 (2001). 26. Feng, G. et al. Imaging neuronal subsets in transgenic mice expressing multiple spectral variants of GFP. Neuron 28, 41–51 (2000). 27. Brown, M. C., Booth, C. M., Lunn, E. R. & Perry, V. H. Delayed response to denervation in muscles of C57BL/Ola mice. Neuroscience 43, 279–283 (1991). 28. Sagot, Y., Tan, S. A., Hammang, J. P., Aebischer, P. & Kato, A. C. GDNF slows loss of motoneurons but not axonal degeneration or premature death of pmn/pmn mice. J. Neurosci. 16, 2335–2341 (1996). 29. Burne, J. F., Staple, J. K. & Raff, M. C. Glial cells are increased proportionally in transgenic optic nerves with increased numbers of axons. J. Neurosci. 16, 2064–2073 (1996). 30. Koegl, M. et al. A novel ubiquitination factor, E4, is involved in multiubiquitin chain assembly. Cell 96, 635–644 (1999). 31. Saigoh, K. et al. Intragenic deletion in the gene encoding ubiquitin carboxyterminal hydrolase in gad mice. Nat. Genet. 23, 47–51 (1999).

1205



articles

32. Hamilton, M. H., Tcherepanova, I., Huibregtse, J. M. & McDonnell, D. P. Nuclear import/export of hRPF1/Nedd4 regulates the ubiquitin-dependent degradation of its nuclear substrates. J. Biol. Chem. 276, 26324–26331 (2001). 33. Smith, S. The world according to PARP. Trends Biochem. Sci. 26, 174–179 (2001). 34. Ha, H.C. & Snyder, S.H. Poly(ADP-ribose) polymerase is a mediator of necrotic death by ATP depletion. Proc. Natl. Acad. Sci. USA 96, 13978–13982 (1999). 35. Nagayama, T. et al. Activation of poly(ADP-ribose) polymerase in the rat hippocampus may contribute to cellular recovery following sublethal transient global ischemia. J. Neurochem. 74, 1636–1645 (2000). 36. Smith, K. J., Kapoor, R., Hall, S. M. & Davies, M. Electrically active axons degenerate when exposed to nitric oxide. Ann. Neurol. 49, 470–476 (2001). 37. Di Lisa, F. & Ziegler, M. Pathophysiological relevance of mitochondria in NAD(+) metabolism. FEBS Lett. 492, 4–8 (2001). 38. Frei, R. et al. Loss of distal axons and sensory Merkel cells and features indicative of muscle denervation in hindlimbs of P0-deficient mice. J. Neurosci. 19, 6058–6067 (1999). 39. Mi, W., Conforti, L. & Coleman, M. P. Genotyping methods to detect a unique neuroprotective factor for axons (WldS). J. Neurosci. Methods (in press). 40. Valente, E.M. et al. Localization of a novel locus for autosomal recessive earlyonset parkinsonism, PARK6, on human chromosome 1p35–p36. Am. J. Hum. Genet. 68, 895–900 (2001).

1206

41. Fujiki, M., Zhang, Z., Guth, L. & Steward, O. Genetic influences on cellular reactions to spinal cord injury: activation of macrophages/microglia and astrocytes is delayed in mice carrying a mutation (WldS) that causes delayed Wallerian degeneration. J. Comp. Neurol. 371, 469–484 (1996). 42. Gunning, P., Leavitt, J., Muscat, G., Ng, S. Y. & Kedes, L. A human beta-actin expression vector system directs high-level accumulation of antisense transcripts. Proc. Natl. Acad. Sci. USA 84, 4831–4835 (1987). 43. Karnovsky, M. J. A formaldehyde-glutaraldehyde fixative of high osmolarity for use in electron microscopy. J. Cell Biol. 27, 137A (1965). 44. Reynolds, E. E. The use of lead citrate at high pH as an electron opaque stain in electron microscopy. J. Cell Biol. 17, 208 (1963). 45. Costanzo, E. M., Barry, J. A. & Ribchester, R. R. Co-regulation of synaptic efficacy at stable polyneuronally innervated neuromuscular junctions in reinnervated rat muscle. J. Physiol. (Lond.) 521, 365–374 (1999). 46. Costanzo, E. M., Barry, J. A. & Ribchester, R. R. Competition at silent synapses in reinnervated skeletal muscle. Nat. Neurosci. 3, 694–700 (2000). 47. Ribchester, R. R., Mao, F. & Betz, W. J. Optical measurements of activitydependent membrane recycling in motor nerve terminals of mammalian skeletal muscle. Proc. R. Soc. Lond. B Biol. Sci. 255, 61–66 (1994). 48. Balducci, E. et al. Assay methods for nicotinamide mononucleotide adenylyltransferase of wide applicability. Anal. Biochem. 228, 64–68 (1995).



articles


GABAB receptor activation enhances mGluR-mediated responses at cerebellar excitatory synapses Moritoshi Hirono1, Tohru Yoshioka1,2 and Shiro Konishi3 1 Department of Molecular Neurobiology, Advanced Research Institute for Science and Engineering, Waseda University, Tokyo 169-8555, Japan 2 Department of Molecular Neurobiology, School of Human Sciences, Waseda University, Tokorozawa 359-1192, Japan 3 Laboratory of Molecular Neurobiology, Mitsubishi Kagaku Institute of Life Sciences and CREST (JST), 11-Minamiooya, Machida-shi, Tokyo 194-8511, Japan

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

Published online: 12 November 2001, DOI: 10.1038/nn764 Metabotropic γ-aminobutyric acid type B (GABAB) and glutamate receptors (mGluRs) are postsynaptically co-expressed at cerebellar parallel fiber (PF)–Purkinje cell (PC) excitatory synapses, but their functional interactions are unclear. We found that mGluR1 agonist-induced currents and [Ca2+]i increases in PCs were enhanced following co-activation of GABAB receptors. A GABAB antagonist and a G-protein uncoupler suppressed these effects. Low-concentration baclofen, a GABAB agonist, augmented mGluR1-mediated excitatory synaptic current produced by stimulating PFs. These results indicate that postsynaptic GABAB receptors functionally interact with mGluR1 and enhance mGluR1mediated excitatory transmission at PF–PC synapses. The interaction between the two types of metabotropic receptors provides a likely mechanism for regulating cerebellar synaptic plasticity.

GABA is the main inhibitory neurotransmitter in the central nervous system, and inhibitory GABAergic synapses are endowed with transmitter receptors including ionotropic GABAA and GABA C receptors and metabotropic GABA B receptors (GABABRs)1. Ionotropic GABA receptors exhibit considerable molecular diversity2, whereas recent cloning has revealed that the GABA BR gene encodes R1a and R2 subunits that form heterodimers to function3–6. GABABRs are coupled with Gi/o proteins either to inhibit neurotransmission presynaptically7,8 or to decrease the excitability postsynaptically by opening G-proteincoupled inwardly rectifying K + (GIRK) channels 9,10 . The GABABR-mediated Gi/o protein activation also depresses adenylyl cyclase and reduces Ca2+ channel currents11,12. However, correlations between these GABABR-mediated pharmacological actions and synaptic events are not completely understood. In the cerebellar cortex, radioautographic and immunocytochemical studies have shown that GABABR binding sites densely occur in the molecular layer, where the PCs extend their dendritic branches 13,14 . The postsynaptic localization of GABABRs on PCs was reported by in situ hybridization3–6. An electron microscopy study showed that GABABRs are present at the extra-postsynaptic sites of excitatory connections between parallel fibers (PFs) and PCs5,15. GABABRs also suppress a synaptic process called rebound potentiation of inhibitory transmission following PC depolarization 16 . However, it remains uncertain what physiological role the GABA BRs have at the PF–PC excitatory postsynaptic sites. Long-term depression (LTD) at PF–PC synapses has been proposed as a cellular mechanism of synaptic plasticity closely associated with motor learning17,18. Induction of LTD requires activation of type 1 metabotropic glutamate receptors (mGluR1), triggering a signaling cascade that nature neuroscience • volume 4 no 12 • december 2001

includes Gq protein-mediated activation of phospholipase C (PLC), hydrolysis of phosphatidylinositol-4,5-bisphosphate (PIP2) to inositol 1,4,5-triphosphate (IP3) and diacylglycerol, leading to intracellular Ca2+ concentration ([Ca2+]i) increase and protein kinase C (PKC) stimulation18–20. Interestingly, cellular localization of mGluR1 is similar to that of GABABRs: morphological studies have shown that mGluR1 is also expressed abundantly at the extra-postsynaptic sites of PF–PC synapses21–23. It is, therefore, reasonable to expect that GABABRs would interact with mGluR1 and modulate mGluR1-mediated physiological responses at these synaptic sites. Therefore, the aim of the present study was to explore the cross-talk between mGluR1 and GABABRs expressed by PCs in acute slices from the mouse cerebellum, using whole-cell recordings combined with Ca2+-signal imaging. We found that the activation of GABABRs by the exogenous agonist baclofen enhanced both mGluR1-mediated inward currents and Ca2+ signals in PCs. More importantly, we showed that endogenous GABA released by electrical stimulation in the cerebellar cortex mimicked the effect of the GABABR agonist, which augmented the mGluR1mediated slow excitatory synaptic current elicited by PF stimulation. Therefore, the cross-talk between mGluR1 and GABABR revealed in this study seems to call for a revision of our view that GABAB receptors serve an exclusively inhibitory role in chemical signaling at central synapses.

RESULTS GABAB activation enhanced mGluR current Iontophoretic application of the nonselective mGluR agonist 1S,3R-ACPD (1S,3R-1-aminocyclopentane-1,3-dicarboxylic acid) produced an inward current in cerebellar PCs voltage-clamped 1207


articles

a


b

Fig. 1. Enhancement of 1S,3R-ACPD-induced inward current by GABA. (a) Inward current responses were induced by iontophoretic application of 1S,3R-ACPD to a Purkinje cell at a constant interval of 40 s. GABA (100 µM) was applied by perfusion for 3 min as indicated by horizontal bar. The record was obtained from a Purkinje cell held at –60 mV in the presence of bicuculline (30 µM) and TTX (0.5 µM). (b) 1S,3RACPD-induced responses indicated by 1, 2 and 3 in (a) are displayed on a fast time base.

at –60 mV, which is close to the resting potential24–26. The 1S,3RACPD-induced current was suppressed by (S)-4-carboxyphenylglycine (4CPG), a selective antagonist for group I mGluRs including mGluR1 and mGluR5 (ref. 26), whereas 1S,3R-ACPD did not produce any detectable current in PCs of mGluR1-deficient mice19, which indicates that the inward current resulted from activation of mGluR1. This mGluR1-mediated current increased markedly when GABA, an inhibitory transmitter substance, was co-applied by superfusion (Fig. 1). The increase of the 1S,3R-ACPD-current amplitude following 100 µM GABA co-application was 161 ± 8% of the control response (n = 4) in an artificial cerebrospinal fluid (ACSF) that contained tetrodotoxin (TTX, 0.5 µM) and the GABAA receptor

a

d

b e

c

1208

antagonist bicuculline (30 µM). Therefore, it is likely that the enhancement of the 1S,3R-ACPD-induced current in the presence of GABA is mediated by GABABRs but not by GABAA receptors in PCs, which is consistent with the slow outward current observed during GABA application. We then examined whether the GABABR agonist baclofen reproduces the augmentation of the 1S,3R-ACPD response. When applied by perfusion, baclofen (3 µM) increased the amplitude of the 1S,3RACPD-induced current up to 226 ± 11% of the control response (n = 15; Fig. 2). To test whether the GABABR-induced enhancement is specific to the mGluR1-mediated response, we compared the effects of baclofen on inward current responses produced by 1S,3RACPD and AMPA (α-amino-3-hydroxy-5-methyl-4-isoxasole propionic acid), an ionotropic GluR agonist. Baclofen selectively enhanced the 1S,3R-ACPD response without appreciably affecting the AMPA-induced current (Fig. 2a–c). Baclofen also produced an outward current (36 ± 4 pA, n = 15), as observed with GABA application. The baclofen-induced enhancement of the 1S,3R-ACPD response was reversible, subsiding over a period of approximately 20 minutes after the agonist was washed out. In a current-clamp mode, baclofen hyperpolarized PCs by –2.1 ± 2.2 mV (n = 4, Fig. 2d) and increased the amplitude of 1S,3R-ACPD-induced depolarization from 10.6 ± 0.7 mV to 14.8 ± 1.3 mV (Fig. 2e; n = 4, p < 0.05). The baclofen-induced enhancement of the mGluR1-mediated response was completely abolished by the selective GABABR antagonist CGP62349 (Fig. 3a) with a half-maximal inhibitory concentration IC50 of about 15 nM. The 1S,3R-ACPD-induced current remained almost unchanged in the presence of 100 nM CGP62349 (97 ± 2% of the control current, n = 4), which indicates that tonic activation of GABABRs by endogenous GABA released spontaneously is insufficient for a detectable enhancement of the mGluR1-mediated response. We then investigated whether activation of G-protein-coupled receptors other than GABABRs also enhances the mGluR1mediated inward current in PCs. The 1S,3R-ACPD-induced Fig. 2. Selective enhancement of mGluR1-mediated current following GABABR activation. (a) 1S,3R-ACPD- and AMPA-current responses were alternately induced by iontophoretic applications from separate microelectrodes to a single Purkinje cell. The GABABR agonist baclofen (3 µM) was applied by perfusion during the period indicated by the horizontal bar. (b) Sequential 1S,3R-ACPD- and AMPA-induced currents indicated as 1–3 in (a) are displayed on a fast time base. (c) Time course of the effect of GABABR activation on the 1S,3R-ACPD- and AMPA-induced current responses. The amplitude of both responses is given as a percentage of the control immediately before baclofen application. (d) A current-clamp recording of 1S,3RACPD-induced depolarization in a Purkinje cell and the effects of baclofen (3 µM) on the 1S,3RACPD-response and membrane potential. (e) Expanded and superimposed records of 1S,3R-ACPDinduced depolarizations before and during baclofen application. Each response was recorded as indicated by 1 and 2 in (d). nature neuroscience • volume 4 no 12 • december 2001



articles

Fig. 3. Blockade of baclofen-induced enhancement of the mGluR1mediated current response by the GABABR antagonist CGP62349. (a) 1S,3R-ACPD was iontophoretically applied at a time point indicated by arrows at a constant interval in the presence of CGP62349 (30 nM). Responses were recorded from a single PC before (left) and during perfusion of baclofen (3 µM, middle) and after washing out the GABABR agonist (right). (b) Comparison of the effects of Gi/o-coupled receptor agonists on the 1S,3R-ACPD-induced current response in PCs. The number in parentheses represents the number of PCs in which the effects of individual agonists were determined: baclofen (3 µM), serotonin (5-HT, 30 µM), carbachol (CCh, 100 µM) and adenosine (10 µM). ***P < 0.001, one-way ANOVA with Tukey’s post-test.

a

b

current was not significantly altered by application of carbachol (CCh), serotonin (5-HT) or adenosine (Fig. 3b). However, muscarinic, 5-HT and adenosine receptors occur in the cerebellar cortex27–29, and 5-HT and adenosine elicit facilitation and inhibition of cerebellar GABAergic transmission, respectively, when applied by superfusion, as in this study30,31. It seems, therefore, that the enhancement of the mGluR1-mediated response requires selective activation of GABABRs in PCs. Characterization of GABABR-mediated enhancement First, the current–voltage relationship of the 1S,3R-ACPDinduced response was compared before and during GABABR activation. The 1S,3R-ACPD-induced current was obtained by subtracting current responses produced by a constant voltage ramp in the absence and presence of 1S,3R-ACPD (Fig. 4a and b). Baclofen increased the 1S,3R-ACPD-induced current, whereas its reversal potential was almost identical before and after baclofen application (–5.1 ± 4.2 mV and –2.8 ± 3.6 mV, respectively, n = 7, p = 0.68). The degree of the GABABR-mediated enhancement of the 1S,3R-ACPD current did not change in the membrane potential range examined (Fig. 4c), indicating that the GABABR-mediated enhancement of mGluR1-activated currents is independent of the membrane potential. Furthermore, the effect of GABABR activation did not depend on the amplitude of the 1S,3R-ACPD responses, as there was no correlation between the initial amplitude of 1S,3R-ACPD-induced currents (140 to 280 pA) and the extent of the baclofen-induced enhancement of 1S,3R-ACPD currents (Fig. 4d). Extrapolation of the baclofen-induced current response showed its reversal potential of –93.6 ± 5.0 mV (n = 5), which was close to the K+ equilibrium potential (–96.6 mV) predicted from the Nernst equation. In the presence of a GIRK channel inhibitor, Ba2+ (1 mM), baclofen still induced an outward current, the extent of which was almost comparable to that of the control response (43.3 ± 5.8 pA, n = 4, p = 0.46). Furthermore, Ba2+ had little effect on the GABABR-mediated enhancement of 1S,3R-ACPD-induced current (197 ± 22% of the control response, n = 4, p = 0.32; example, Fig. 4g). The GIRK current in principal neurons of the amygdala was, in fact, almost completely blocked by 1 mM Ba2+ when applied by superfusion to slice preparations (ref. 32 and unpublished observations). Therefore, it seems that the GABABR-mediated enhancement and outward currents are not due to the activation of GIRK channels. The baclofen-induced outward current and enhancement of mGluR1-mediated response were dose-dependent with half maximal effective concentrations (EC50) of approximately 1.01 and 0.94 µM, respectively (Fig. 4e), and pooled data showed that there was a modest correlation between both responses (Fig. 4f). In addition to the two effects, baclofen markedly inhibited PF-stimulation-evoked excitatory postsynaptic currents (EPSCs), and nature neuroscience • volume 4 no 12 • december 2001

the IC50 of PF-EPSC inhibition was approximately 0.77 µM, which is consistent with the value reported previously33,34. The effective concentration ranges of baclofen for inhibiting PF-EPSCs as well as causing the mGluR1-current enhancement and outward currents were very similar (Fig. 4e). Effect of baclofen on mGluR1-mediated [Ca2+]i transients Application of the mGluR agonist 1S,3R-ACPD increased [Ca2+]i in PCs as reported previously24–26. To determine the effects of baclofen on the mGluR1-induced [Ca2+]i increase, we performed simultaneous whole-cell recordings and intracellular Ca2+ imaging in PCs using a Ca 2+ indicator, fura-2. Baclofen (3 µM) enhanced not only the inward current but also the [Ca2+]i elevation produced in response to iontophoretic application of 1S,3R-ACPD (Fig. 5): the 1S,3R-ACPD-current and [Ca2+]i transients measured in the distal dendrites of PCs were increased to 180 ± 22% (n = 4, p < 0.01) and 321 ± 71% (n = 4, p < 0.05) of the control responses, respectively (Fig. 5c). 1S,3R-ACPD caused a larger increase in [Ca2+]i at distal dendrites than at proximal dendrites of the PC. Furthermore, baclofen increased the basal level of [Ca2+]i in three of four PCs tested. The averaged F340/F380 ratio reflecting the basal [Ca2+]i increased during the GABABR agonist application and recovered to the control level after the agonist was washed out (Fig. 5d). This effect was particularly significant in the recording site at proximal dendrites. The effect of baclofen on [Ca2+]i is not compatible with the previous observation that GABABR activation inhibits Ca2+ influx through P-type Ca2+ channels in PCs11. A possible involvement of P/Q-type Ca2+ channel inhibition in the modulation of mGluR1 response was excluded as the P/Q-type Ca2+ channel blocker ω-agatoxin IVA (100 nM) did not enhance the 1S,3R-ACPD-induced inward current but, rather, slightly reduced its amplitude to 82.8 ± 6.2% of the control response (n = 5, p < 0.05). Thus, it seems that GABABR activation increased the basal [Ca2+]i level by influencing internal Ca 2+ store through mGluR1-mediated and G-protein-coupled signaling pathways. An analogous mechanism has been proposed for the modulation of the basal [Ca2+]i level elicited by the activation of other metabotropic receptors, such as opioid receptors, and adenosine A1 and NPYY1 receptors35. Another possibility would be that baclofen enhances the [Ca2+]i 1209



articles

a

b

d

e

c

f

g

Fig. 4. Voltage-independent facilitation of GABABR-mediated enhancement of the 1S,3R-ACPD-induced currents, and comparisons of presynaptic and postsynaptic actions induced by baclofen. (a) 1S,3R-ACPD was iontophoretically applied to a Purkinje cell (PC) at 40-s intervals, and the holding potential of the PC (–60 mV) was shifted by constant voltage ramps from –130 to 0 mV for 1560 ms at the time points as indicated by (1, 1´) and (2, 2´) during and after the 1S,3R-ACPD-induced response, respectively. This sequence of ramp commands was repeated before (1, 2) and during baclofen (3 µM) application (1´, 2´). (b) 1S,3R-ACPD-induced currents determined by subtraction of current–voltage relationships produced by ramp commands as indicated in (a). (c) The amplitude ratios of 1S,3R-ACPD-induced currents before and during baclofen application calculated as (1´ – 2´)/(1 – 2) × 100 (%) are plotted against the membrane potential range determined in PCs (n = 5). (d) Relationship between the extent of baclofen-induced enhancement of mGluR1-mediated currents and the amplitude of 1S,3R-ACPD-currents before baclofen (3 µM) application in individual PCs. The straight line indicates a regression line with a correlation coefficient, r = –0.029, indicating no correlation between the two responses. (e) Comparison of dose–response relationships for the baclofen-induced enhancement of the mGluR1-mediated current, outward current and inhibition of PF-mediated EPSCs. The 50% effective doses(EC50) of baclofen were determined for the 3 responses (n = 3–15): the increase of the mGluR current expressed as a percentage of that induced by 100 µM baclofen (white circles and solid line, EC50 ≈ 0.94 µM), the amplitude of baclofen-induced outward current response expressed as a percentage of that induced by 100 µM baclofen (black circles and dashed line, EC50 ≈ 1.01 µM) and percentage inhibition of PF-EPSC by baclofen (white squares and dotted line, IC50 ≈ 0.77 µM). The Hill coefficient (n) determined from each dose–response curve was 0.84, 0.92 and 0.94, respectively. (f) Relationship between baclofen-induced enhancement of mGluR1mediated current and outward current response. The straight line indicates a regression line with a correlation coefficient was r = 0.558. (g) Effects of Ba2+ on the baclofen-induced outward current response and enhancement of mGluR1-mediated current. Ba2+ (1 mM) and baclofen (3 µM) were applied by perfusion during the period indicated by horizontal bars. Downward deflections represent the inward currents induced by iontophoretic application of 1S,3R-ACPD at a constant interval, as in (a).

rise via tonic activation of mGluR1 by glutamate released spontaneously from excitatory nerve terminals. We next examined the mechanisms underlying the baclofeninduced enhancement of the mGluR1 response. First, we investigated whether G i/o activation is required for the GABABR-mediated enhancement. Treatment of cerebellar slices with N-ethylmaleimide (NEM), a Gi/o inhibitor9, at a concentration of 50 µM for 15 minutes significantly reduced the extent of the baclofen-induced enhancement of the 1S,3R-ACPDinduced current (127 ± 17% of the baseline after NEM treatment, versus 226 ± 11% of the baseline in the control ACSF, n = 4, p < 0.001; Fig. 6a and b). However, the NEM treatment did not cause any significant effects on the 1S,3R-ACPD-current 1210

(89 ± 10% of the control response, n = 4, p = 0.39) and the AMPA-current (98 ± 6% of the control response, n = 3, p = 0.92). This finding suggests that G-proteins, presumably G i/o, are responsible for the GABABR-mediated enhancement. One target of Gi/o proteins linked with GABABRs might be adenylyl cyclase, as GABABR agonists reduce the level of intracellular cyclic AMP by inhibiting adenylyl cyclase12. However, the baclofen-induced increase in the 1S,3R-ACPD current was not significantly affected by treatment with either forskolin (30 µM) for 20 minutes or 8Br-cAMP (500 µM) for 35 minutes (data not shown). Treatment with protein kinase inhibitors H-7 (20 µM) or H-8 (20 µM) for 20 minutes also had no effect (data not shown). Another possible target of Gi/o proteins is PLC. Gi/o nature neuroscience • volume 4 no 12 • december 2001


articles


a

b

c

d

Fig. 5. Simultaneous recordings of GABABR activation-mediated enhancement of 1S,3R-ACPD-induced current and [Ca2+]i transients in a PC. (a) [Ca2+]i transients were fluorometrically measured at proximal (P, white) and distal dendrites (D, red) of the PC held at –60 mV starting 25 min after loading the Ca2+ indicator fura-2 (1 mM) via the recording electrode. The fura-2-loaded PC was viewed with a fluorescence image produced by 380 nm excitation wavelength (1). Pseudocolor ratio images were recorded before 1S,3R-ACPD application (2), during 1S,3R-ACPD application in the absence (3) and presence (4) of 3 µM baclofen. (b) Effects of baclofen on 1S,3R-ACPD-induced Ca2+ signals (top) and current responses (bottom). 1S,3R-ACPD was iontophoretically applied to a single PC at a constant interval as indicated by dots, and baclofen (3 µM) was applied by perfusion during the period indicated by a horizontal bar. The numbers 2–4 indicate the time points where the images in (a) were obtained, and the fluorescence ratio curves indicated by red and black were measures at distal and proximal dendritic sites, respectively. (c) Time courses of baclofen-induced enhancement of the mGluR1-mediated current (white circles) and [Ca2+]i transient, F340/F380 ratio (black circles) in PCs. Each response is expressed as a mean percentage of the control response determined immediately before baclofen (3 µM) application (n = 4). (d) Changes in basal [Ca2+]i level in PCs induced by baclofen. Each plot represents the mean ± s.e.m. of the F340/F380 ratios determined before (70–90 s) and during baclofen application (150–170 s) and after washing out of the drug (310–330 s), respectively, in each of four different PCs. *p < 0.05, ***p < 0.001, one-way ANOVA tested for the values before and during baclofen application.

proteins are linked to [Ca2+]i increases via activation of IP3 receptors35,36, and stimulation of GABABRs in the cerebellar cortex causes activation of PLC via G-proteins, thereby resulting in modulation of GABAA receptor fuctions37. We therefore tested the possible involvement of PLC and IP3 receptors in the GABABRmediated response. A selective PLC inhibitor, U73122 (10 µM), infused into PCs via a patch electrode, significantly suppressed the baclofen-induced enhancement of the 1S,3R-ACPD current when the effect was determined at least 30 minutes after intracellular application of the PLC inhibitor (n = 5, p < 0.05; Fig. 6d). Application of an IP3 receptor modulator, heparin (300 units/ml), also suppressed the GABABR-mediated enhancement (n = 4, p < 0.01). Furthermore, intracellular infusion of the Ca2+ chelator BAPTA (35 mM) significantly reduced the extent of the 1S,3R-ACPD current enhancement (n = 5, p < 0.05). The baclofen-induced outward current was also significantly reduced by all the three treatments (data not shown). The pharmacological manipulations used here caused only partial suppression of the mGluR1-mediated current response per se as previously reported 26,38 (Fig. 6c). Taken together, it is likely that the baclofen-induced basal [Ca2+]i rise resulting from Ca2+ release nature neuroscience • volume 4 no 12 • december 2001

from intracellular stores is responsible for the enhancement of the mGluR1 response. The baclofen-induced outward current might be attributable to a Ca2+-activated K+ current. Dual effects of baclofen on mGluR1-mediated EPSC In the presence of ionotropic glutamate and GABA receptor antagonists, repetitive stimulation of PFs produces in PCs a slow excitatory synaptic response that is mediated by group I mGluRs26,38–40. A crucial test is to determine whether GABABR activation enhances synaptically evoked mGluR-mediated responses (Fig. 7). Activation of PFs with 10 stimuli at 100 Hz produced fast EPSCs with a gradual increase in amplitudes that were followed by a slow inward EPSC. Fast and slow components of the synaptic responses were AMPA-receptor- and mGluR1mediated EPSCs, respectively, because the former was almost completely blocked by CNQX (30 µM) and the latter was abolished by the group I mGluR antagonist 4CPG (500 µM)26. Application of baclofen at a low concentration of 0.3 µM increased the amplitude of slow EPSCs to 118 ± 6% of the control (n = 9, p < 0.01; Fig. 7a–c). In contrast, baclofen at the same concentration decreased the amplitude of fast EPSCs to 89.2 ± 5.7% of the 1211



articles

Fig. 6. Effects of signal transduction a modulators on the GABABR-induced enhancement of the mGluR1-mediated current. (a) Effects of NEM treatment on 1S,3R-ACPDand AMPA-current responses and baclofen-induced enhancement of 1S,3R-ACPD-response. NEM (50 µM) was applied by perfusion. The responses were recorded at the time points indicated by 1 to 3 in (b). (b) Time courses of the NEM effects on 1S,3R-ACPD-current and GABABRinduced enhancement of 1S,3RACPD-response. The amplitude of c 1S,3R-ACPD-currents is expressed as a percentage of the controls immediately before NEM application (black squares) and baclofen application without NEM treatment (white circles), respectively. (c) Effects of the PLC inhibitor U73122 (10 µM) and the IP3 receptor inhibitor heparin (300 units/ml) infusions on the time course change of the 1S,3R-ACPDinduced current amplitudes. Each data point represents the mean ± s.e.m. of a percentage to the amplitude of initial 1S,3R-ACPD-current determined 2 min after obtaining whole-cell access in 5 to 12 PCs. (d) Effects of NEM (50 µM), U73122 (10 µM), heparin (300 units/ml) and BAPTA (35 mM) infusions on the baclofen-induced enhancement of 1S,3R-ACPD-induced current. Each compound except NEM was applied for at least 30 min by infusing into PCs via the recording electrode, and then the application of 3 µM baclofen was initiated. NEM was applied by superfusion for 15 min, and the effect of baclofen on 1S,3R-ACPD-induced current response was determined in the presence of NEM and expressed as a percentage of the control 1S,3RACPD-response before the GABABR agonist application. The number in parentheses represents the number of independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, one-way ANOVA with Tukey’s post-test.

control (n = 9). When the concentration was increased to 1–30 µM, baclofen inhibited both fast and slow EPSCs to a similar degree (42.1 ± 8.1% and 43.9 ± 6.5% inhibition at 10 µM, respectively, n = 9; Fig. 7d and e). The dose–response relationships for the baclofen-induced enhancement and inhibition of the slow mGluR1-mediated EPSC are shown in Fig. 8a. The enhancement might be explained by postsynaptic interaction between GABABRs and mGluR1 activated by the excitatory transmitter released from PF terminals. The inhibitory action of baclofen might be due to a presynaptic mechanism of reducing the amount of transmitter released39. It is likely, therefore, that postsynaptic GABABRs are more susceptible to baclofen than are presynaptic GABABRs. In the following two series of experiments, we further determined whether endogenous GABA released from cerebellar interneurons can enhance the mGluR1-mediated slow EPSCs. First, we examined the effect of a GABA uptake inhibitor, SKF89976A, on the slow EPSCs. Application of SKF89976A at a concentration of 10 µM increased the amplitude of slow EPSCs in all PCs examined (112 ± 5% of the control, n = 4, p < 0.05), whereas the GABA uptake inhibitor at a higher concentration of 100 µM decreased the slow EPSC amplitude by 44.5 ± 6.8% (n = 4, data not shown). The observations suggest that endogenous GABA causes enhancement and inhibition of the mGluR1mediated transmission via activation of postsynaptic and 1212

b

d

presynaptic GABABRs, respectively, depending on the concentration of GABA released into the synaptic cleft. Second, we examined the effect of the GABABR antagonist CGP62349 on the mGluR1-mediated EPSC. The slow EPSCs evoked by repetitive stimulation of PFs gradually decreased to 74.5 ± 5.5% of the control after application of CGP62349 at a concentration of 300 nM (n = 6, p < 0.01) and the inhibitory effect persisted over 10 minutes after the drug was washed out (Fig. 8b and c). In contrast, the GABABR antagonist caused a slight increase in amplitude of the fast EPSC (109 ± 7.9% of the control, n = 6). These observations are consistent with the possibility that the mGluR1-mediated slow EPSC could be enhanced by the activation of postsynaptic GABABRs due to endogenous GABA released during repetitive stimulation of PFs, as PFs form excitatory connections with GABA-containing interneurons to elicit GABA release. The modulation of fast EPSCs by endogenous GABA seems to be dominated by the inhibitory action of presynaptic GABABRs.

DISCUSSION G-protein-coupled GABABRs and mGluR1 display analogous distribution profiles at cerebellar PF–PC excitatory synapses with both receptors being localized mainly at extrasynaptic sites along the PC dendrites5,15,21–23. The physiological functions of presynaptic GABABRs in PF terminals have been well studied33,34, but relatively little is known about the functions of postsynaptic GABABRs on PCs. Therefore, we investigated functional interactions between postsynaptic GABABRs and mGluR1 in PCs. We found that activation of GABABRs by the exogenous agonist baclofen elicited a profound enhancement of the mGluR1 agonist 1S,3R-ACPD-induced current response and [Ca2+]i rise in PCs (Figs. 2 and 5) and that endogenous GABA synaptically released from cerebellar interneurons following PF stimulation could enhance the mGluR1-mediated synaptic response evoked by the PF excitatory transmitter (Fig. 8). Our data suggest that the GABA B R-mediated enhancement of mGluR1 response requires Ca2+ release from internal stores through a signaling nature neuroscience • volume 4 no 12 • december 2001


articles


a

b

d

c

e

Fig. 7. Enhancement by baclofen of mGluR1-mediated slow EPSCs produced in response to PF stimulation. (a) Fast and slow EPSCs were evoked by repetitive stimulation (100 Hz for 100 ms) of the PF at a constant interval of 60 s in the presence of CNQX (10 µM), AP5 (30 µM) and bicuculline (50 µM) and recorded from a PC. The synaptic responses were recorded before (top) and during baclofen (0.3 µM) application (middle), and after the drug was washed out (bottom). (b) Superimposed PF-mediated slow EPSCs in the control and 0.3 µM baclofen-containing ACSF are displayed on a fast time base. (c) Time course of the effects of baclofen on fast and slow EPSCs produced in PCs by PF stimulation. The amplitude of both EPSCs is expressed as a percentage of the control amplitude determined immediately before baclofen application (n = 9). (d) Inhibitory effects of a higher concentration of baclofen (10 µM) on PF-mediated slow EPSCs. (e) Time course of baclofen-induced inhibition of fast and slow EPSCs recorded from PCs. The amplitude of both responses was expressed as a percentage of the control determined immediately before 10 µM baclofen application (n = 4).

mechanism that involves G i/o-coupled PLC activation. This GABABR-mediated modulation of synaptic processes seemed to be specific to the mGluR1 responses, because the ionotropic AMPA-type GluR-mediated current response was not affected by the GABA BR agonist baclofen (Fig. 2). Furthermore, the enhancement of mGluR1-mediated responses is a unique property of metabotropic GABABRs, as other G-protein-linked receptors including serotonin, adenosine and muscarinic receptors were devoid of this capability in PCs. Two possible mechanisms may explain the cross-talk between GABABR and mGluR1 revealed in this study. First, because not only the mGluR1-activated current but also the mGluR1-induced [Ca2+]i increase in PCs were enhanced following GABABR activation, Ca2+ mobilization from internal stores through Gi/olinked PLC activation and IP3 formation might be critical in the GABABR-mGluR1 interaction. This notion was supported by the following observations: treatment with the Gi/o inhibitor NEM markedly attenuated the GABABR-mediated enhancement of the current response; infusion of the PLC inhibitor U73122 and the IP3 antagonist heparin into PCs suppressed these GABABRmGluR1 interactions; infusion of the Ca2+ chelator BAPTA also inhibited the effects of GABABR activation on mGluR1 responses; and application of the GABABR agonist baclofen increased basal [Ca2+]i (Fig. 5). Thus, it seems that GABABR activation may be linked to intracellular Ca2+ stores to modulate [Ca2+]i and enhance the mGluR1 functions including the current response and Ca2+ elevation, presumably through a cooperative upregulation of G-protein-coupled receptor signaling by Gβγ subunits as reported in β-adrenergic receptor-GABABR synergistic interactions41. This is compatible with the finding that mGluR1-mediated excitation at PF–PC synapses is markedly increased by activation of the climbing fiber input via a transient increase in nature neuroscience • volume 4 no 12 • december 2001

[Ca2+]i40. Studies have suggested that the regulation of Ca2+ release from internal stores in presynaptic and postsynaptic neurons profoundly influences short- and long-term plasticity at cerebellar synapses42–44. Another possibility is that the GABABR-mediated enhancement of the mGluR1-activated current and Ca 2+ signals are attributable to two independent mechanisms, although there is no direct evidence supporting this. GABABR activation produced a substantial outward current that was resistant to treatment with Ba2+, an inhibitor of GABABR-linked GIRK channels. A previous study using a genetic knockout of GIRK channels clearly demonstrated that presynaptic GABABR-mediated inhibition of neurotransmission is independent of GABABR-mediated GIRK channel activation10. In our study, GABABRs associated with the enhancement of synaptically evoked mGluR1 response exhibited the sensitivity higher than those associated with presynaptic inhibitory actions (Fig. 8), although the exogenous mGluR agonist 1S,3R-ACPD-induced response and PF–PC transmission were affected by baclofen in a similar concentration range. Therefore, it remains to be determined whether separate receptor subtypes with distinct signaling pathways are involved in the conventional GABABR-mediated presynaptic inhibition and postsynaptic GIRK channel activation, and in the mechanism of GABABR-mGluR1 cross-talk found in this study. Two possible physiological consequences might be associated with the GABABR–mGluR1 interaction at PF–PC synapses. First, the interaction may increase the excitability of PCs, as it is assumed that the GABABR activation enhances the slow depolarizing synaptic potential mediated through mGluR1 activation by the excitatory transmitter glutamate released from PF nerve terminals. If this is the case, GABA may serve dual functions at PF–PC synapses, either as an excitatory modulator caus1213


articles


a

b

c

Fig. 8. Dose dependency of baclofen-induced enhancement and inhibition of mGluR1-mediated slow EPSCs, and the effects of synaptic GABABR activation on fast and slow EPSCs tested by using the GABABR antagonist CGP62349. (a) Dose–response relationships between baclofen-induced facilitatory and inhibitory actions on mGluR1-mediated slow EPSCs following repetitive PF stimulation were determined in PCs. The number in parentheses represents the number of PCs in which effects of baclofen were tested. (b) CGP62349-induced increase in PF-mediated fast EPSC and decrease in slow EPSC amplitude. The synaptic responses following repetitive PF stimulation were recorded from a PC in the control (left) and in 300 nM CGP62349-containing ACSF (right). (c) Time courses of CGP62349-induced effects on fast and slow EPSCs. The amplitudes of both EPSCs are expressed as a percentage of the control amplitude determined immediately before application of the GABABR antagonist (n = 6).

ing the postsynaptic transient facilitation of PF excitatory inputs to PCs, or as a classical transmitter eliciting presynaptic inhibition of the ionotropic GluR-mediated fast excitatory transmission. Second, the GABABR-mGluR1 cross-talk may be critical in synaptic plasticity associated with the cerebellar function, as mGluR1 has been implicated as an important molecule in the induction of LTD at these synapses19,20, which is proposed as a key mechanism underlying motor coordination within in the cerebellar system17,18. Furthermore, studies have suggested that mGluR1-mediated Ca2+ release from internal stores in PCs acts as a coincidence detection mechanism for PF and CF activations, thereby leading to induction of LTD at PF–PC synapses44. Therefore, simultaneous activation of mGluR1 and GABABRs would enhance this mechanism. The physiological significance of GABABRs in synaptic plasticity has also been shown in other brain regions. Presynaptic GABABRs seem to be involved in longterm potentiation (LTP) in the hippocampus45,46. In addition, postsynaptic GABABRs contribute to long-term regulation of synaptic strength at GABAergic inhibitory synapses in the visual cortex47. In this case, GABABRs and adrenoceptors seem to act in concert to further enhance heterosynaptic monoaminergic LTP, possibly through GABA BR-mediated facilitation of monoamine-induced IP 3 formation. Negative regulation of synaptic plasticity, which involves postsynaptic GABABRs, has been reported at inhibitory synapses in the cerebellar cortex: the activation of GABABRs in PCs downregulates the long-lasting increase, or ‘rebound potentiation,’ of GABAA receptor sensitivity following depolarization of PCs 16 . The postsynaptic GABABR–mGluR1 interaction identified in this study provides a prominent example of the modulatory role of GABABRs in synaptic plasticity at excitatory PF–PC synapses. Thus, GABABRs localized in presynaptic and postsynaptic neurons seem to be significantly involved in long-term regulation of synaptic efficacy at various synapses in the central nervous system. 1214

Interaction between different transmitter receptor systems is one emerging feature of neurotransmission at central synapses. For instance, dopamine D and somatostatin SST5 receptors form heterodimers to create a novel receptor with augmented functional activity48. Another example has been demonstrated for dopamine D5 and GABAA γ2 receptors49: The GABAA-ligandgated channels complex with D5 receptors via direct binding, thereby enabling mutually inhibitory functional interactions between the two receptor systems. Cross-talk between neurotransmitter-gated cation channels has been reported for heterologous expression systems and cultured neurons, where structurally distinct nicotinic receptor and purinergic P2X2 receptor channels influence each other with regard to cross-inhibition between the two channels50. Conformational spread from one receptor to its neighbors is proposed as a possible mechanism for the cross-talk. As exemplified by this cross-talk, together with the GABABR-mGluR1 interaction identified in this study, the interplay between distinct receptor systems can provide a powerful mechanism to influence chemical signaling at central nervous system synapses.

METHODS Electrophysiology. BDF1 mice (4–5 weeks old) were anesthetised with pentobarbital, and sagittal slices (180–200 µm thick) of the cerebellar vermis were prepared using a vibrating microtome (Microslicer DTK1000, Dosaka, Kyoto, Japan). Whole-cell recordings were obtained from PCs visually identified under Nomarski optics using a water immersion objective (40×, NA 0.75, Zeiss, Germany). Slices were superfused with ACSF containing 138.6 mM NaCl, 3.35 mM KCl, 21 mM NaHCO3, 0.6 mM NaH2PO4, 9.9 mM glucose, 2.5 mM CaCl2, and 1 mM MgCl2 and were gassed with a mixture of 95% O2 and 5% CO2 (pH 7.4). TTX (0.5 µM) was added to the ACSF to block Na+ spikes and synaptic activity except in experiments investigating synaptic responses initiated by focal stimulation within the cerebellar cortex. Patch pipettes (3 to 6 MΩ) were filled with an internal solution containing 150 mM KCH3SO3, nature neuroscience • volume 4 no 12 • december 2001



articles

5 mM KCl, 0.1 mM K-EGTA, 5.0 mM Na-HEPES, 3.0 mM Mg-ATP and 0.4 mM Na-GTP (pH 7.4). KCH3SO3 in the patch pipette solution was replaced with an equimolar amount of the calcium chelator BAPTA (1,2bis-(2-aminophenoxy)ethane-N,N,N´,N´-tetraacetic acid; 35 mM) to examine the effects of the intracellular application of BAPTA on GABABR-mediated responses. The series resistance was 8–16 MΩ, which was monitored throughout the recording and was not compensated except in experiments investigating synaptic responses. Membrane currents were recorded in a whole-cell configuration using an EPC-7 amplifier (List Electronic, Darmstadt, Germany) and pCLAMP software (Axon Instruments, Foster City, CA), digitized, and stored on a computer disk for off-line analysis. 1S,3R-ACPD (100 mM) and AMPA (10 mM) were applied by iontophoresis (10–20 µA, 100–200 ms) through microelectrodes placed in the vicinity of PC soma from which recordings were obtained. As the 1S,3RACPD-induced current response tended to increase during the initial 15 min after breaking into the whole-cell configuration (Fig. 6c), the effect of GABABR activation was determined thereafter. Furthermore, the effect of a GABABR agonist, baclofen, on 1S,3R-ACPD-responses was tested only once in each slice, because the effect of the GABABR agonist on the mGluR1mediated responses attenuated with repeated applications. Synaptic responses were evoked by stimulation (30–50 V, 0.1–0.2 ms) via a glass microelectrode filled with ACSF and placed within the molecular layer. PFmediated EPSCs were identified based on paired-pulse facilitation. A PFand mGluR1-mediated slow EPSC was evoked by repetitive stimulation (10 pulses at 100 Hz) in an ACSF that contained 6-cyano-7-nitroquinoxaline2,3-dione (CNQX; 10 µM), D-2-amino-5- phosphonopentanoic acid (AP5; 30 µM) and bicuculline (50 µM). U73122, heparin and BAPTA were added to the patch pipette solution and infused into PCs during whole-cell recordings. Other inhibitors and receptor antagonists were applied by perfusion. Ca2+ imaging. The Ca2+ indicator fura-2 was added to the patch pipette solution at a concentration of 1 mM and loaded into the cells for 25–40 min. Fluorescence Ca2+ ratio imaging was done by excitation of the indicator at 340:380 nm and paired emission images were acquired using a cooled CCD camera (C4880, Hamamatsu Photonics, Hamamatsu, Japan) at 510 nm. Paired emission images were recorded every 2 s and ratio images were calculated using a digital image acquisition system and image processing software (ARGUS 50/CA, Hamamatsu Photonics). In all experiments, visual identification of cells and Ca2+ fluorescence imaging were done using a water immersion objective (LUMPlanFI 60×, NA 0.90, Olympus, Tokyo, Japan) capable of passing 340 nm light efficiently. Drugs. 1S,3R-ACPD and SKF89976A were obtained from Tocris Cookson (Bristol, UK); baclofen, heparin and TTX, from Wako (Osaka, Japan); ω-agatoxin IVA, from Peptide Institute (Osaka, Japan); and fura-2, from Dojindo (Kumamoto, Japan). CGP62349 was a gift from Novartis Pharma (Basel, Switzerland). All other chemicals were from Sigma (St. Louis, Missouri). Statistics. Each value is expressed as mean ± s.e.m. Unless otherwise stated, levels of significance were determined by unpaired Student’s t-test between groups.

ACKNOWLEDGEMENTS We thank J. Bockaert, T. Murakoshi, D. Rusakov, F. Saitow and K. Yoshioka for comments on the manuscript and Novartis Pharma (Basel, Switzerland) for the gift of CGP62349. This work was supported in part by a Grant-in-Aid 0727910 (T.Y.) from the Ministry of Education, Science, Sports and Culture of Japan, and Grant-in-Aids 96L00310 (T.Y.) and 12780603 (M.H.) from the Japan Society for Promotion of Science. S.K. is a research director of CREST, JST (Core Research for Evolutional Science and Technology, Japan Science and Technology Corporation).

RECEIVED 31 JULY; ACCEPTED 5 OCTOBER 2001 1. Bowery, N. G. GABAB receptor pharmacology. Annu. Rev. Pharmacol. Toxicol. 33, 109–147 (1993).


2. Barnard, E. A. et al. Subtypes of GABAA receptors: classification on the basis of subunit structure and receptor function. Pharmacol. Rev. 50, 291–313 (1998). 3. Kaupmann, K. et al. Expression cloning of GABAB receptors uncovers similarity to metabotropic glutamate receptors. Nature 386, 239–246 (1997). 4. Jones, K. A. et al. GABAB receptors function as a heteromeric assembly of the subunits GABABR1 and GABABR2. Nature 396, 674–679 (1998). 5. Kaupmann, K. et al. GABAB-receptor subtypes assemble into functional heteromeric complexes. Nature 396, 683–687 (1998). 6. Kuner, R. et al. Role of heteromer formation in GABAB receptor function. Science 283, 74–77 (1999). 7. Thompson, S. M., Capogna, M. & Scanziani, M. Presynaptic inhibition in the hippocampus. Trends Neurosci. 16, 222–227 (1993). 8. Wu, L. G. & Saggau, P. Presynaptic inhibition of elicited neurotransmitter release. Trends Neurosci. 20, 204–212 (1997). 9. Sodickson, D. L. & Bean, B. P. GABAB receptor-activated inwardly rectifying potassium current in dissociated hippocampal CA3 neurons. J. Neurosci. 16, 6374–6385 (1996). 10. Luscher, C., Jan, L. Y., Stoffel, M., Malenka, R. C. & Nicoll, R. A. G proteincoupled inwardly rectifying K+ channels (GIRKs) mediate postsynaptic but not presynaptic transmitter actions in hippocampal neurons. Neuron 19, 687–695 (1997). 11. Mintz, I. M. & Bean, B. P. GABAB receptor inhibition of P-type Ca2+ channels in central neurons. Neuron 10, 889–898 (1993). 12. Kerr, D. I. & Ong, J. GABAB receptors. Pharmacol. Ther. 67, 187–246 (1995). 13. Martinelli, G. P., Holstein, G. R., Pasik, P. & Cohen, B. Monoclonal antibodies for ultrastructural visualization of L-baclofen-sensitive GABAB receptor sites. Neuroscience 46, 23–33 (1992). 14. Turgeon, S. M. & Albin, R. L. Pharmacology, distribution, cellular localization, and development of GABAB binding in rodent cerebellum. Neuroscience 55, 311–323 (1993). 15. Fritschy, J. M. et al. GABAB-receptor splice variants GB1a and GB1b in rat brain: developmental regulation, cellular distribution and extrasynaptic localization. Eur. J. Neurosci. 11, 761–768 (1999). 16. Kawaguchi, S. & Hirano, T. Suppression of inhibitory synaptic potentiation by presynaptic activity through postsynaptic GABAB receptors in a Purkinje neuron. Neuron 27, 339–347 (2000). 17. Ito, M. Long-term depression. Annu. Rev. Neurosci. 12, 85–102 (1989). 18. Linden, D. J. & Connor, J. A. Long-term synaptic depression. Annu. Rev. Neurosci. 18, 319–357 (1995). 19. Conquet, F. et al. Motor deficit and impairment of synaptic plasticity in mice lacking mGluR1. Nature 372, 237–243 (1994). 20. Ichise, T. et al. mGluR1 in cerebellar Purkinje cells essential for long-term depression, synapse elimination, and motor coordination. Science 288, 1832–1835 (2000). 21. Nusser, Z., Mulvihill, E., Streit, P. & Somogyi, P. Subsynaptic segregation of metabotropic and ionotropic glutamate receptors as revealed by immunogold localization. Neuroscience 61, 421–427 (1994). 22. Luján, R., Roberts, J. D., Shigemoto, R., Ohishi, H. & Somogyi, P. Differential plasma membrane distribution of metabotropic glutamate receptors mGluR1 alpha, mGluR2 and mGluR5, relative to neurotransmitter release sites. J. Chem. Neuroanat. 13, 219–241 (1997). 23. Mateos, J. M. et al. Immunolocalization of the mGluR1b splice variant of the metabotropic glutamate receptor 1 at parallel fiber–Purkinje cell synapses in the rat cerebellar cortex. J. Neurochem. 74, 1301–1309 (2000). 24. Staub, C., Vranesic, I. & Knöpfel, T. Responses to metabotropic glutamate receptor activation in cerebellar Purkinje cells: induction of an inward current. Eur. J. Neurosci. 4, 832–839 (1992). 25. Linden, D. J., Smeyne, M. & Connor, J. A. Trans-ACPD, a metabotropic receptor agonist, produces calcium mobilization and an inward current in cultured cerebellar Purkinje neurons. J. Neurophysiol. 71, 1992–1998 (1994). 26. Hirono, M., Konishi, S. & Yoshioka, T. Phospholipase C-independent group I metabotropic glutamate receptor-mediated inward current in mouse Purkinje cells. Biochem. Biophys. Res. Commun. 251, 753–758 (1998). 27. Goodman, R. R., Kuhar, M. J., Hester, L. & Snyder, S. H. Adenosine receptors: autoradiographic evidence for their location on axon terminals of excitatory neurons. Science 220, 967–969 (1983). 28. Jaarsma, D., Levey, A. I., Frostholm, A., Rotter, A. & Voogd, J. Lightmicroscopic distribution and parasagittal organisation of muscarinic receptors in rabbit cerebellar cortex. J. Chem. Neuroanat. 9, 241–259 (1995). 29. Miquel, M. C. et al. Postnatal development and localization of 5-HT1A receptor mRNA in rat forebrain and cerebellum. Brain Res. Dev. Brain Res. 80, 149–157 (1994). 30. Konishi, S. & Mitoma, H. in The Role of Adenosine in the Nervous System (ed. Okada, Y.) 89–95 (Elsevier, New York, 1997). 31. Mitoma, H. & Konishi, S. Monoaminergic long-term facilitation of GABAmediated inhibitory transmission at cerebellar synapses. Neuroscience 88, 871–883 (1999). 32. Yamada, J., Saitow, F., Satake, S., Kiyohara, T. & Konishi, S. GABAB receptormediated presynaptic inhibition of glutamatergic and GABAergic transmission in the basolateral amygdala. Neuropharmacology 38, 1743–1753 (1999). 33. Takahashi, M., Kovalchuk, Y. & Attwell, D. Pre- and postsynaptic determinants of EPSC waveform at cerebellar climbing fiber and parallel fiber to Purkinje cell synapses. J. Neurosci. 15, 5693–5702 (1995).

1215



articles

34. Dittman, J. S. & Regehr, W. G. Contributions of calcium-dependent and calcium-independent mechanisms to presynaptic inhibition at a cerebellar synapse. J. Neurosci. 16, 1623–1633 (1996). 35. Selbie, L. A. & Hill, S. J. G protein-coupled-receptor cross-talk: the finetuning of multiple receptor-signalling pathways. Trends Pharmacol. Sci. 19, 87–93 (1998). 36. Jin, W., Lee, N. M., Loh, H. H. & Thayer, S. A. Opioids mobilize calcium from inositol 1,4,5-trisphosphate-sensitive stores in NG108-15 cells. J. Neurosci. 14, 1920–1929 (1994). 37. Hahner, L., McQuilkin, S. & Harris, R. A. Cerebellar GABAB receptors modulate function of GABAA receptors. FASEB J. 5, 2466–2472 (1991). 38. Tempia, F., Miniaci, M. C., Anchisi, D. & Strata, P. Postsynaptic current mediated by metabotropic glutamate receptors in cerebellar Purkinje cells. J. Neurophysiol. 80, 520–528 (1998). 39. Batchelor, A. M. & Garthwaite, J. Novel synaptic potentials in cerebellar Purkinje cells: probable mediation by metabotropic glutamate receptors. Neuropharmacology 32, 11–20 (1993). 40. Batchelor, A. M. & Garthwaite, J. Frequency detection and temporally dispersed synaptic signal association through a metabotropic glutamate receptor pathway. Nature 385, 74–77 (1997). 41. Bourne, H. R. & Nicoll, R. Molecular machines integrate coincident synaptic signals. Cell 72 Suppl., 65–75 (1993). 42. Llano, I. et al. Presynaptic calcium stores underlie large-amplitude miniature

1216

43. 44. 45. 46. 47.

48. 49. 50.

IPSCs and spontaneous calcium transients. Nat. Neurosci. 3, 1256–1265 (2000). Miyata, M. et al. Local calcium release in dendritic spines required for longterm synaptic depression. Neuron 28, 233–244 (2000). Wang, S. S., Denk, W. & Hausser, M. Coincidence detection in single dendritic spines mediated by calcium release. Nat. Neurosci. 3, 1266–1273 (2000). Davies, C. H., Starkey, S. J., Pozza, M. F. & Collingridge, G. L. GABA autoreceptors regulate the induction of LTP. Nature 349, 609–611 (1991). Mott, D. D. & Lewis, D. V. Facilitation of the induction of long-term potentiation by GABAB receptors. Science 252, 1718–1720 (1991). Komatsu, Y. GABAB receptors, monoamine receptors, and postsynaptic inositol trisphosphate-induced Ca2+ release are involved in the induction of long-term potentiation at visual cortical inhibitory synapses. J. Neurosci. 16, 6342–6352 (1996). Rocheville, M. et al. Receptors for dopamine and somatostatin: formation of hetero-oligomers with enhanced functional activity. Science 288, 154–157 (2000). Liu, F. et al. Direct protein–protein coupling enables cross-talk between dopamine D5 and GABAA receptors. Nature 403, 274–280 (2000). Khakh, B. S., Zhou, X., Sydes, J., Galligan, J. J. & Lester, H. A. State-dependent cross-inhibition between transmitter-gated cation channels. Nature 406, 405–410 (2000).



articles


Long-term depression in the nucleus accumbens: a neural correlate of behavioral sensitization to cocaine Mark J. Thomas1, Corinne Beurrier1, Antonello Bonci2 and Robert C. Malenka1 1 Nancy Pritzker Laboratory, Department of Psychiatry and Behavioral Sciences, Stanford University School of Medicine, Palo Alto, California 94304, USA 2 Ernest Gallo Clinic and Research Center, Department of Neurology, University of California, San Francisco, California 94110, USA

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

Published online: 5 November 2001, DOI: 10.1038/nn757 A compelling model of experience-dependent plasticity is the long-lasting sensitization to the locomotor stimulatory effects of drugs of abuse. Adaptations in the nucleus accumbens (NAc), a component of the mesolimbic dopamine system, are thought to contribute to this behavioral change. Here we examine excitatory synaptic transmission in NAc slices prepared from animals displaying sensitization 10–14 days after repeated in vivo cocaine exposure. The ratio of AMPA (α-amino-3-hydroxy5-methyl-4- isoxazole propionic acid) receptor- to NMDA (N-methyl-D-aspartate) receptor-mediated excitatory postsynaptic currents (EPSCs) was decreased at synapses made by prefrontal cortical afferents onto medium spiny neurons in the shell of the NAc. The amplitude of miniature EPSCs at these synapses also was decreased, as was the magnitude of long-term depression. These data suggest that chronic in vivo administration of cocaine elicits a long-lasting depression of excitatory synaptic transmission in the NAc, a change that may contribute to behavioral sensitization and addiction.

Drug addiction is a pathological behavior characterized by compulsive drug seeking and drug ingestion despite severe adverse consequences. Animal models of addiction mimic several of the core features of addiction in humans and therefore can be used to study the neural mechanisms underlying this pathological form of experience-dependent behavioral plasticity. Because of advances in molecular neurobiology, much is known about how drugs of abuse interact with and modify their molecular targets. Furthermore, the molecular adaptations that occur in specific brain regions in response to acute and chronic administration of drugs of abuse are being determined at a rapid pace1. There is a relative paucity of information, however, about the changes in synapses and circuits that occur as a consequence of these druginduced molecular changes; this information is critical for a thorough understanding of the neural mechanisms of addiction. A key feature of addiction is the intensification of drug craving that occurs in human addicts with repeated drug exposure. A prominent model for this behavioral change is the long-lasting increase in locomotor response to drugs of abuse following repeated exposures. This increased response, termed behavioral sensitization, is thought to reflect adaptations in neural circuits that determine the incentive value of external stimuli rendering the circuits hypersensitive or sensitized2. Many of the neural adaptations that have been identified following psychostimulant administration take place in the mesolimbic dopamine system, major components of which are the ventral tegmental area (VTA) and the nucleus accumbens (NAc). Modifications in the VTA are involved in the induction of behavioral sensitization, and modifications in the NAc are involved in its long-term maintenance3,4. For example, repeated injection of psychostimulants into the VTA nature neuroscience • volume 4 no 12 • december 2001

induces behavioral sensitization, whereas in sensitized animals, the injection of psychostimulants into the NAc is sufficient to elicit sensitized responses (for review, see refs. 3, 4). Much of the initial work on the adaptations that mediate behavioral sensitization appropriately focused on pre- and postsynaptic changes in dopaminergic transmission. Recently, however, evidence has accumulated that excitatory inputs to the VTA and NAc are critical. Consistent with the VTA’s involvement in triggering sensitization, repeated electrical stimulation of excitatory cortical afferents to the VTA induces sensitization that, when elicited by systemic drug exposure, is blocked by local injection of glutamate receptor antagonists into the VTA5. Indeed, in vivo administration of cocaine elicits a robust enhancement of excitatory synaptic transmission in this structure6. On the other hand, the expression of behavioral sensitization is blocked by inhibiting excitatory synaptic transmission in the NAc7 or by lesions of the excitatory cortical afferents to this structure8 (but see ref. 9). What changes occur in the NAc that account for its importance in behavioral sensitization? The major cell type in the NAc (>95% of total) is the medium spiny neuron that receives glutamatergic inputs from a variety of cortical and subcortical limbic areas, including the hippocampus, prefrontal cortex and amygdala. Because these neurons have high resting membrane potentials and are generally quiescent, they depend on these excitatory inputs to generate output to their main targets, the VTA and ventral pallidum. It is therefore reasonable to hypothesize that changes in the efficacy of these inputs would have a significant effect on the functioning of mesolimbic dopamine circuitry and would thereby alter the behavioral 1217


articles

saline

cocaine

)

(

cocaine , )

Synaptic transmission in cocaine-treated mice To determine whether changes in the efficacy of AMPA receptor (AMPAR)-mediated synaptic transmission had occurred as a result of the chronic cocaine treatment, we first measured field EPSPs from the NAc10 in slices prepared from the animals one day after the final cocaine challenge. The NAc is commonly divided into two components, the shell and the core, which are distinguished both anatomically and functionally11. Therefore, data were divided according to the location of the recording. Despite using a range of stimulus intensities, we could not detect any difference in the size of the field EPSPs in either the shell or core between cocaineand saline-treated groups (data not shown). Differences in the placement of stimulating and recording electrodes and the density of afferent fibers may be considerable sources of experimental variability when attempting to measure differences in synaptic strength between slices. To reduce the sliceto-slice variability that might have obscured any cocaine-induced changes in synaptic efficacy, we repeated our analysis using a more sensitive assay that compared the relative contributions of AMPARs and NMDA receptors (NMDARs) to EPSCs. This procedure involved evoking EPSCs while holding cells at +40 mV in the absence and then in the presence of the NMDAR antagonist D-APV (50 µM)6 (see Methods). Using this assay, we observed a decrease in the AMPAR/NMDAR ratio in cells in the NAc shell from cocaine-treated compared to saline-treated mice (Fig. 2a–c; cocaine, 0.68 ± 0.09, n = 12; saline 0.97 ± 0.09, n = 9; p < 0.05)

Distance traveled / session (cm)

4000

3000

2000

1000

0 1

2

3

4

5

Injection day

6

7

10–14 test days

Fig. 1. Behavioral sensitization induced by repeated cocaine administration. Mean (± s.e.m.) locomotor activity in response to saline and cocaine injections. Locomotor activity was monitored for 15 min immediately following each injection.

response to a drug of abuse. To test this hypothesis, we examined excitatory synaptic transmission in slices of NAc prepared from mice in which behavioral sensitization was induced by repeated in vivo administration of cocaine. This in vivo treatment caused a long-lasting depression of synaptic a Shell strength at synapses made by prefrontal cortical afferents onto medium spiny neurons in the shell of the NAc.

Fig. 2. Repeated cocaine administration induces a decrease in the AMPAR/NMDAR ratio of synaptic currents in the shell but not in the core of NAc. (a) Sample EPSCs from saline and cocainetreated animals. Scale bars, 40 ms and 20 pA. (b, d) Cumulative probability plots of AMPAR/NMDAR ratio in shell (b) and core (d) neurons from saline (n = 9 shell; n = 8 core) and cocaine-treated (n = 12 shell; n = 7 core) mice. (c, e) Mean AMPAR/NMDAR ratio in shell (c) and core (e) neurons from saline and cocaine-treated mice. *p < 0.05.

1218

cocaine AMPA

b Shell 1.0

Cumulative probability

Long-lasting locomotor sensitization to cocaine To induce behavioral sensitization, we paired repeated cocaine injections with exposure of the animals to a distinct test environment. After two days of saline injections to habituate the animals to the activity box, the immediate locomotor response to a fixed dose of cocaine (15 mg/kg) increased dramatically across days of testing (Fig. 1; distance traveled, day 3, saline, 418 ± 22 cm, n = 51; cocaine, 1176 ± 90 cm, n = 47; day 7, saline, 462 ± 36 cm, n = 51; cocaine, 3896 ± 191 cm, n = 47; p < 0.001). To test whether this procedure produced long-lasting sensitization, we administered a challenge dose of cocaine to both saline- and cocaine-treated groups 10 to 14 days following the last dose of the initial treatment regimen. Mice pretreated with cocaine showed a much greater locomotor response to cocaine than did saline-pretreated animals (cocaine, 4660 ± 204 cm, n = 47; saline, 1556 ± 170 cm, n = 51; p < 0.001). These results indicate that the initial five-day exposure to cocaine caused behavioral sensitization that lasted for at least two weeks.

NMDA

saline

RESULTS

c 1.0

saline cocaine

0.8 0.6 0.4 0.2 0.0

*

0.8 0.6 0.4 0.2 0.0

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6

saline

cocaine

saline

cocaine

AMPAR/NMDAR ratio

d Core

e 1.0

1.0



5000

AMPAR/NMDAR ratio

saline ,

saline cocaine

AMPAR/NMDAR ratio

(

0.8 0.6 0.4 0.2 0.0 0.2

0.8 0.6 0.4 0.2 0.0

0.4

0.6

0.8

1.0

1.2

1.4

AMPAR/NMDAR ratio



articles

a

c

1.0 0.8 0.6 0.4

saline cocaine

0.2

Mean mEPSC amplitude (pA)


b

0.0

20

15

10

5

0 0

10

20

30

saline

40

cocaine

Amplitude (pA)

e Mean mESPC frequency (Hz)

d 1.0 0.8 0.6 0.4 0.2 0.0 0

200

400

600

800

1000

10 8

Shell

a

mean PPF ratio

3.0 2.5

cocaine saline

2.0 1.5 1.0 0.5 0

50

100

150

200

Interstimulus interval (ms)

b

Core 3.0

mean PPF ratio

but not in cells in the core (Fig. 2d and e; cocaine, 0.76 ± 0.11, n = 7; saline 0.78 ± 0.12, n = 8; p > 0.05). Because a single dose of cocaine can elicit behavioral sensitization and an increase in the AMPAR/NMDAR ratio in the VTA6, we wondered if the single injection of cocaine received by the saline-treated mice may have affected the AMPAR/NMDAR ratio and thereby influenced our results. Therefore, in another set of experiments, mice were given either one cocaine or one saline injection and AMPAR/NMDAR ratios for shell neurons were determined the next day. There was no significant difference between the mean AMPAR/NMDAR ratios of these groups (data not shown) indicating that, unlike the VTA, a single cocaine exposure does not affect this measurement in the NAc shell. A decrease in the AMPAR/NMDAR ratio likely reflects a decrease in AMPAR function and/or number, an increase in NMDAR function and/or number, or a combination of these. Such a change, however, does not rule out that changes in transmitter release mechanisms may be induced by in vivo cocaine exposure as well. To assess whether changes in the probability of transmitter release (Pr) occurred in the cocaine-treated animals, we compared the magnitude of the facilitation that occurs in response to paired-pulse stimulation, a measure that routinely



Fig. 3. Repeated cocaine had no significant effect on paired-pulse facilitation in either shell or core. (a, b) Mean paired-pulse facilitation (PPF) values in shell (a; saline, n = 7; cocaine, n = 9) and in core (b; saline, n = 9; cocaine, n = 8) are shown for different interstimulus intervals. A sample trace from a shell neuron in the cocaine group is shown (50 ms ISI; calibration bars are 20 ms, 20 pA). Values did not differ significantly between groups at any interval tested (p > 0.05; ANOVA).

2.5 2.0 1.5 1.0 0.5 0

50

100

150

200

Interstimulus interval (ms)

changes with changes in Pr12. The paired-pulse ratio did not differ between cocaine- and saline-treated mice when tested at a variety of intervals in either shell or core neurons following the cocaine treatment regimen (Fig. 3), suggesting that significant changes in Pr did not occur in cocaine-treated mice. Changes in mEPSC amplitude in cocaine-treated mice To address whether AMPAR function and/or number was modified in the NAc shell of cocaine-treated mice, we examined miniature AMPAR-mediated EPSCs (mEPSCs). We found no difference between cocaine and saline-treated mice in either the frequency or the amplitude of mEPSCs in shell neurons (Fig. 4). This suggested that the decrease in the AMPAR/NMDAR ratio might be due to an increase in the function and/or number of NMDARs. Testing this possibility is difficult because the relatively high frequency of mEPSCs in this preparation (∼7 Hz), the increased ‘noise’ generated by opening of NMDARs by ambient glutamate13 and the slow rise-time of NMDAR events make it problematic to reliably detect NMDAR-mediated mEPSCs. To circumvent these problems, we bathed slices in Mg 2+-free solution to record mEPSCs containing both AMPAR and NMDAR components using the fast rise-time of the AMPAR component to reliably detect the mEPSCs. We then applied D-APV and collected AMPAR mEPSCs from the same cell. Subtracting average mEPSCs in the presence of D-APV from those collected in its absence yielded an average NMDAR mEPSC for each cell (Fig. 5a and b). This mea-

6 4 2 0

saline

Inter-event Interval (ms)


cocaine

Fig. 4. Repeated cocaine administration had no effect on the amplitude or frequency of shell AMPAR mEPSCs. (a) Sample traces of mEPSCs in a shell neuron in a cocaine-treated mouse. Scale bars, 80 ms and 10 pA. (b, d) Cumulative amplitude (b) and inter-event interval (d) distributions of shell mEPSCs obtained in cocaine and saline-treated mice (n = 400 events per cell, 8 cells in each group). (c, e) Means of mEPSCs amplitude (c) and frequency (e) p > 0.05. 1219


articles

2+

a

b

zero Mg

dual AMPA NMDA

2+

c

d

5

saline cocaine

Mean amplitude (pA)


A limitation of our analysis of mEPSCs is that medium spiny neurons receive excitatory projections from several 0.8 sources and there was no way of knowing from which 3 0.6 synapses the mEPSCs were generated. If the changes in the AMPAR/NMDAR ratio of the evoked EPSC occurred 2 0.4 primarily at the stimulated synapses, presumably those 1 0.2 made by cortical afferents, we may have missed detecting a change in mEPSCs. To circumvent this problem, we 0 0.0 1 2 3 4 5 6 7 8 9 10 replaced the Ca2+ in our bathing solution with Sr2+ and saline cocaine e Amplitude (pA) applied short bursts of afferent stimulation that causes the asynchronous release of vesicles from the set of 120 NMDA synapses that were activated (Fig. 6a)16. Using this techsaline 100 nique, it was possible to preferentially sample mEPSCs cocaine 80 from the same subset of synapses that were activated to 60 yield the evoked EPSC17. Although no significant differ40 ence in the frequency of asynchronous quantal events 20 between shell neurons from cocaine and saline groups was observed (cocaine, 25 ± 3 Hz, n = 11; saline, 0 23 ± 2 Hz, n = 12; p > 0.05), the mean amplitude distri–20 bution of quantal events in the cocaine group was signif–40 0 1 2 3 4 5 icantly shifted to the left compared to the saline group Time (min) (Fig. 6b; n = 11, 12; Kolmogorov–Smirnov test; p < 0.05) and their mean amplitude was also decreased (Fig. 6c) (cocaine, 12.4 ± 0.4 pA; saline 14.0 ± 0.6 pA, p < 0.05). These surement did not differ significantly in shell neurons from results indicate that AMPAR-mediated quantal events generated cocaine- and saline-treated mice (Fig. 5c and d), suggesting that by the synapses activated by cortical afferents are significantly smallNMDAR function and/or number is not altered at synapses that er in the cocaine-treated group, suggesting that the decrease in contain both AMPARs and NMDARs. AMPAR/NMDAR ratio is due, at least in part, to a reduction in Another plausible explanation for the decrease in the AMPAR function and/or number specifically at these synapses. AMPAR/NMDAR ratio in cocaine-treated mice is that cocaine causes an increase in the proportion of synapses that contain only NMDARs, so-called silent synapses14. Such synapses would not Changes in LTD in cocaine-treated mice have been sampled in our experiments examining dual-compoGiven that the synapses made by cortical afferents onto medium nent mEPSCs, but may have been recruited when we evoked spiny neurons can express NMDAR-dependent LTD18, a form of EPSCs at +40 mV. An initial finding that led to the proposal of synaptic plasticity that is associated with a reduction in AMPAR silent synapses is that the coefficient of variation (CV) of EPSCs function and number17,19,20, we predicted that the expression of was higher for the AMPAR component of EPSCs relative to the LTD would be reduced in cocaine-treated mice. We first tested NMDAR component15. When we examined CV ratios in shell whether the triggering of LTD at NAc synapses was significantly impaired in cocaine-treated mice by delivering an LTD-inducing neurons from cocaine and saline-treated mice, we found that in train of stimuli (5 Hz, 3 min) while monitoring field EPSPs. This 13 of 19 cells examined, the AMPAR/NMDAR CV ratio was induced a modest amount of LTD that did not differ between the greater than 1, suggesting that some synapses on a proportion of two groups (cocaine, 79 ± 6% of baseline, n = 6; saline, 85 ± 5% of medium spiny neurons may contain only NMDARs. However, baseline, n = 6; data not shown). To determine whether the magthere was no significant difference between the groups in this nitude of LTD was reduced, as would be expected if the decrease in ratio (cocaine, 1.39 ± 0.16, n = 11; saline, 1.15 ± 0.09, n = 8; p > synaptic strength in cocaine-treated mice was due to mechanisms 0.05). As a final test to determine whether the function or numshared with LTD, we made whole-cell recordings from shell neuber of NMDARs increased in cocaine-treated mice, we bathrons and used a strong LTD induction protocol that generated a applied NMDA (10 µM, 30 s) and recorded the change in holding near-saturating amount of LTD (3 bouts of 5 Hz, current (Fig. 5e). Again, there was no significant difference 3 min stimulation paired with depolarization to –50 mV). This between cocaine and saline-treated groups (cocaine, 82 ± 16 pA, elicited robust LTD in saline-treated animals (Fig. 7; n = 5; saline, 76 ± 17 pA, n = 4; p > 0.05). 1.0

4

Change in holding current (pA)


zero Mg + APV

Fig. 5. Repeated cocaine administration did not affect NMDAR mEPSC amplitude or response to NMDA in shell neurons. (a) Samples of mEPSCs recorded at –65 mV in zero Mg2+ solution in the absence (top) and presence (bottom) of D-APV (50 µM) in a cocaine-treated mouse. Scale bars, 100 ms and 20 pA. (b) Sample averaged traces of mEPSCs obtained in each condition plus the subtracted trace that yielded an average NMDAR mEPSC. Scale bars, 10 ms and 2 pA. (c) Cumulative probability plot for NMDAR mEPSC amplitude (n = 9 cocaine; n = 7 saline). (d) Means of NMDAR mEPSCs amplitude (p > 0.05). (e) Bath application of NMDA (10 µM) for 30 s elicited similar change in holding current in neurons held at +40 mV (cocaine, n = 5; saline, n = 4; p > 0.05).

1220



DISCUSSION

saline

Amplitude (pA)

–500 –400 –300 –200 –100 0 0

Amplitude (pA)

10

20

30

40

50

60

70

30

40

50

60

70

cocaine

–500 –400 –300 –200 –100 0 0

10

20

Time (min)

b

140

saline cocaine

120

100

80

60

0 0

10

20

30

40

50

Time (min)


60

2+

Sr

c saline cocaine

1.0

Psychostimulant-induced behavioral sensitization is thought to model some of the core features of addiction2–4, as well as the development of drug-elicited psychosis21. It is of interest not only because it serves as a model for the pathological changes that occur as a consequence of repeated exposure to drugs of abuse but also because it is a robust form of experience-dependent behavioral plasticity. Indeed, sensitized animals can remain hypersensitive to the locomotor and rewarding actions of drugs for periods of months to years2–4. Over the last few years, evidence has accumulated that modifications in excitatory neural circuitry in the VTA and NAc may be particularly important in this form of experience-dependent plasticity. However, excitatory synaptic transmission following repeated drug exposure has not been examined directly. In this study, we measured excitatory synaptic responses in slices of the NAc prepared from animals in which behavioral

a

2+

Ca

b Cumulative probability

66 ± 7% of baseline, n = 7), but much smaller LTD in the cocaine group (Fig. 7; 84 ± 4% of baseline, n = 8, p < 0.05).

a

15

Mean mEPSC amplitude (pA)

Fig. 6. Repeated cocaine treatment decreased the amplitude of AMPAR quantal EPSCs in cortical synapses onto NAc shell neurons. (a) Sample traces of evoked EPSCs in the presence of Ca2+ or Sr2+. Scale bars, 50 ms and 10 pA. (b) Cumulative amplitude distributions of AMPAR quantal EPSCs in Sr2+ solution for saline and cocaine-treated mice (n = 300 events per cell; saline, n = 11; cocaine, n = 12). Distributions are significantly different (p < 0.05; Kolmogorov–Smirnov test). (c) Mean amplitude of quantal EPSCs in Sr2+ solution for saline and cocaine-treated mice. Cocaine treatment induced a significant decrease in Sr2+-induced AMPAR mEPSC amplitude (p < 0.05; t-test).

EPSC amplitude (% of baseline)


articles

0.8

0.6

0.4

0.2

14

*

13 12 11 10 9 0

0.0 5

10

15

20

25

30

35

saline

cocaine

Amplitude (pA)

sensitization was induced by repeated in vivo treatment with cocaine. We found that cocaine treatment decreased synaptic strength at excitatory synapses made by cortical afferents onto medium spiny neurons and that this occurred in the shell region of the NAc but not in the core. Changes in NMDAR-mediated synaptic responses and in the probability of transmitter release were not detected. Consistent with the idea that the cocaineinduced decrease in synaptic strength shares expression mechanisms with LTD, we found that the magnitude of LTD was significantly reduced in the cocaine-treated mice. Our observations are consistent with several previous findings. First, the single-unit responses of NAc neurons to glutamate are decreased for at least 14 days following a cocaine or amphetamine sensitization regimen22. Although a decrease in AMPAR function and/or number, as our results suggest, could explain this result, changes in voltage-dependent conductances following chronic cocaine exposure 23 may also contribute. Second, a decrease in the levels of the AMPAR subunits, GluR1 and GluR2 is observed in the NAc 14 days following amphetamine-induced sensitization24,25 (but see ref. 26). Third, the inward current generated by AMPA/kainate application to acutely dissociated striatal neurons is decreased in cells from animals that have received chronic cocaine27. Fourth, viral-mediated overexpression of GluR2 in the NAc, a manipulation that would be expected to decrease AMPAR function, increases behavioral sensitivity to cocaine as measured by place conditioning28. The functional importance of

Fig. 7. Repeated cocaine treatment decreased the magnitude of LTD in shell neurons. (a) Examples of individual experiments in saline (top) and cocaine-treated mice (bottom) displaying the time course of EPSCs before and after 3 bouts of 5 Hz, 3 min synaptic stimulation paired with depolarization of the cell to –50 mV. Traces shown were collected during the baseline period and 20 min following the last bout. Scale bars, 20 ms and 50 pA. (b) Summary graph of the average LTD elicited in saline (n = 7) and cocaine-treated (n = 8) mice. LTD was significantly decreased in cocaine-treated animals compared with saline animals (p < 0.05). 1221



articles

a decrease in the strength of excitatory inputs to the NAc is supported by the findings that virtually all drugs of abuse decrease the firing of NAc neurons29 and that such decreases have been associated with enhanced locomotor activity30. Furthermore, animals will self-administer glutamate receptor antagonists directly into the NAc shell31, suggesting that decreases in excitatory drive are reinforcing. It has been proposed that cocaine-induced behavioral sensitization involves an enhancement, not a depression, of excitatory synaptic transmission in the NAc4. This conclusion, however, is based primarily on microdialysis measurements of extracellular glutamate concentration, an assay that is not a direct measure of synaptic strength. The NAc is commonly divided into two regions: the core, which is considered a functional extension of the dorsal striatum, and the shell, which is thought to be a transitional region between the striatum and the extended amygdala32. There is evidence that the shell is particularly important for mediating the effects of drugs of abuse as well as behavioral sensitization. Certain drugs of abuse are preferentially self-administered in the shell compared to core31,33,34. Similarly, drug-induced increases in extracellular dopamine levels may preferentially occur in the shell35,36. It has also been reported that injections of D1 receptor antagonists into the shell reduce the reinforcing effects of cocaine37. In the context of drug-induced locomotor activity, three weeks after repeated cocaine exposure, injection of amphetamine into the shell, but not the core, elicits sensitized locomotor responses38. Conversely, lesions of the shell markedly impair the locomotor effects of amphetamine as well as its rewarding properties39. There is also evidence, however, that excitatory synaptic transmission in the core is important for the expression of behavioral sensitization3,4 and thus, our results should not be taken to indicate that modifications in this structure are not also important. All of our findings are consistent with the hypothesis that chronic in vivo cocaine administration induces a form of LTD that shares expression mechanisms with the NMDAR-dependent LTD previously described at these synapses18. We were unable to detect any change in NMDAR properties but we would caution that our assays might have been relatively insensitive, especially if such changes are restricted to synapses formed by a subset of NAc afferents. The relationship between our findings and the increase in spine density and proportion of branched spines that occurs following psychostimulant-induced sensitization40,41 is unclear. Although an increase in spines would suggest that the total afferent input to medium spiny neurons was increased, nothing is known about the detailed properties of these new spines. Indeed, if they contained small numbers of AMPARs or were functionally silent, they could contribute to our observed decrease in the AMPAR/NMDAR ratio. We have demonstrated that an LTD-like process in the NAc may contribute to the reorganization of neural circuitry that underlies behavioral sensitization to cocaine, and thus may be an important factor in the development of addiction. Also, the decrease in the ability of synaptic activity to elicit LTD in the NAc shell may contribute to the drug-induced behavioral changes. Together with the recent demonstration of a cocaine-induced LTP in the VTA6, these results suggest that like other forms of experience-dependent behavioral plasticity42–46, drug-induced changes in behavior may be due, at least in part, to modifications of synaptic efficacy in critical neural circuits. The sort of approaches taken here should further our understanding of how the molecular adaptations caused by drugs of abuse lead to the changes in the behavior of neural circuitry that ultimately must underlie addiction. Furthermore, because of shared molecular and circuit mechanisms, drug-induced behavioral changes such 1222

as psychostimulant-induced behavioral sensitization offer relatively simple models for understanding the neural mechanisms underlying many forms of experience-dependent plasticity, including learning and memory1–4,47.

METHODS Treatment regimen and locomotor activity. Male C57/Bl6 mice (24–26 days old) were given intraperitoneal injections of either saline (0.9% NaCl) or saline with cocaine (15 mg/kg). Immediately following each injection, horizontal locomotor activity was monitored in open-field chambers (Med Associates, St. Albans, Vermont) for 15 min. After two days of saline injections, mice were divided into groups that received five daily injections of either cocaine or saline. Following 10–14 days without injections, both groups received cocaine injections and locomotor activity was assessed. Brain slices were prepared on the following day. All procedures were approved by the Institutional Animal Care and Use Committee. Electrophysiology. Sagittal slices of the NAc (200–250 µm) were prepared with a vibratome (Leica, Nussloch, Germany) as described10. Slices (four per animal) were placed in a holding chamber and allowed to recover for at least 1 h before being placed in the recording chamber and superfused with bicarbonate-buffered solution (ACSF) saturated with 95% O 2 /5% CO 2 and containing 119 mM NaCl, 2.5 mM KCl, 1.0 mM NaH 2 PO 4 , 1.3 mM MgCl 2 , 2.5 mM CaCl 2 , 26.2 mM NaHCO 3 and 11 mM glucose (at 28–30°C). Picrotoxin (100 µM) was added to block GABAA receptor-mediated IPSCs. Cells were visualized using infrareddifferential interference contrast video microscopy. Whole-cell voltageclamp recordings were made using an Axopatch1D amplifier (Axon Instruments, Foster City, California). Electrodes (3–8 MΩ) contained 117 mM cesium gluconate, 2.8 mM NaCl, 20 mM HEPES, 0.4 mM EGTA, 5 TEA-Cl, 2.5 MgATP, and 0.25 MgGTP, pH 7.2–7.4 (285–295 mOsm) for whole-cell experiments and ACSF for field recordings. Series resistance (10–40 MΩ) and input resistance were monitored on-line with a 4-mV depolarizing step (50 ms) given with every afferent stimulus. Medium spiny neurons were identified by their morphology and high resting membrane potential (–75 to –85 mV). We examined core neurons in slices in which rostral and caudal limbs of the anterior commisure as well as caudate/putamen were present. Shell neurons were examined in medial NAc slices that did not contain dorsal striatal tissue. Stainless steel bipolar microelectrodes were placed at the prelimbic cortex–NAc border to stimulate afferents preferentially from the prelimbic cortex. Afferents were stimulated at 0.1 Hz except where noted. Neurons were voltage-clamped at –80 mV except where noted. Data were filtered at 1–2 kHz, digitized at 2–5 kHz and collected on-line using custom software (Igor Pro; Wavemetrics, Lake Oswego, Oregon). EPSC amplitudes were calculated by taking the mean of a 1–2 ms (AMPAR EPSCs) or 3–4 ms (NMDAR EPSCs) window around the peak and comparing this with the mean of an 8-ms window immediately before the stimulation artifact. AMPAR/NMDAR ratios were computed by taking the average of EPSCs at +40 mV (25–30 EPSCs) in the absence and presence of D-APV (50 µM). The average response in the presence of D-APV (AMPAR-only) was subtracted from that seen in its absence and an average NMDAR EPSC was calculated. The peak of the AMPAR EPSC was divided by the peak of the NMDAR EPSC to yield an AMPAR/NMDAR ratio. To compute AMPAR and NMDAR CVs, short latency (AMPAR; 2–4 ms following stimulus) and long latency (NMDAR; 45–50 ms following stimulus) amplitude measurements (25–30 per cell) were made at +40 mV and amplitude variances were divided by amplitude means after subtracting the variance of the recording noise. Miniature EPSCs (300–600 per cell) were collected in the presence of either tetrodotoxin (TTX, 1.5 µM) or lidocaine hydrochloride (0.6–0.8 mM). Dual-component mEPSCs were collected in the additional presence of 20 µM glycine, in the absence of added Mg2+ and at a holding potential of –65 mV. In strontium experiments, AMPAR-mediated quantal events were collected during a 400-ms period beginning 50 ms following each stimulus of a 2-Hz, 10-pulse train delivered once every 30 s in bath solution containing D-APV (50 µM), 2.6 mM MgCl2, 2.5 mM Sr2+ and no Ca2+. Quantal events were analyzed using Minianalysis software (Synaptosoft, Decatur, Georgia) with detection paranature neuroscience • volume 4 no 12 • december 2001



articles

meters set at greater than 4 pA amplitude and less than 3 ms rise time and verified by eye. For each cell, a random stretch of 300 mEPSCs was taken for constructing cumulative probability plots and calculating mean mEPSC amplitudes. LTD was induced by delivering a single train (5 Hz, 3 min; field experiments) or multiple trains (3 × 5 Hz, 3 min, 5 min inter-train interval) paired with depolarization to –50 mV. In over 90% of experiments, data acquisition and analysis were performed blindly without knowledge of the treatment history of the slices. On over 75% of recording days, a similar amount of data was collected from both cocaine and saline groups. Results are presented as mean ± s.e.m. Statistical significance was assessed using two-tailed Student’s t-tests, ANOVA for paired-pulse experiments, or Kolmogorov–Smirnov tests for comparing the cumulative probability histograms. Traces in figures have had stimulus artifacts removed and are averages of 10–12 consecutive responses.

ACKNOWLEDGEMENTS This work was supported by grants from NIDA (R.M., M.T.) and the State of California for medical research on alcohol and substance abuse through the University of California, San Francisco (A.B.). We thank T. Robinson, S. Nicola and G. Hjelmstad for comments on the paper and D. Saal for help with some experiments.

RECEIVED 6 AUGUST; ACCEPTED 16 OCTOBER 2001 1. Nestler, E. J. Molecular basis of long-term plasticity underlying addiction. Nat. Rev. Neurosci. 2, 119–128 (2001). 2. Robinson, T. E. & Berridge, K. C. The neural basis of drug craving: an incentive-sensitization theory of addiction. Brain. Res. Brain. Res. Rev. 18, 247–291 (1993). 3. Wolf, M. E. The role of excitatory amino acids in behavioral sensitization to psychomotor stimulants. Prog. Neurobiol. 54, 679–720 (1998). 4. Vanderschuren, L. J. & Kalivas, P. W. Alterations in dopaminergic and glutamatergic transmission in the induction and expression of behavioral sensitization: a critical review of preclinical studies. Psychopharmacology (Berl.) 151, 99–120 (2000). 5. Schenk, S. & Snow, S. Sensitization to cocaine’s motor activating properties produced by electrical kindling of the medial prefrontal cortex but not of the hippocampus. Brain Res. 659, 17–22 (1994). 6. Ungless, M. A., Whisler, J. L., Malenka, R. C. & Bonci, A. Single cocaine exposure in vivo induces long-term potentiation in dopamine neurons. Nature 411, 583–587 (2001). 7. Pierce, R. C., Bell, K., Duffy, P. & Kalivas, P. W. Repeated cocaine augments excitatory amino acid transmission in the nucleus accumbens only in rats having developed behavioral sensitization. J. Neurosci. 16, 1550–1560 (1996). 8. Pierce, R. C., Reeder, D. C., Hicks, J., Morgan, Z. R. & Kalivas, P. W. Ibotenic acid lesions of the dorsal prefrontal cortex disrupt the expression of behavioral sensitization to cocaine. Neuroscience 82, 1103–1114 (1998). 9. Li, Y. & Wolf, M. E. Ibotenic acid lesions of prefrontal cortex do not prevent expression of behavioral sensitization to amphetamine. Behav. Brain. Res. 84, 285–289 (1997). 10. Nicola, S. M., Kombian, S. B. & Malenka, R. C. Psychostimulants depress excitatory synaptic transmission in the nucleus accumbens via presynaptic D1-like dopamine receptors. J. Neurosci. 16, 1591–1604 (1996). 11. Zahm, D. S. Functional-anatomical implications of the nucleus accumbens core and shell subterritories. Ann. NY Acad. Sci. 877, 113–128 (1999). 12. Zucker, R. S. Short-term synaptic plasticity. Annu. Rev. Neurosci. 12, 13–31 (1989). 13. Sah, P., Hestrin, S. & Nicoll, R. A. Tonic activation of NMDA receptors by ambient glutamate enhances excitability of neurons. Science 246, 815–818 (1989). 14. Malenka, R. C. & Nicoll, R. A. Silent synapses speak up. Neuron 19, 473–476 (1997). 15. Kullmann, D. M. Amplitude fluctuations of dual-component EPSCs in hippocampal pyramidal cells: implications for long-term potentiation. Neuron 12, 1111–1120 (1994). 16. Goda, Y. & Stevens, C. F. Two components of transmitter release at a central synapse. Proc. Natl. Acad. Sci. USA 91, 12942–12946 (1994). 17. Oliet, S. H., Malenka, R. C. & Nicoll, R. A. Bidirectional control of quantal size by synaptic activity in the hippocampus. Science 271, 1294–1297 (1996). 18. Thomas, M. J., Malenka, R. C. & Bonci, A. Modulation of long-term depression by dopamine in the mesolimbic system. J. Neurosci. 20, 5581–5586 (2000). 19. Carroll, R. C., Beattie, E. C., von Zastrow, M. & Malenka, R. C. Role of AMPA receptor endocytosis in synaptic plasticity. Nat. Rev. Neurosci. 2, 315–324 (2001).


20. Lee, H. K., Barbarosie, M., Kameyama, K., Bear, M. F. & Huganir, R. L. Regulation of distinct AMPA receptor phosphorylation sites during bidirectional synaptic plasticity. Nature 405, 955–959 (2000). 21. Segal, D. S. & Schuckit, M. A. in Stimulants: Neurochemical, Behavioral and Clinical Perspectives (ed. Creese, I.) 131–167 (Raven, New York, 1983). 22. White, F. J., Hu, X. T., Zhang, X. F. & Wolf, M. E. Repeated administration of cocaine or amphetamine alters neuronal responses to glutamate in the mesoaccumbens dopamine system. J. Pharmacol. Exp. Ther. 273, 445–454 (1995). 23. Zhang, X. F., Hu, X. T. & White, F. J. Whole-cell plasticity in cocaine withdrawal: reduced sodium currents in nucleus accumbens neurons. J. Neurosci. 18, 488–498 (1998). 24. Lu, W., Chen, H., Xue, C. J. & Wolf, M. E. Repeated amphetamine administration alters the expression of mRNA for AMPA receptor subunits in rat nucleus accumbens and prefrontal cortex. Synapse 26, 269–280 (1997). 25. Lu, W. & Wolf, M. E. Repeated amphetamine administration alters AMPA receptor subunit expression in rat nucleus accumbens and medial prefrontal cortex. Synapse 32, 119–131 (1999). 26. Churchill, L., Swanson, C. J., Urbina, M. & Kalivas, P. W. Repeated cocaine alters glutamate receptor subunit levels in the nucleus accumbens and ventral tegmental area of rats that develop behavioral sensitization. J. Neurochem. 72, 2397–2403 (1999). 27. Bibb, J. A. et al. Effects of chronic exposure to cocaine are regulated by the neuronal protein Cdk5. Nature 410, 376–380 (2001). 28. Kelz, M. B. et al. Expression of the transcription factor deltaFosB in the brain controls sensitivity to cocaine. Nature 401, 272–276 (1999). 29. Wise, R. A. Drug-activation of brain reward pathways. Drug Alcohol. Depend. 51, 13–22 (1998). 30. Pennartz, C. M., Groenewegen, H. J. & Lopes da Silva, F. H. The nucleus accumbens as a complex of functionally distinct neuronal ensembles: an integration of behavioural, electrophysiological and anatomical data. Prog. Neurobiol. 42, 719–761 (1994). 31. Carlezon, W. A. Jr. & Wise, R. A. Rewarding actions of phencyclidine and related drugs in nucleus accumbens shell and frontal cortex. J. Neurosci. 16, 3112–3122 (1996). 32. McGinty, J. F., ed. Advancing from the Ventral Striatum to the Extended Amygdala: Implications for Neuropsychiatry and Drug Abuse Vol. 877 (New York Academy of Sciences, New York, New York, 1999). 33. Carlezon, W. A. Jr., Devine, D. P. & Wise, R. A. Habit-forming actions of nomifensine in nucleus accumbens. Psychopharmacology (Berl.) 122, 194–197 (1995). 34. McKinzie, D. L., Rodd-Henricks, Z. A., Dagon, C. T., Murphy, J. M. & McBride, W. J. Cocaine is self-administered into the shell region of the nucleus accumbens in Wistar rats. Ann. NY Acad. Sci. 877, 788–791 (1999). 35. Pontieri, F. E. et al. Psychostimulant drugs increase glucose utilization in the shell of the rat nucleus accumbens. Neuroreport 5, 2561–2564 (1994). 36. Pontieri, F. E., Tanda, G. & Di Chiara, G. Intravenous cocaine, morphine, and amphetamine preferentially increase extracellular dopamine in the “shell” as compared with the “core” of the rat nucleus accumbens. Proc. Natl. Acad. Sci. USA 92, 12304–12308 (1995). 37. Caine, S. B., Heinrichs, S. C., Coffin, V. L. & Koob, G. F. Effects of the dopamine D-1 antagonist SCH 23390 microinjected into the accumbens, amygdala or striatum on cocaine self-administration in the rat. Brain Res. 692, 47–56 (1995). 38. Pierce, R. C. & Kalivas, P. W. Amphetamine produces sensitized increases in locomotion and extracellular dopamine preferentially in the nucleus accumbens shell of rats administered repeated cocaine. J. Pharmacol. Exp. Ther. 275, 1019–1029 (1995). 39. Parkinson, J. A., Olmstead, M. C., Burns, L. H., Robbins, T. W. & Everitt, B. J. Dissociation in effects of lesions of the nucleus accumbens core and shell on appetitive pavlovian approach behavior and the potentiation of conditioned reinforcement and locomotor activity by D-amphetamine. J. Neurosci. 19, 2401–2411 (1999). 40. Robinson, T. E. & Kolb, B. Persistent structural modifications in nucleus accumbens and prefrontal cortex neurons produced by previous experience with amphetamine. J. Neurosci. 17, 8491–8497 (1997). 41. Robinson, T. E. & Kolb, B. Alterations in the morphology of dendrites and dendritic spines in the nucleus accumbens and prefrontal cortex following repeated treatment with amphetamine or cocaine. Eur. J. Neurosci. 11, 1598–1604 (1999). 42. Rogan, M. T., Staubli, U. V. & LeDoux, J. E. Fear conditioning induces associative long-term potentiation in the amygdala. Nature 390, 604–607 (1997). 43. McKernan, M. G. & Shinnick-Gallagher, P. Fear conditioning induces a lasting potentiation of synaptic currents in vitro. Nature 390, 607–611 (1997). 44. Moser, E. I., Krobert, K. A., Moser, M. B. & Morris, R. G. Impaired spatial learning after saturation of long-term potentiation. Science 281, 2038–2042 (1998). 45. Andersen, P., Moser, E., Moser, M. B. & Trommald, M. Cellular correlates to spatial learning in the rat hippocampus. J. Physiol. (Paris) 90, 349 (1996). 46. Rioult-Pedotti, M. S., Friedman, D. & Donoghue, J. P. Learning-induced LTP in neocortex. Science 290, 533–536 (2000). 47. Hyman, S. E. & Malenka, R. C. Addiction and the brain: the neurobiology of compulsion and its persistence. Nat. Rev. Neurosci. 2, 695–703 (2001).

1223


articles


Endogenous nicotinic cholinergic activity regulates dopamine release in the striatum Fu-Ming Zhou, Yong Liang and John A. Dani Division of Neuroscience, Baylor College of Medicine, Houston, Texas 77030-3498, USA Correspondence should be addressed to J.A.D. ([email protected])

Published online: 19 November 2001, DOI: 10.1038/nn769 Dopamine is vital for coordinated motion and for association learning linked to behavioral reinforcement. Here we show that the precise overlap of striatal dopaminergic and cholinergic fibers underlies potent control of dopamine release by ongoing nicotinic receptor activity. In mouse striatal slices, nicotinic antagonists or depletion of endogenous acetylcholine decreased evoked dopamine release by 90%. Nicotine at the concentration experienced by smokers also regulated dopamine release. In mutant mice lacking the β2 nicotinic subunit, evoked dopamine release was dramatically suppressed, and those mice did not show cholinergic regulation of dopamine release. The results offer new perspectives when considering nicotine addiction and the high prevalence of smoking in schizophrenics.

Dopaminergic mechanisms of the striatum are intimately involved in motor coordination, complex issues of behavioral reinforcement, and disorders such as schizophrenia and Parkinson’s disease1–7. The striatum receives the densest dopaminergic innervation in the mammalian brain, which arises from neurons located in the substantia nigra (SN) and ventral tegmental area (VTA) of the midbrain8,9. In addition, the striatum is densely innervated by local cholinergic interneurons10,11 that are tonically active and release acetylcholine (ACh)12,13. Histochemical studies have indicated that nicotinic acetylcholine receptors (nAChRs) are expressed on dopaminergic nerve terminals in the striatum14-17. Exogenous nicotinic agonists affect dopamine release in the striatum18,19, but the action of endogenous ACh release is not well understood. To investigate cholinergic influence over dopamine release, carbon-fiber microelectrodes were placed into mice striatal brain slices, and fast cyclic voltammetry was used to monitor the concentration of action-potential-dependent dopamine release in real time. We found that cholinergic interneurons acting via nAChRs containing the β2 subunit potently regulate dopamine release. Furthermore, in the concentration range experienced by smokers, nicotine acts within the striatum and influences evoked dopamine release. Because the dopaminergic and nicotinic mechanisms of the striatum are so intimately linked, the results have broad implications for understanding nicotine addiction and the predilection for smoking by schizophrenic patients.

RESULTS Dopamine and ACh fibers overlap in the striatum There is dense expression of tyrosine hydroxylase (TH, indicating dopamine synthesis), choline acetyltransferase (ChAT, indicating ACh synthesis), and acetylcholinesterase (AChE, indicating ACh degradation) in the striatum20. The anatomical 1224

distribution of TH, ChAT and AChE was determined in coronal sections (Fig. 1). The highest density of each enzyme was in the striatum, extending dorsally into the olfactory tubercle. There is precise overlap in the distribution of the enzymes, and at the highest magnification individual cholinergic interneurons are seen embedded within the intertwined dopamine and ACh fibers (Fig. 1b). The precise overlap and close proximity of dopamine fibers and cholinergic enzymes led us to the following hypothesis. The high ChAT activity suggests ACh must be released often, and the high AChE indicates that ACh must be broken down extremely rapidly. The two enzymes together suggest fast nicotinic mechanisms. The nAChRs are very susceptible to desensitization21, which can be avoided if the repeatedly released ACh is removed rapidly by the abundant AChE. Depletion of ACh reduces dopamine release Fast-scan cyclic voltammetry with carbon-fiber microelectrodes was used to monitor dopamine release on the subsecond time scale. A bipolar simulating electrode was placed in the striatum about 150 µm from the carbon-fiber microelectrode. Normally, dopamine release was electrically evoked every 2.5 min at 60% of the maximal response. Under those conditions, the dopamine signal was stable for over two hours. The evoked dopamine concentration at the tip of the carbon-fiber microelectrode was estimated to be 1.62 ± 0.12 µM (mean ± s.e.m., n = 29) in the dorsal striatum, 1.46 ± 0.13 µM (n = 18) in the nucleus accumbens (NAc) core, and 1.44 ± 0.13 µM (n = 17) in the NAc shell. In addition to electrically evoked dopamine release, we also measured spontaneous dopamine release in the slice. Both the spontaneous and evoked dopamine release was prevented by 0.5 µM tetrodotoxin (n = 4) and by removing Ca2+ (n = 3), indicating the release was Ca2+ dependent and action potential dependent. nature neuroscience • volume 4 no 12 • december 2001


articles


a

b

Fig. 1. Dense and overlapping distribution of ACh and dopamine in the striatum. (a) Bright-field photomicrographs show TH, ChAT and AChE in the NAc and dorsal striatum as demonstrated by TH and ChAT antibody staining and AChE histochemical staining. Arrows, anterior commissure. CC, corpus callosum, CPu, caudate putamen, NAc, nucleus accumbens, S, septum. (b) Pictures (left) taken from one slice show immunofluorescence double labeling for TH (green) and ChAT (red) at low magnification. The area in the white boxes is expanded to high magnification (right), revealing dense fiber tracts and two cholinergic interneurons. Scale bars, 50 µm.

To test for the importance of cholinergic activity, we prevented endogenous ACh release by depleting ACh stores with (±)-vesamicol, a well characterized inhibitor of vesicular ACh transport that consequently exhausts ACh release 22 . The (±)-vesamicol (2 µM) presumably diminished ACh release and inhibited evoked dopamine release by 81 ± 4.5% (n = 6; Fig. 2a and b). The maximum inhibition was 90.5 ± 4.3% in 5 µM or 10 µM (±)-vesamicol (n = 12). The median inhibitory concentration (IC50) was estimated to be 1 µM (Fig. 2c). These results show that intact cholinergic activity is necessary for normal dopamine release induced by action potentials. Nicotinic activity facilitates dopamine release When muscarinic acetylcholine receptors were inhibited by 1 or 2 µM atropine, the evoked dopamine release slightly increased by 4.8 ± 3.2% (n = 6; Fig. 3a). In contrast, the non-specific nicotinic antagonist mecamylamine a (1 µM) decreased evoked dopamine release by 83 ± 4.8%

Fig. 2. Inhibition of ACh vesicular transport by vesamicol reduces dopamine release in the striatum. (a) The three dopamine responses were electrically evoked under control conditions, during 2 µM (±)-vesamicol application, and after recovery following 2 h of wash. The dopamine responses were constructed from voltammograms that were obtained at the rate of 10 Hz. The voltammogram (right) was obtained at the peak of evoked dopamine response under control conditions. The same results were obtained with or without atropine (1–2 µM, data not shown). (b) Summary of the effect of 2 µM (±)vesamicol (n = 6). (c) The concentration dependence for inhibition of dopamine release produced by (±)-vesamicol was fitted by an Hill equation with IC50 of 1 µM. nature neuroscience • volume 4 no 12 • december 2001

b

(n = 5). Maximal inhibition of dopamine release was by 92 ± 3.7% (n = 16) in 5 to 20 µM mecamylamine (Fig. 3b). Furthermore, mecamylamine’s inhibition of dopamine release was not altered by atropine (n = 3). In the absence of electrical stimulation, spontaneous, action-potential-dependent dopamine release could be monitored (Fig. 3c). When nAChRs were inhibited by mecamylamine (1 or 5 µM), the spontaneous dopamine release was inhibited below our resolution (n = 6). Microdialysis studies and measures of radioactive dopamine release from minced striatal brain slices indicated that nicotine increases basal or ambient dopamine levels, and that increase was resistant to dihydro-β-erythroidine (DHβE) and to blockade of action potentials by tetrodotoxin23. Fast cyclic voltammetry measures dopamine release driven by action potentials on a fast time scale, which is different from the measures of basal dopamine obtained from samples taken over much longer times24. Because of the importance for addiction25, we bath-applied nicotine and measured the effect on evoked dopamine. Under our experimental conditions, 50 nM nicotine decreased evoked dopamine release by 73 ± 4.8% (n = 4; Fig. 4a and b). The maximum inhibition was 90.5 ± 3.8% in 100 nM (n = 6) and 92.2 ± 3.5% in 500 nM (n = 5). The IC50 was estimated to be 30 nM (Fig. 4c). In the absence of electrical stimulation, the same concentrations of nicotine inhibited spontaneous, action-potential-dependent dopamine release (n = 4; Fig. 4d). AChE inhibition reduces dopamine release Based upon the distribution and the density of AChE in the striatum (Fig. 1), we reasoned that AChE might be important for the ongoing nAChR activity that enhances dopamine release. To test this idea, we used a potent AChE inhibitor, ambenonium26. Bath application of 0.1 µM (n = 6), 0.5 µM (n = 3) or 1 µM (n = 3) ambenonium gradually decreased dopamine release by 90.5 ± 2.6% (Fig. 5). The effect of ambenonium was reversed upon prolonged wash. This result demonstrates that AChE activity is essential for the ongoing nicotinic facilitation of dopamine release. By increasing extracellular ACh27, AChE inhibition may increase nAChR desensitization, as suggested by the results with bath-applied nicotine. This interpretation was supported by results obtained with a puffer pipette containing 1 mM ACh that was positioned near to the carbon-fiber voltammetry electrode. Under control conditions (without inhibition of AChE), puffs of ACh applied just before and during the electrical stimulation decreased dopamine release by 16 ± 2.4% (n = 6). This result

c

1225



articles

Fig. 3. Nicotinic but not muscarinic ACh receptors regulate dopamine release in the striatum. (a) The dopamine responses were evoked under control conditions, during 1 µM atropine application, and after recovery following a prolonged wash. The voltammogram (right) was obtained at the dopamine peak of the control trace. (b) The dopamine responses were evoked under control conditions, during 1 µM mecamylamine application, and after recovery following a prolonged wash. The voltammogram (right) is from the dopamine peak of the control trace. (c) Without electrical stimulation, spontaneous dopamine release is shown under control conditions and during application of 5 µM mecamylamine. The voltammogram was obtained at the peak of the event indicated by the arrow.

a

b

again is consistent with desensitization of nAChRs caused by c the excess ACh. Mecamylamine is a non-specific nAChR inhibitor (Fig. 3), but DHβE is a specific inhibitor of β2* nAChRs28,29. DHβE decreased dopamine release by 47 ± 5.3% (20 nM, n = 4) and 90 ± 5.2% (100 nM, n = 4; Fig. 6a and b), and maximum inhibition was 91.6 ± 4.3% (1,000 nM, n = 4). The IC50 was estimated to be 20 nM (Fig. 6c). The results were verified using mutant mice lacking the β2 subunit28,30. It was much more difficult to electrically evoke dopamine release from β2-null mice (Fig. 7a). The evoked dopamine concentration was only 0.31 ± 0.07 µM (n = 14), which is an 80% decrease relative to wild-type littermates. Furthermore, 100 nM and 1,000 nM DHβE (n = 4), 10 µM mecamylamine (n = 3, data not shown), and 1,000 nM nicotine (n = 6) no longer had any effect (Fig. 7b). Although electrically evoked dopamine release was depressed by elimination of the nAChR β2 subunit, dopamine release induced by a depolarizing solution of 30 mM KCl produced the same dopamine signal in slices from β2-null mice (37.6 ± 3.8 µM, n = 4) and wild-type mice (38.2 ± 3.4 µM, n = 6). This result indicates that the dopamine content in β2-null mice was normal. The effect of nAChRs is specific to the β2 subunit because inhibition of α7* nAChRs had no detectable effect. In the presence of the α7*-specific inhibitor, 20 nM methyllycaconitine29, evoked dopamine release was 98.6 ± 2.3% of control level (n = 3).

a

b

d

1226

c

DISCUSSION Striatal dopamine and ACh fibers form an intertwined meshwork that is the densest in the brain, and these fibers are associated with the densest expression of AChE8,9,11. Tonic activity of the cholinergic interneurons releases ACh12,13, and the AChE rapidly terminates the ACh signal. This situation optimizes ongoing nAChR activity by avoiding desensitization. Our results show that dopamine release caused by action potentials is potently regulated by β2* nAChR activity. Nicotine, at the concentrations achieved by smokers31, also decreases action-potential-dependent dopamine release in the NAc, suggesting that nicotine is acting like an antagonist by causing desensitization32. This hypothesis is reasonable because the β2* nAChRs that regulate dopamine release have a high affinity for nicotine and are readily desensitized by those concentrations33. The finding was unexpected, however, because older studies using in vivo microdialysis or loading of slices with radiolabeled dopamine have shown that nicotine increases the basal level of dopamine18,34. That increase, however, is at a very low concentration of dopamine and is often action potential independent. Microdialysis is a slow process, often taking 10 minutes per sample. The sample is collected over a relatively large volume with probes of about 250–500 µm diameter. The radiolabeled and microdialysis samples report dopamine released from multiple sources and provide an average baseline dopamine concentration that escapes reuptake or breakdown, giving estimates near 4 nM (ref. 36).

Fig. 4. Bath-applied nicotine reduces action-potential-dependent dopamine release. (a) The dopamine responses were electrically evoked under control conditions, during 50 nM nicotine application, and after recovery following a prolonged wash. The voltammogram (right) was obtained at the dopamine peak of the control trace. (b) Summary of the effect of 50 nM nicotine (n = 4). (c) The concentration dependence for inhibition of dopamine release produced by nicotine was fitted by an Hill equation with IC50 of 30 nM. (d) Without electrical stimulation, spontaneous dopamine release is shown under control conditions and during application of 50 nM nicotine. The voltammogram was obtained at the peak of the event indicated by the arrow. nature neuroscience • volume 4 no 12 • december 2001


articles

a


b

Fig. 5. An AChE inhibitor, ambenonium, reduces dopamine release in the striatum. (a) The dopamine responses were evoked under control conditions, during 100 nM ambenonium application, and after recovery following a prolonged wash. The voltammogram (right) was obtained at the dopamine peak of the control trace. (b) Summary of the effect of 100 nM ambenonium in 6 experiments, 3 of which lasted long enough to recover upon washing.

Fast voltammetry in our present study detects dopamine that is released by action potentials, and the measurement estimates dopamine concentration before uptake and diffusion have much effect. The voltammetry measurement is from a smaller volume, with an electrode of 10 µm diameter, and it detects dopamine signals at speeds and concentrations that are indicative of direct neuronal activity. These two very different measurements of extracellular dopamine dynamics suggest that nicotine has multiple actions when analyzed in the dopamine target area of the NAc. The unexpected result that nicotine strongly inhibits actionpotential-dependent dopamine release in the target has broad implications, particularly when considering nicotine addiction. A simplification of a commonly held view of nicotine addiction is the following: nicotine elevates dopamine in the NAc, and that elevation reinforces continued use32. However, our understanding of dopamine participation in reinforcement processes is far from complete. It is clear that dopamine in the NAc is not a direct indication of reward. More sophisticated theories suggest that the dopamine signal conveys novelty and/or prediction error during the ongoing process of learning adaptive behaviors as an animal continually updates a construct of environmental

a

a

b

c

Fig. 6. A specific inhibitor of β2* nAChRs, DHβE, potently reduces evoked dopamine release. (a) The dopamine responses were evoked under control conditions, during 0.1 µM DHβE application, and after recovery following a prolonged wash. The voltammogram (right) was obtained at the dopamine peak of the control trace. (b) Summary of the effect of 0.1 µM DHβE (n = 5). (c) The concentration dependence for inhibition of dopamine release produced by DHβE was fitted by an Hill equation with IC50 of 20 nM.

saliency1,7,32,34,37. The complexity of the dopamine signal is exemplified by a recent study38. Rats learned to press a lever causing intracranial self-stimulation of the midbrain dopamine areas. After learning the task, the rats continued self-simulations as if it were pleasurable, but the self-simulations no longer increased the dopamine concentration at the target (NAc). Dopamine was released only during the initial phase while the rats were learning. Thus, even when the dopamine neurons are stimulated, other regulatory processes can ultimately control dopamine release. Our results identify a nicotinic cholinergic mechanism that regulates dopamine release at the target. Ongoing β2* nAChRs activity in the NAc seems important for dopamine release driven by afferent action potentials, but nicotine desensitizes those β2* nAChRs35. Although this result is contrary to the simplest view of nicotine addiction, it offers a new clue to understand the high prevalence of smoking by schizophrenic patients. Schizophrenics have impaired voluntary or sustained attention, and the positive symptoms of schizophrenia, such as delusions and disorganized behavior, are associated with an excess of dopamine in the striatum39. Schizophrenic patients are usually treated with neuroleptics, which inhibit specific dopamine receptors and ultimately decrease dopamine signaling. Nicotine can transiently improve attention and some aspects of the positive symptoms in schizophrenic patients 40,41. Such nicotine-induced improvements are thought to account, at least partially, for the extraordinarily high rate of smoking observed in schizophrenics42,43. However, it was difficult to explain why nicotine could help schizophrenics if it were increasing dopamine levels. Our finding

b Fig. 7. β2-null mice have decreased dopamine release, and the release is not regulated by DHβE or nicotine. (a) The dopamine responses were evoked under control conditions, during 0.1 µM DHβE application, and during 1 µM nicotine application. The detected dopamine concentration was depressed in the β2-null mice to about 0.3 µM compared to about 1.5 µM in WT mice. The voltammogram (right) was obtained at the dopamine peak of the control trace. (b) Summary data showing the lack of effect by 0.1 or 1 µM DHβE and by 1 µM nicotine (n = 4). nature neuroscience • volume 4 no 12 • december 2001

1227


articles

that nicotine decreases action-potential-evoked dopamine release may underlie nicotine’s transient positive influence for schizophrenic patients.


METHODS Wild-type C57BL/6J (Jackson Laboratory, Bar Harbor, Maine) and β2-null28 mice were used at 3 to 6 months of age. Horizontal or coronal striatal slices (400 µm) were cut on a vibratome, held at room temperature, and studied at 30°C in 125 mM NaCl, 2.5 mM KCl, 1.3 mM MgCl2, 2.5 mM CaCl2, 26 mM Na2HPO3, 1.25 mM NaHCO3 and 10 mM glucose saturated with 95% O2 and 5% CO2. Horizontal slices appeared more robust than coronal sections, such that spontaneous dopamine events were recorded in about 10% of horizontal slices but none in the coronal slices. Animal care was in accord with Baylor College of Medicine’s animal care committee. Homemade carbon-fiber (P55S, Amoco Polymers, Greenville, South Carolina) microelectrodes were used for fast-scan cyclic voltammetry44. Stable, low-noise and sensitive electrodes were selected, especially for recording spontaneous dopamine release. An Axopatch 200B amplifier and pClamp 8 (Axon Instruments, Foster City, California) were used to acquire and analyze data. Scans of the microelectrode potential (12 ms duration, 10 Hz) were from 0 mV to –400 mV to 1,000 mV to –400 mV to 0 mV against an Ag/AgCl reference electrode at a rate of 300 mV/ms. The signal that formed the voltammogram was sampled at 50 kHz, and the net current due to the electrochemical reaction was obtained by digital subtraction before and after a stimulation. The voltammograms are of the proper form, having oxidation current peaks between 500 to 700 mV and the reduction current peaks between –230 to –330 mV. Dopamine is more than 90% of striatal monoamines45, and the voltammetry signal was calibrated against fresh solutions of 1–10 µM dopamine. Intrastriatal stimuli were delivered using bipolar tungsten electrodes with resistances of about 0.5 MΩ. The two tips of the stimulating electrode were about 100–200 µm away from each other. The tip of the recording electrode was about 150 µm away from the two tips of the stimulating electrode. For wild-type striatal slices, stimuli were 1 ms in duration and 2–6 V to achieve about 60% maximum amplitude, but the same results and conclusions were obtained with higher and lower stimulus intensities. For β2-null striatal slices, stimuli were 1 ms in duration and 6–12 V, and stimulation and recording sites were adjusted to obtain ≈ 80% maximum amplitude. ChAT and TH immunohistochemistry and AChE histochemistry were adapted from published methods46–48. For immunohistochemistry, brains were fixed in 0.1 M phosphate buffer containing 4% (w/v) paraformaldehyde and 14% (v/v) picric acid. Coronal sections (30 µm) were cut on a cryostat, incubated with either goat anti-ChAT or rabbit anti-TH antibodies, followed by peroxidae-conjugated antibodies, and were finally treated with chromogen diaminobenzidine. For ChAT and TH double immunofluorescence, after treatment with the primary antibody mixture, the secondary antibody mixture containing Cy2-conjugated donkey anti-rabbit and rhodamine-conjugated donkey anti-goat IgG antisera was used. For AChE staining, sections were incubated for 15 h in 100 ml solution containing 50 mM sodium acetate, 4 mM copper sulfate, 16 mM glycine, 116 mg S-acetylthiocholine iodide and 3 mg ethopropazine. The sections were then rinsed and developed with 1% sodium sulphide. All images were captured with a digital camera and processed in Adobe PhotoShop.

ACKNOWLEDGEMENTS We thank A. Beaudet and A. Orr-Urtreger for providing the mutant mice, and Y. Schmitz and D. Sulzer for advice with the techniques. The work was supported by the National Institute on Drug Abuse (DA09411 and DA12661), the National Institute of Neurological Disorders and Stroke (NS21229) and the NARSAD (to F.M.Z.).

RECEIVED 5 SEPTEMBER; ACCEPTED 29 OCTOBER 2001 1. Berke, J. D. & Hyman, S. E. Addiction, dopamine, and the molecular mechanisms of memory. Neuron 25, 515–532 (2000).

1228

2. Breiter, H. C. et al. Acute effects of cocaine on human brain activity and emotion. Neuron 19, 591–611 (1997). 3. Grace, A. A. Gating of information flow within the limbic system and the pathophysiology of schizophrenia. Brain Res. Rev. 31, 330–341 (2000). 4. Kalivas, P. W. & Nakamura, M. Neural systems for behavioral activation and reward. Curr. Opin. Neurobiol. 9, 223–227 (1999). 5. Graybiel, A. M., Aosaki, T., Flaherty, A. W. & Kimura, M. The basal ganglia and adaptive motor control. Science 265, 1826–1831 (1994). 6. Lang, A. E. & Lozano, A. M. Parkinson’s disease. N. Engl. J. Med. 339, 1130–1143 (1998). 7. Schultz, W., Dayan, P. & Montague, P. R. A neural substrate of prediction and reward. Science 275, 1593–1599 (1997). 8. Anden, N. E., Fuxe, K., Hamberger, B. & Hokfelt, T. A quantitative study on the nigro-neostriatal dopamine neuron system in the rat. Acta. Physiol. Scand. 67, 306–312 (1966). 9. Björklund, A. & Lindvall, O. in Classical Transmitters in the CNS Part I (eds. Björklund, A. & Hökfelt, T.) 55–122 (Elsevier, Amsterdam, 1984). 10. Butcher, L. L. & Woolf, N. J. in Classical Transmitters and Transmitter Receptors in the CNS (eds. Björklund, A., Hökfelt, T. & Kuhar, M. J.) 1–50 (Elsevier, Amsterdam, 1984). 11. Woolf, N. J. Cholinergic systems in mammalian brain and spinal cord. Prog. Neurobiol. 37, 475–524 (1991). 12. Aosaki, T., Kimura, M. & Graybiel, A. M. Temporal and spatial characteristics of tonically active neurons of the primate’s striatum. J. Neurophysiol. 73, 1234–1252 (1995). 13. Bennett, B. D. & Wilson, C. J. Spontaneous activity of neostriatal cholinergic interneurons in vitro. J. Neurosci. 19, 5586–5596 (1999). 14. Colquhoun, L. M. & Patrick, J. W. Pharmacology of neuronal nicotinic acetylcholine receptor subtypes. Adv. Pharmacol. 39, 191–220 (1997). 15. Hill, J. A. Jr., Zoli, M., Bourgeois, J. P. & Changeux, J. P. Immunocytochemical localization of a neuronal nicotinic receptor: the β2-subunit. J. Neurosci. 13, 1551–1568 (1993). 16. Schwartz, R. D., Lehmann, J. & Kellar, K. J. Presynaptic nicotinic cholinergic receptors labeled by [3H]acetylcholine on catecholamine and serotonin axons in brain. J. Neurochem. 42, 1495–1498 (1984). 17. Jones, I. W., Bolam, J. P. & Wonnacott, S. Presynaptic localisation of the nicotinic acetylcholine receptor beta2 subunit immunoreactivity in rat nigrostriatal dopaminergic neurones. J. Comp. Neurol. 439, 235–247 (2001). 18. Marshall, D. L., Redfern, P. H. & Wonnacott, S. Presynaptic nicotinic modulation of dopamine release in the three ascending pathways studied by in vivo microdialysis: comparison of naive and chronic nicotine-treated rats. J. Neurochem. 68, 1511–1519 (1997). 19. Johnson, J. H., Zhao, C., James, J. R. & Rosecrans, J. A. Individual variability of dopamine release from nucleus accumbens induced by nicotine. Brain Res. Bull. 51, 249–253 (2000). 20. Descarries, L., Gisiger, V. & Steriade, M. Diffuse transmission by acetylcholine in the CNS. Prog. Neurobiol. 53, 603–625 (1997). 21. McGehee, D. S. & Role, L. W. Physiological diversity of nicotinic acetylcholine receptors expressed by vertebrate neurons. Annu. Rev. Physiol. 57, 521–546 (1995). 22. Prior, C., Marshall, I. G. & Parsons, S. M. The pharmacology of vesamicol: an inhibitor of the vesicular acetylcholine transporter. Gen. Pharmacol. 23, 1017–1022 (1992). 23. Sacaan, A. I., Dunlop, J. L. & Lloyd, G. K. Pharmacological characterization of neuronal acetylcholine gated ion channel receptor-mediated hippocampal norepinephrine and striatal dopamine release from rat brain slices. J. Pharmacol. Exp. Ther. 274, 224–230 (1995). 24. Wu, Y., Pearl, S. M., Zigmond, M. J. & Michael, A. C. Inhibitory glutamatergic regulation of evoked dopamine release in striatum. Neuroscience 96, 65–72 (2000). 25. Dani, J. A. & Heinemann, S. Molecular and cellular aspects of nicotine abuse. Neuron 16, 905–908 (1996). 26. Hodge, A. S., Humphrey, D. R. & Rosenberry, T. L. Ambenonium is a rapidly reversible noncovalent inhibitor of acetylcholinesterase, with one of the highest known affinities. Mol. Pharmacol. 41, 937–942 (1992). 27. Vinson, P. N. & Justice, J. B. Jr. Effect of neostigmine on concentration and extraction fraction of acetylcholine using quantitative microdialysis. J. Neurosci. Methods 73, 61–67 (1997). 28. Xu, W. et al. Multiorgan autonomic dysfunction in mice lacking the beta2 and the beta4 subunits of neuronal nicotinic acetylcholine receptors. J. Neurosci. 19, 9298–9305 (1999). 29. Alkondon, M. & Albuquerque, E. X. Diversity of nicotinic acetylcholine receptors in rat hippocampal neurons. I. Pharmacological and functional evidence for distinct structural subtypes. J. Pharmacol. Exp. Ther. 265, 1455–1473 (1993). 30. Picciotto, M. R. et al. Acetylcholine receptors containing the β2 subunit are involved in the reinforcing properties of nicotine. Nature 391, 173–177 (1998). 31. Henningfield, J. E., Stapleton, J. M., Benowitz, N. L., Grayson, R. F. & London, E. D. Higher levels of nicotine in arterial than in venous blood after cigarette smoking. Drug Alcohol Depend. 33, 23–29 (1993). 32. Dani, J. A., Ji, D. & Zhou, F. M. Synaptic plasticity and nicotine addiction. Neuron 31, 349–352 (2001). 33. Fenster, C. P., Rains, M. F., Noerager, B., Quick, M. W. & Lester, R. A.



articles

34. 35. 36.


37. 38. 39. 40.

Influence of subunit composition on desensitization of neuronal acetylcholine receptors at low concentrations of nicotine. J. Neurosci. 17, 5747–5759 (1997). Di Chiara, G. Role of dopamine in the behavioural actions of nicotine related to addiction. Eur. J. Pharmacol. 393, 295–314 (2000). Mansvelder, H. D. & McGehee, D. S. Long-term potentiation of excitatory inputs to brain reward areas by nicotine. Neuron 27, 349–357 (2000). Parsons, L. H. & Justice, J. B. Jr. Extracellular concentration and in vivo recovery of dopamine in the nucleus accumbens using microdialysis. J. Neurochem. 58, 212–218 (1992). Balfour, D. J., Wright, A. E., Benwell, M. E. & Birrell, C. E. The putative role of extra-synaptic mesolimbic dopamine in the neurobiology of nicotine dependence. Behav. Brain Res. 113, 73–83 (2000). Garris, P. A. et al. Dissociation of dopamine release in the nucleus accumbens from intracranial self-stimulation. Nature 398, 67–69 (1999). Abi-Dargham, A. et al. Increased baseline occupancy of D2 receptors by dopamine in schizophrenia. Proc. Natl. Acad. Sci. USA 97, 8104–8109 (2000). Adler, L. E., Hoffer, L. D., Wiser, A. & Freedman, R. Normalization of auditory physiology by cigarette smoking in schizophrenic patients. Am. J. Psychiatry 150, 1856–1861 (1993).


41. Rezvani, A. H. & Levin, E. D. Cognitive effects of nicotine. Biol. Psychiatry 49, 258–267 (2001). 42. Adler, L. E. et al. Schizophrenia, sensory gating, and nicotinic receptors. Schizophr. Bull. 24, 189–202 (1998). 43. Dalack, G. W., Healy, D. J. & Meador-Woodruff, J. H. Nicotine dependence in schizophrenia: clinical phenomena and laboratory findings. Am. J. Psychiatry 155, 1490–1501 (1998). 44. Kawagoe, K., Zimmerman, J. B. & Wightman, R. M. Principles of voltammetry and the microelectrode surface states. J. Neurosci. Methods 48, 225–240 (1993). 45. Schenk, J. O., Miller, E., Rice, M. E. & Adams, R. N. Chronoamperometry in brain slices: quantitative evaluations of in vivo electrochemistry. Brain Res. 277, 1–8 (1982). 46. Chang, H. T. Dopamine-acetylcholine interaction in the rat striatum: a duallabeling immunocytochemical study. Brain Res. Bull. 21, 295–304 (1988). 47. Koelle, G. G. & Friedenwald, J. S. A histochemical method for localizing cholinesterase activity. Proc. Soc. Exp. Biol. Med. 70, 617–622 (1949). 48. Schwartz, M. L. & Mrzljak, L. Cholinergic innervation of the mediodorsal thalamic nucleus in the monkey: ultrastructural evidence supportive of functional diversity. J. Comp. Neurol. 327, 48–62 (1993).

1229


articles


Differential synaptic processing separates stationary from transient inputs to the auditory cortex Marco Atzori1, Saobo Lei1, D. Ieuan P. Evans1, Patrick O. Kanold2, Emily Phillips-Tansey1, Orinthal McIntyre1 and Chris J. McBain1 1 LCMN/NICHD/NIH, Rm 5A72, Bldg 49, Convent Drive, Bethesda, Maryland 20892-4495, USA 2 Department of Biomedical Engineering, Johns Hopkins University, Baltimore, Maryland 21205, USA

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

Published online: 5 November 2001, DOI: 10.1038/nn760 Sound features are blended together en route to the central nervous system before being discriminated for further processing by the cortical synaptic network. The mechanisms underlying this synaptic processing, however, are largely unexplored. Intracortical processing of the auditory signal was investigated by simultaneously recording from pairs of connected principal neurons in layer II/III in slices from A1 auditory cortex. Physiological patterns of stimulation in the presynaptic cell revealed two populations of postsynaptic events that differed in mean amplitude, failure rate, kinetics and short-term plasticity. In contrast, transmission between layer II/III pyramidal neurons in barrel cortex were uniformly of large amplitude and high success (release) probability (Pr). These unique features of auditory cortical transmission may provide two distinct mechanisms for discerning and separating transient from stationary features of the auditory signal at an early stage of cortical processing.

Sensory cortices perform sophisticated pattern recognition tasks, elaborating the thalamic input via a variety of functionally different synapses across the six layers. The role of the auditory cortex in processing and modifying auditory signals, however, is largely unknown. Thalamic inputs to primary auditory cortex (A1) show several types of responses to acoustic stimuli, which fall into two broad classes of ‘transient’ and ‘sustained’ firing patterns1–3. Sustained responses saturate and can even be inhibited at increasing sound pressure level (SPL), whereas transient responses do not saturate or decrease at increasing SPL, suggesting differential mechanisms of synaptic processing. How these afferent signals are processed by downstream elements is unknown and may rely on distinct intrinsic conductances, synaptic inputs, ensemble coding or a combination of these or other factors. Neurons within layer II/III of the auditory cortex receive direct thalamic input in addition to inputs from both layer IV spiny stellate neurons and other layer II/III neurons4–6. Output from layer II/III pyramidal neurons project to local layer V and layer II/III in neighboring and contralateral cortical areas, feeding all subsequent stages of auditory signal analysis. Anatomical studies have identified pyramidal cells in layer II/III of the auditory cortex with different types of axonal recurrent collateral patterns4,5,7 raising the possibility of differential functional interactions within auditory columns8. We wanted to test the hypothesis that such intrinsic connections had the potential for functionally differentiating between either transient or sustained firing patterns. Here we demonstrate the existence of two modes of transmission between pyramidal neurons of layer II/III of the auditory cortex using paired whole-cell recordings. These two modes of transmission differed with respect to their success probability, EPSC amplitude, 1230

kinetics and short-term plasticity. In contrast, identical experiments done between layer II/III pyramidal cells in barrel cortex reveal a uniformly homogeneous population of connections with large amplitude EPSCs and high release probability, indicating that the properties of layer II/III auditory cortex connections do not represent two populations of connections common to other sensory cortical areas. We propose that these two synaptic networks implement two spectrally different tasks: the low probability synapses support ensemble-encoded narrow-band, long-lasting input, and the high probability synapses serve as reliable event detectors and wide-band signal analyzers.

RESULTS We recorded from over 100 pairs of layer II/III pyramidal neurons, which resulted in 37 synaptic connections in 35 pairs (two were reciprocally connected). All recordings were made at room temperature (∼22°C) with the presynaptic cell held under current clamp and postsynaptic cell held under voltage clamp. Synaptic currents evoked by single action potentials in the presynaptic cell (stimulus frequency, 0.06 Hz) were blocked by the AMPA receptor antagonist DNQX (10 µM, n = 6), confirming that connections were glutamatergic in nature. Synaptic events were insensitive to exogenous application of philanthotoxin (2 µM, n = 11, data not shown), indicating that synapses contained predominantly GluR2-containing, Ca2+-impermeable AMPA receptors9–14. In 24 of 37 connections, transmission occurred with a low probability (‘weak connections’), with most presynaptic action potentials failing to produce detectable postsynaptic events (Figs. 1a and 2a). The remaining 13 connections demonstrated synaptic transmission with a higher success probability with nature neuroscience • volume 4 no 12 • december 2001


articles


a

b

c

only a small number of failures of transmission (‘strong connections’; Figs. 1b and 2a). A plot of the probability of transmission success (Pr = 1 – failures) between all connections revealed a histogram composed of two distinct clusters (Fig. 2a). One cluster, which we defined as ‘low-probability connections’ (LPCs), had Pr values less than 0.4 (mean ± s.e.m., 0.13 ± 0.02). We defined the second cluster, which had Pr values greater than a 0.5 (mean ± s.e.m., 0.68 ± 0.02), as ‘high probability connections’ (HPCs). To determine whether the Pr Fig. 2. Differing properties of HPCs and LPCs. (a) Left, histogram plots of release probability (Pr1) for synaptic responses in all connections (n = 37). The discontinuous distribution reveals two modes of transmission, one with low probability and one with high probability (24/37 and 13/37 connections, respectively). The largest population comprised those with a high initial release probability (HPC, open columns), with mean success probability (Pr) of 0.68 (right). The second class, low probability connections (LPC, black columns) had a mean probability across connections of 0.13. The Pr distribution was best fit by the sum of two Poissonian distributions (continuous line). Right three panels, mean values for the release probability, EPSC amplitudes measured in the absence of event failures (Anf) and total EPSC amplitude (A, measurement including failures) of the two groups. (b) 10–90% rise time (left) and decay time constant (τ, right) versus Pr1. Individual experiments are shown by small symbols. The mean values across all data are indicated by large symbols. LPC, triangles; HPC, circles. (c) Left, paired-pulse ratio A2/A1 versus success probability of the first EPSC. All but two HPCs demonstrated paired-pulse depression, whereas LPCs could show either depression or facilitation. Right, CV2 analysis suggests that presynaptic mechanisms (indicated by hatched areas) largely determine the response to paired-pulse stimuli. In this figure, error bars represent standard deviation to allow better resolution of the data scatter. nature neuroscience • volume 4 no 12 • december 2001

Fig. 1. Two layer II/III cortico–cortical excitatory connection phenotypes. (a) Representative example of a ‘weak connection’ synapse. Top, single traces showing EPSCs evoked by paired-pulse stimulation (stimuli separated by 50 ms). Average (n = 50) EPSCs and presynaptic spikes are in red. Bottom, amplitude histograms from 50 trials for the first and second synaptic responses. Synaptic failures and events are reported in black and red, respectively. (b) Representative example of a ‘strong connection’ synapse. Setup for figure is the same as for (a). Note scale differences between (a) and (b). Bottom, EPSC amplitude histograms from 70 trials. (c) Camera lucida drawing of the corresponding neurons used in panel (b). Cell bodies were located within layer III, whereas dendrites (black) projected deep into layers I–IV. The axon (indicated in red) made many local collaterals before projecting to layer V.

dataset (Fig. 2a) represented a single population with a broad Pr distribution, data were fit with homogeneous, single Gaussian and single Poissonian distributions. None of these fits gave acceptable χ2 values. In contrast, when the dataset were fit by the sum of two Gaussians or the sum of two first-order Poissonian dis-

b

c

1231



articles

Fig. 3. Short-term plasticity at LFCs a and HFC connections. High- and low-P connections demonstrate significantly different modes of short-term plasticity in response to trains of action potentials. Eight representative individual records in response to 20-Hz trains of presynaptic action potentials (lower traces) in LPCs (a) and HPCs (b). The mean response is shown below in bold for both types of connections. Amplitude histograms for each of the four synaptic responses in the spike train (n = 100 events for each histogram for both the LPC and the HPC) are shown below the individual traces. LPCs had smaller but stationary amplitudes during the course of the stimulus train. In contrast, HPCs possessed a c larger amplitude response whose mean amplitude shifted to lower values as the train progressed. The number of failures increased during the stimulus train. (c–e) Average EPSC amplitude, probability and non-failure amplitude for all connections within the two groups (n = 13 for HPCs and n = 24 for LPCs). Whereas the HPCs showed a reduction in mean amplitude, probability and nonfailure amplitude during the spike train, LPC parameters were constant throughout the stimulus train. Solid lines are single exponential fits of the experimental data. The numeric values for τ are 31 ± 4 ms, 56 ± 28 ms, and 80 ± 11 ms for (c), (d) and (e), respectively.

tributions, the χ2 values (5.7 and 0.8, respectively) were well within the χ20.05 range (<11.1 and <14.1, respectively). This suggests that the data do not represent an evenly distributed single population of synaptic connections, but rather, indicate that HPC and LPCs reflect synapses with two distinct modes of transmission between layer II/III pyramidal neurons. Throughout the remainder of this report, all data were obtained from 24 cell pairs for LPCs and 13 cell pairs for HPCs, unless specified otherwise. Biophysical and paired-pulse properties We next wanted to determine whether LPCs and HPCs differed in properties other than their release probability. We observed no obvious differences in the resting membrane potentials (LPCs, V r = –59.6 ± 1.5 mV; HPCs, –61.3 ± 1.6 mV), input resistance of the postsynaptic cells (LPCs, Rinput = 228 ± 38 MΩ, n = 11; HPCs, 243 ± 53 MΩ, n = 6), or latency of EPSC onset (LPCs, mean latency, 2.9 ± 0.3 ms, n = 19; HPCs, 2.3 ± 0.2, n = 11). Analysis of EPSC kinetics revealed that the 10–90% rise time differed between the two populations (LPCs, 1.5 ± 0.2 ms, n = 19; HPCs, 2.0 ± 0.3 ms, n = 13, p < 0.01) but not the decay time constant (LPCs, 8.4 ± 0.7 ms, n = 19; HPCs, 10.3 ± 1.1 ms, n = 13, p = 0.16; Fig. 2b). We also observed significant differences between the mean EPSC amplitudes including failures (LPCs, A = 1.1 ± 0.5 pA; HPCs, A = 6.9 ± 1.9 pA) and excluding failures (LPCs, EPSC nf = 4.7 ± 0.9 pA, n = 22; HPCs, 9.3 ± 2.6 pA, n = 13, p = 0.046; Fig. 2a). We next determined whether the paired-pulse ratio (PPR) of synaptic transmission in response to two closely time action potentials (50 ms) in the presynaptic cell were similar for LPCs and HPCs. At LPCs, the amplitude of the second EPSC demon1232

b

d

e

strated either facilitation or depression in response to pairedpulse stimuli. The paired-pulse ratio (PPR) from all connections ranged from 0.18 to 5.3 (Fig. 2c). However, across all recordings, the mean amplitude of the second EPSC was not significantly different from the first EPSC amplitude (EPSC1 = 1.1 ± 0.5 pA; EPSC2 = 1.1 ± 0.2 pA). In contrast, all but two of the HPCs displayed strong depression (EPSC1 = 6.9 ± 1.9 pA; EPSC2 = 3.5 ± 0.8 pA; Fig. 2c), with a mean PPR in the range of 0.23–1.13. The ratio Pr2/Pr1 (Pr1,2, success probability of the first and second EPSC) also differed between the two connection types: Pr2/Pr1 was 1.28 ± 0.17 for LPCs, and 0.78 ± 0.06 for HPCs. Evidence that presynaptic mechanisms determine the observed response to paired-pulse stimuli in both LPCs and HPCs was suggested by CV2 (amplitude mean/variance) analysis16 (Fig. 2c). We next determined whether any of the measured parameters, Pr, EPSC amplitude, EPSCnf or the paired pulse ratio, were correlated with the developmental age range used throughout our experiments (postnatal day 18–28). None of these parameters were correlated with the postnatal age of animals (Pr versus age, r = 0.16; mean amplitude, r = 0.31; EPSCnf, r = 0.3; paired-pulse ratio, r = –0.09; data not shown). Similarly, no correlation existed between the intersomatic distances between the pairs of layer II/III pyramidal cells chosen for study (intersomatic distances measured from post hoc anatomical analysis) and the Pr (r = 0.04). Short-term plasticity in response to spike trains Units in A1 recorded in vivo have an upper firing frequency limit between 40 and 60 Hz17, and cortical integration times are of the order of 20–150 ms18. Moreover, in vivo cortical units respond to repeated auditory stimuli up to frequencies of about 20 Hz19. Therefore, trains of spikes 150 ms long at 20 Hz (repeated 30–150 times at 15-second intervals, to allow complete recovery from depression or facilitation) were selected to simulate physiological stimuli. The postsynaptic response to the presynaptic spike train markedly difnature neuroscience • volume 4 no 12 • december 2001


articles

Fig. 4. Time course of recovery from short-term depression at HPCs. To determine the time course of recovery from depression in HPCs, experiments similar to those shown in Fig. 4 were repeated, but the initial 20-Hz train was followed by stimuli 1 and 2 s after the termination of the train. (a) Example of HPC activity at 3 time points during the train and 2 s following termination of the train. Each panel represents 10 traces evoked by single action potentials (bottom). Left and middle traces represent the first and fourth events in the train and illustrate the significant depression of the event amplitude. Right, recovery of the event amplitude at a 2-s time point after termination of the train. (b, c) The mean time course of the recovery for the EPSC amplitude and probability, respectively (n = 6 connections).


a

b

c

fered between LPCs and HPCs. On average, LPCs displayed no change in mean amplitude, Pr or EPSCnf of any EPSCs occurring during the train (Fig. 3a, c–e). In contrast, HPCs displayed depression of all three parameters when comparing the fourth to the first event in the train (Fig. 3b–e). In all cases, the observed amplitude (or probability) depression was best described by a single exponential function (Fig. 3c–e). Despite the observation of significant short-term depression of transmission during trains of stimuli at HFCs, the mean EPSC amplitude remained significantly greater than that of the corresponding LFCs (Fig. 3c). The mean value of the fourth EPSC was 1.1 ± 0.2 pA for LPCs, and 2.7 ± 0.7 pA for HPCs. Similarly, despite the decrease in Pr during successive events in the train, Pr4 (success probability of the fourth response) in HPCs remained significantly higher than that of the LPCs (HPCs, Pr4 = 0.5 ± 0.1; LPCs, 0.1 ± 0.2; Fig. 3d). The stable response of LPCs is qualitatively different from the depressing response observed in HPCs, and suggests that HPCs are not simply the numerical summation of multiple LPCs and that each represents a distinct mode of transmission between auditory cortex pyramidal cells. Fig. 5. Differential dependence of short-term plasticity on [Ca2+]o in HPCs. (a–d) Representative experiment demonstrating the effect of reducing [Ca2+]o on short-term plastic properties. (e, f) Mean data from five experiments at high-probability connections. (a–d) Lowering [Ca2+]o from 1.5 to 0.5 mM decreased both the mean amplitude and the success probability of EPSCs occurring early in the train. In contrast, the amplitudes of synaptic responses occurring later in the train were not significantly different from those recorded in 1.5 mM [Ca2+]o (mean change in EPSC4 amplitude, 5 ± 2%). (c, e) This redistribution of EPSC amplitude acts to minimize the short-term depression observed in control [Ca2+]o. The influence of [Ca2+]o on short-term plasticity is more clearly observed when the two data sets are normalized to their respective first EPSC amplitudes (e, inset), which clearly illustrates that the depressing response of HPCs in control [Ca2+]o is converted to a ‘stationary’ response in low [Ca2+]o. (d) A comparison of the Pr in the two [Ca2+]o conditions suggests that a reduction of Pr1 and Pr2 contributes to the reduction of EPSC1 and EPSC2 amplitude observed in panel (c). (f) Examination of the mean Pr data reveals that in general, the largest reduction in Pr was associated with Pr1. nature neuroscience • volume 4 no 12 • december 2001

Time course of recovery from short-term depression To determine the time course of recovery from short-term depression in HPCs, experiments similar to those described above were repeated, but the initial 20 Hz train was then followed by stimuli 1–2 seconds after the train. In 6 HPCs (Fig. 4), the depression of the EPSC mean amplitude occurring during the 20-Hz conditioning train did not fully recover until about 2 seconds after termination of the 20-Hz spike train (EPSC1 = 6.9 ± 0.8 pA, EPSC 4 = 2.9 ± 1.3 pA, EPSC 1second = 2.5 ± 0.5 pA, EPSC2seconds = 5.5 ± 1.3 pA; Fig. 4b and c). The time course for amplitude recovery did not parallel that of the recovery of Pr (Fig. 4c) (mean PrEPSC1 = 0.70 ± 0.05, PrEPSC4 = 0.41 ± 0.04, PrEPSC1second = 0.42 ± 0.07, PrEPSC2seconds = 0.53 ± 0.10). These data suggest that two distinct mechanisms with distinct kinetics may underlie recovery from short-term depression. [Ca2+]o-dependence of short-term plasticity Several features of synaptic transmission, such as release probability and the degree of depression or facilitation in response to

a

b

c

d

e

f

1233


articles


Fig. 6. Differential dependence of short term plasticity on [Ca2+]o in LPCs. Greater than twofold elevation in [Ca2+]o fails to convert the stationary response of LPC short-term plasticity into a depressing one. Figure setup is identical to Fig. 5. At LPC connections, an elevation of [Ca2+]o from 1.5 to 3.5 mM results in an increase in the mean amplitude and Pr of EPSCs occurring early in the train (c–f). However, this increase in EPSC amplitude or Pr was not accompanied by any alteration of the short-term plastic properties of the synapse (e, inset).

a

trains of stimuli, are tightly regulated by the extracellular calcium ion concentration ([Ca2+]o)20,21. For example, at connections between hippocampal pyramidal neurons, the degree of paired-pulse depression increases with elevation of [Ca2+]o22, indicating that shortterm plasticity is strongly influenced by the initial presynaptic release probability. To determine whether c the differential response to brief trains of stimuli of HPCs and LPCs resulted from differences in the initial Pr, we next examined the role of [Ca2+]o in shaping the synaptic response to short trains of presynaptic activity. Specifically, we wanted to determine whether the pattern of short-term plasticity at HPCs could be converted to activity characteristic of LPCs by simply decreasing [Ca2+]o, and vice versa. e In 5 HPC recordings, [Ca2+]o was decreased from 1.5 to 0.5 mM (whereas [Mg2+]o was raised from 1.5 to 2.5 mM). Such a reduction in Ca2+ and elevation of Mg2+ allowed us to manipulate the amplitude and P r of EPSC1 to approximate LPCs. The reduction in [Ca2+]o reduced the mean EPSC 1 amplitude by 40 ± 8% (Fig. 5), but had little effect on events occurring later in the train (mean change in EPSC4 amplitude, 5 ± 2%; Fig. 5c and e). This redistribution of EPSC amplitude reduced the short-term depression observed in control [Ca2+]o: in low-[Ca2+]o, EPSC1 and EPSC4 amplitudes were not significantly different (EPSC1 = 7.3 ± 0.8 pA versus EPSC4 = 5.4 ± 1.2 pA, p = 0.28). The short-term plastic properties of HPCs in low [Ca2+]o were similar to those observed at LPCs in 1.5 mM [Ca2+]o (compare with Fig. 3c). This was more clearly illustrated when the two data sets were normalized to their first EPSC amplitudes (Fig. 5e, inset), which showed that a depressing response at HPCs is converted to a ‘stationary’ response in lowered [Ca2+]o. Lowering [Ca2+]o significantly reduces the Pr of EPSC1 and EPSC2 (Fig. 5a and b). This reduction of Pr1 and Pr2 likely contributed to the smaller EPSC1 and EPSC2 amplitude (Fig. 5c). Across all experiments, the largest reduction in Pr of all events in the train was associated with P r1 (mean P r1low[Ca2+]o/mean Pr1ctrl[Ca2+]o = 0.46 ± 0.15, Fig. 5f). This reduction in Pr of events early in the train endows HPC synapses with transmission properties reminiscent of LPCs. In contrast, the pattern of short-term plasticity at LPCs was not altered by elevating [Ca2+]o. In 5 experiments, [Ca2+]o was elevated from 1.5 to 3.5 mM (and [Mg2+]o was decreased to 0.5 mM). Under these conditions, we observed an increase in the mean amplitude and Pr of EPSCs occurring early in the train (Fig. 6; mean increase in EPSC1 amplitude, 160 ± 98% of control). However, this increase in EPSC amplitude did not alter the pattern of short-term plasticity at these synapses (that is, we were unable to convert a ‘stationary’ response to a ‘depressing’ one; Fig. 6c and e and inset, compare Fig. 5c and e). Indeed, on no occasion did we observe a LPC that demonstrated HPC1234

b

d

f

like short-term plasticity in elevated [Ca2+]o conditions. These data suggest that although transmission in LPCs is sensitive to changes in [Ca2+]o, LPCs could not be ‘converted’ to HPCs by simply changing their Pr. Equivalent transmission is not observed in barrel cortex To determine whether LPCs and HPCs were a general property of synapses between layer II/III pyramidal neurons in other sensory cortical areas, similar recordings were made from pyramidal neurons in barrel cortex slices23 across an identical age range. Synaptic transmission between connected pairs of layer II/III pyramidal neurons from barrel cortex possessed properties fundamentally different from equivalent neurons in auditory cortex (Fig. 7). From a total of 18 connected pairs, the mean EPSC amplitude in response to single presynaptic stimulation was 19.6 ± 1.8 pA (n = 18), almost 3 times greater than the mean amplitude of connections between HPC in auditory cortex and about 20 times greater than LPC amplitudes (compare Figs. 1 and 2). The probability of synaptic transmission was also significantly higher at barrel cortex connections (mean Pr = 0.93 ± 0.01, n = 18; Fig. 7e) with few failures of synaptic transmission occurring (Fig. 7a, b, d; compare with HPC mean P r = 0.68). Consequently, the mean EPSC amplitude, excluding failures (mean, 20.9 ± 1.9 pA) was significantly greater than observed in either auditory cortex HPC (9.3 pA) or LPC (4.7 pA). Responses to short trains of presynaptic stimuli were invariably depressing (Fig. 7) with the mean EPSC amplitude, mean EPSC amplitude excluding failures and Pr all showing signifinature neuroscience • volume 4 no 12 • december 2001


articles


a

b

d

c

e

Fig. 7. Equivalent connections were not observed in barrel cortex. Experiments identical to those illustrated in Fig. 3 were made between pairs of connected neurons recorded from layer II/III pyramidal neurons in barrel cortex. (a) Representative single experiment and panels (b–d) illustrate the mean pooled data from 18 connected pairs. (a–c) In all cases, connections between layer II/III pyramidal neurons possessed large EPSC amplitudes far in excess of those observed in the layer II/III auditory connection. (a) Seven sample traces of synaptic activity in response to trains of 20-Hz presynaptic action potentials (bottom). An average of 30 traces (third from bottom) illustrates that this connection possessed short-term depression in response to the 20-Hz train. In the presence of the AMPA receptor antagonist, DNQX, all transmission was blocked, indicating the glutamatergic nature of this connection. Right, amplitude histograms constructed from cell shown on left. Failures are indicated by open columns and successes by solid columns. (b–e) Mean data from 18 connections. In all cases, the EPSC amplitude (b) and EPSC amplitude excluding failures (c) showed strong depression in response to the train of 20-Hz stimuli. (d, e) The mean initial release probability of connections in barrel cortex was extremely high, with little evidence for synaptic failures (compare with Figs. 1–3). The mean Pr decreased during the train of presynaptic stimuli. (e) Distribution of Pr1 from all 18 connections show that all connections possess a uniformly high Pr1. Solid line is fit by single Gaussian function.

the incidence of synchronous events being similar in LPCs (58%) and HPCs (64%). In contrast, only 18% of non-connected pairs received synchronous spontaneous synaptic activity (Fig. 8d). These data suggest the existence of synaptically connected ensembles whose components may be activated in concert24. cant depression during the 20-Hz train of stimuli (Fig. 7b–d). Despite demonstrating significant reduction in the mean Pr during the course of the stimulus train (Pr1 = 0.93 ± 0.01 versus Pr4 = 0.70 ± 0.03, n = 18), Pr4 was still significantly greater in barrel cortex that Pr4 in HPC auditory cortex (∼0.5). Similarly, the mean value of the EPSC4 was 10.9 ± 1.3 pA (n = 18) in barrel cortex compared to 1.1 pA and 2.7 pA for LPC and HPC in auditory cortex. Taken together, these data demonstrate that connections between layer II/III pyramidal neurons in barrel cortex possess properties distinct from those observed in auditory cortex. Moreover, these data suggest that HPC and LPCs do not represent a homogeneous property of synaptic connections common to other sensory cortical areas. A high level of connectivity within layer II/III In recordings from auditory cortex, spontaneous excitatory synaptic activity often occurred synchronously on both the presynaptic and postsynaptic neuron, suggesting that common sources may innervate layer II/III neurons. We next determined whether the incidence of such synchronous synaptic activity preferentially occurred on pairs of cells that were connected by LPC or HPCs compared to pairs of cells that were not synaptically connected. Correlated spontaneous EPSCs were detected primarily in recordings from connected cells (Fig. 8). The synchronous nature of spontaneous events detected in cell pairs was confirmed by the presence of a sharp peak in cross-correlograms (Fig. 8c). The percentage of cells displaying synchronous spontaneous currents was higher in connected cells compared to non-connected cells, with nature neuroscience • volume 4 no 12 • december 2001

DISCUSSION Here we provide functional evidence, based on differences in EPSC amplitude, success probability, kinetics and response to brief trains of stimuli, for two different modes of transmission between excitatory connections of the primary auditory cortex7. The most frequently observed transmission mode, the LPC, was associated with small mean EPSC amplitude, a low event success probability, and on average, a non-decremental response to 20-Hz spike trains. Transmission via HPCs was characterized by a high Pr, a larger mean EPSC amplitude, paired-pulse depression, and marked short-term depression in response to 20-Hz spike trains. The differential response of LPCs and HPCs to short trains of presynaptic stimuli represented the most significant difference between the two modes of transmission. At LPCs, the mean EPSC amplitude and Pr remained constant at each successive event in the train. In contrast, both mean event amplitude and Pr decreased during the 20-Hz train at HPCs. This suggests that whereas the release machinery at LPCs can faithfully support brief trains of presynaptic stimuli, the decrease in both EPSC amplitude and Pr at HPCs favors their involvement in transient episodes. It is uncertain at this time whether LPCs and HPCs represent two distinct classes of connection or represent the extreme ends of a continuum ranging from low- to high-probability connections. Because a uniform distribution for Pr between cortical pyramidal neurons in layer V was reported previously15, we considered the possibility that the clustering of LPCs and HPCs resulted from the finite sample size (n = 37). We considered this unlikely; were the data from a single population of connections, possessing a uniform 1235



articles

Fig. 8. Synchronous spontaneous synaptic input onto connected layer II/III cells. (a) An example of two connected cells (LPC-type). Lower traces indicate presynaptic action potentials; upper traces represent four representative trains of postsynaptic EPSCs. (b) The same two cells display numerous synchronous spontaneous EPSPs (upward deflections)/EPSCs (downward deflections) that originate from unidentified sources (indicated by arrows). (c) A zero lag in the cross correlation analysis confirms the synchronous nature of the co-occurring synaptic events. (d) Summary histogram showing that the fraction of cells showing simultaneous spontaneous EPSC is high in both HPCs and LPC connected pairs but lower in recordings of two cells showing no connectivity.

distribution of Pr, the probability of observing such a gap in the dataset within a single population would be extremely low (p < 0.0001). In addition, Pr frequency histograms (Fig. 2a) were well fit by the sum of two first-order Poissonian distributions but poorly fit by either a single Poissonian or Gaussian distribution. Thus, we consider it likely that LPCs and HPCs represent specific populations of synapse types whose transmission properties are tuned for specific roles in auditory cortex. This is underscored by the experiments from equivalent connections in barrel cortex. Transmission between layer II/III pyramidal neurons in barrel cortex occurred via synapses that had high release probability (mean Pr = 0.93) and large EPSC amplitudes, and that invariably demonstrated short-term depression of both EPSC amplitude and Pr in response to brief trains of stimuli. Physiological function of HPCs and LPCs In vivo recordings from thalamus show that a limited number of firing patterns exist in response to sustained tones or repetitive clicks. The response to tone stimulation, measured as mean firing rate, is either sustained or transient in nature2. Similarly, the two most frequently encountered patterns, classified according to the response to series of clicks, are the so-called ‘lockers’ and ‘special responders’1. ‘Lockers’ respond to clicks presented at a frequency of up to about 50 Hz, faithfully following the stimulus, whereas ‘special responders’ respond only to the onset of the stimulus train. How these two types of signal are integrated within auditory cortex is largely unknown. We speculate that two synaptic ‘channels’ composed of connections with temporal characteristics observed in the present experiments could faithfully convey to cortical neurons (which fire up to approximately 20 Hz19) information encoded in transient or sustained patterns of thalamic input. Because of the distinct properties of short-term plasticity, we propose that LPCs and HPCs may subserve fundamentally different functions in the auditory network. Input to a set of LPCs could act to support ‘locker’ responses, ensuring fidelity of the cortical output. A cortical unit receiving a sustained train of action potentials from a sufficient number of LPCs could indeed produce a sustained noise-insensitive response by integration of ensemble-coded synaptic excitatory input. On the contrary, supra-threshold input through a set of HPCs would result in a transient, depressing response, largely independent of the duration of the input signal. Thus, the brief activation of a few HPCs may generate ‘special responder’ patterns in cortical units. Consequently, a unit that is postsynaptic to HPCs may be particularly suited for detecting variability in the timing of signal location and frequency. An alternative or complementary role of depressing synapses is the detection of subtle differences of frequency-modulated signals 25. These interpretations do not rule out the involvement of additional mechanisms, such as local inhibition or non-local cortical interactions, commonly invoked to explain shaping of excita1236

a

b

c

d

tory responses. Future experiments are aimed at determining the role of inhibitory neurons within the circuitry. In conclusion, pyramidal cells within layer II/III of the auditory cortex process incoming presynaptic signals using two synaptic networks with unique computational features. HPCs and LPCs networks in layer II/III provide a compelling synaptic substrate for performing spectrally heterogeneous demands of auditory cortical analysis, such as event detection and sound fingerprinting26, possibly related to the ‘what and where’ processing in the auditory pathway27.

METHODS Auditory cortex slices. C57/BL6 male mice (18–28 days old) were anesthetized according to the NIH Animal Care and User Committee guidelines, and their brains were placed for 1–2 min in ice-cold saline solution composed of 130 mM NaCl, 3.5 mM KCl, 24 mM NaHCO3, 1.25 mM NaH2PO4, 0.5 mM CaCl2, 3.0 mM MgCl2 and 10 mM glucose. After removal of the cerebellum, 250-µm-thick coronal slices from the first sixth of the caudal part of the brain, where the primary auditory area A1 lies, were cut at 0–4°C. Slices were then incubated at 32°C in the same solution until used for recordings. Slices were subsequently moved to a recording chamber and superfused with a solution as above except that [Ca2+] and [Mg2+] were both 1.5 mM and maintained at room temperature. Barrel cortex slices. Coronal slices (400 µm) containing the posteromedial barrel subfield were prepared from C57/BL6 mice (P18–P28) using previously described methods23. Mice were anesthetized with forane and decapitated, and the brain was rapidly removed in ice-cold saline solution as described above. The posterior part of the left hemisphere was vertically trimmed away at 50 degrees relative to the midsagittal plane. The trimmed surface was then glued downward to the vibratome (Leica, Deerfield, Illinois) stage and slices were cut from the rostral pole of the right hemisphere parallel to the trimmed surface. The initial 4 slices (about 1600 µm) were discarded and the next 5 slices (about 2000 µm) containing the whisker barrels were saved for recording. The slices were incubated at 32°C in the recording solution for 30 min and then maintained at room temperature until use. The posteromedial barrel subfield was identified by the presence of three to four large barrel-like structures in layer IV, visible under transillumination. Individual pyramidal neurons located within layer II/III were selected for recording based on somata size and position. Recording. For recording pairs of auditory cortical pyramidal neurons in layer II/III, cells were selected according to their position, dorsal to the ectosylvian region, and shape of their cell bodies. Whole-cell patch-clamp recordings were performed using Axopatch-1D amplifiers. Signals were filtered at 2 kHz and digitized at 10 kHz on a PC using a Digidata 1200 interface driven by pClamp8 (Axon Instruments, Foster City, California). The intracellular electrode solution was composed of 90 mM KGluconate, nature neuroscience • volume 4 no 12 • december 2001



articles

12 mM KCl, 10 mM HEPES, 2 mM EGTA, 2.0 mM ATP.Na2, 0.3 mM GTP.Na, 1.0 mM MgCl2 and 0.5% biocytin. This solution was selected to provide a reversal potential of –60 mV for Cl– to minimize the contribution of GABAergic IPSCs. After obtaining cell pairs, suprathreshold electrotonic current pulses (200–300 ms) were delivered in current clamp to allow determination of the neuronal firing pattern to provide tentative identification that cells were pyramidal in nature. In all cases, the presence of an ‘accommodating’ action potential firing pattern in response to depolarizing current injection (postsynaptic cell now held under currentclamp conditions) and post hoc cell identification following biocytin injection (example in Fig. 1c) confirmed that all weakly and strongly connected neuron pairs were between pyramidal neurons. Excitatory synaptic currents were recorded by holding the presynaptic cell in current clamp and postsynaptic one in voltage clamp at a holding potential of –60 mV. Synaptic currents were evoked at 0.06 Hz by delivering current pulses (1–3 ms, 0.3–2.0 nA) to the presynaptic cell held under current clamp. In some experiments, trains of 4 action potentials were generated at 20 Hz, repeated at 15-s intervals (40–140 times; on average, trains were repeated 80 times). In experiments to determine the recovery from short-term depression, additional EPSCs were elicited at 1–2-s intervals commencing after the spike train. Biocytin staining. Following all recordings, slices were immediately transferred to a 24-well plate and fixed in a solution containing 80 mM Na2HPO4, 80 mM NaH2PO4 and 3.5% paraformaldehyde. Biocytin staining was then processed using diaminobenzidine as chromogen, using a standard ABC kit (Vector Labs, Burlingame, California). A light cresyl violet Nissl counterstain was used to identify the cortical layers. Data analysis. The input and series resistance of the postsynaptic cell was constantly monitored by delivering a –5-mV voltage command. Cells whose input or series resistance changed by more than 30% of the initial value were discarded. Lists of events and failures were created for each recorded cell and for each peak amplitude in the train using a semiautomatic algorithm setting a threshold of two standard deviations of the noise amplitude. To measure peak amplitudes, the mean of the amplitude of the failures, corresponding to the peak of the noise, was first subtracted from the event amplitude list. Latencies were determined from the peak of the presynaptic spike to the zero crossing of the straight line that best fitted the rising phase of a sample of 10–20 EPSCs. The same sample was used to calculate rise times (10–90%). Decay time constants were determined using single exponential curve fitting routines. Mean and s.e.m. are reported throughout unless otherwise indicated. Data were considered significant if p < 0.05. CV2 analysis was used to predict the locus of short term synaptic plasticity16. A change is represented as a point in the potentiation–CV2 ratio (π–ρ) graph. CV2 ≡ µ/σ2, where µ ≡ mean amplitude and σ2 ≡ amplitude variance π ≡ µafter/µbefore and ρ ≡ CV2 ratio ≡ CV2before/CV2after A data point in the regions between the ρ = π and the π = 1 straight lines (indicated by hatched area in Fig. 2d) corresponds to a presynaptic locus.

ACKNOWLEDGEMENTS We thank D. Feldman for his assistance in preparing Barrel cortex slices and C. Trouth for cell reconstruction and camera lucida drawings of neurons. This work was supported by an Intramural Research award to C.McB.


RECEIVED 24 SEPTEMBER; ACCEPTED 16 OCTOBER 2001 1. Rouiller, E., de Ribaupierre, Y., Toros-Morel, A. & de Ribaupierre, F. Neural coding of repetitive clicks in the medial geniculate body of the cat. Hear. Res. 5, 81–100 (1981). 2. Phillips, D. P. & Kelly, J. B. Coding of tone-pulse amplitude by single neurons in auditory cortex of albino rat. Hear. Res. 37, 269–280 (1989). 3. Clarey, J. C., Barone, P. & Imig, T. J. in The Mammalian Auditory Pathway: Neurophysiology (eds. Popper, A. N. & Fay, R. R.) 232–334 (Springer, New York, 1992). 4. Winer, J. A. The pyramidal neurons in layer III of cat primary auditory cortex (AI). J. Comp. Neurol. 229, 476–496 (1984). 5. Winer, J. A. Structure of layer II in cat primary auditory cortex (AI). J. Comp. Neurol. 238, 10–37 (1985). 6. Metherate, R. & Aramakis, V. B. Intrinsic electrophysiology of neurons in thalamorecipient layers of developing rat auditory cortex. Dev. Brain Res. 115, 131–44 (1999). 7. Clarke, S., de Ribaupierre, F., Rouiller, E. M. & de Ribaupierre, Y. Several neuronal and axonal types form long intrinsic connections in the cat primary auditory cortical field (AI). Anat. Embryol. (Berl.) 188, 117–138 (1993). 8. Shen, J. X., Xu, Z. M. & Yao, Y. D. Evidence for columnar organization in the auditory cortex of the mouse. Hear. Res. 137, 174–177 (1999). 9. Blaschke, M. et al. A single amino-acid determines the subunit-specific spider toxin block of alpha-amino-3-hydroxy-5-methylisoxazole-4-propionate kainate receptor channels. Proc. Natl. Acad. Sci. USA 90, 6528–6528 (1993). 10. Brackley, P. T., Bell, D. R., Choi, D. K., Nakanishi, K. & Usherwood, P. N. Selective antagonism of native and cloned kainate and NMDA receptors by polyamine-containing toxins. J. Pharmacol. Exp. Ther. 266, 1573–1580 (1993). 11. Herlitze, S. et al. Argiotoxin detects molecular differences in AMPA receptor channels. Neuron 10, 1131–1140 (1993). 12. Washburn, M. S. & Dingledine, R. Block of alpha-amino-3-hydroxy-5methyl-4-isoxazolepropionic acid (AMPA) receptors by polyamines and polyamine toxins. J. Pharmacol. Exp. Ther. 278, 669–678 (1996). 13. Toth, K., & McBain, C. J. Afferent specific innervation of two distinct AMPA receptor subtypes on single hippocampal interneurons. Nat. Neurosci. 1, 572–578 (1998). 14. Dingledine, R., Borges, K., Bowie, D. & Traynelis, S. The glutamate receptor ion channels. Pharmacol. Rev. 51, 7–61 (1999). 15. Markram, H., Lubke, J., Frotscher, M., Roth, A. & Sakmann, B. Physiology and anatomy of synaptic connections between thick tufted pyramidal neurones in the developing rat neocortex. J. Physiol. (Lond.) 500, 409–440 (1997). 16. Faber, D. & Korn, H. Applicability of the coefficient of variation method for analyzing synaptic plasticity. Biophys. J. 60, 1288–1291 (1991). 17. DeCharms, R. C., Blake, D. T. & Merzenich, M. M. Optimizing sound features for cortical neurons. Science 280, 1439–1443 (1998). 18. DeCharms, R. C. & Zador, A. Neural representation and the cortical code. Annu. Rev. Neurosci. 23, 613–647 (2000). 19. Kilgard, M. P. & Merzenich, M. M. Plasticity of temporal information processing in the primary auditory cortex. Nat. Neurosci. 1, 727–731 (1998). 20. Zucker, R. S. Short-term synaptic plasticity. Annu. Rev. Neurosci. 12, 13–31 (1989). 21. Zucker, R. S. Calcium- and activity-dependent synaptic plasticity. Curr. Opin. Neurobiol. 9, 305–313 (1999). 22. Debanne, D., Guerineau, N. C., Gahwiler, B. H. & Thompson, S. M. Paired pulse facilitation and depression at unitary synapses in rat hippocampus: quantal fluctuation affects subsequent release. J. Physiol. (Lond.) 491, 163–176 (1996). 23. Feldman, D. E. Timing-based LTP and LTD at vertical inputs to Layer II/III pyramidal cells in rat barrel cortex. Neuron 27, 45–56 (2000). 24. Schreiner, C. E., Read, H. L. & Sutter, M. L. Modular organization of frequency integration in primary auditory cortex. Annu. Rev. Neurosci. 23, 501–529 (2000). 25. Abbott, L. F., Varela, J. A., Sen, K. & Nelson, S. B. Synaptic depression and control of cortical gain. Science 275, 220–224 (1997). 26. Giraud, A. et al. Representation of the temporal envelope sounds in the human brain. J. Neurophysiol. 22, 1588–1598 (2000). 27. Kaas, J. & Hackett, T. ‘What’ and ‘where’ processing in the auditory cortex. Nat. Neurosci. 2, 1045–1047 (1999).

1237


articles


Inducible, pharmacogenetic approaches to the study of learning and memory Masuo Ohno, Paul W. Frankland, Adele P. Chen, Rui M. Costa and Alcino J. Silva Departments of Neurobiology, Psychiatry and Psychology, Brain Research Institute, University of California, Los Angeles, California 90095-1761, USA Correspondence should be addressed to A.J.S. ([email protected])

Published online: 19 November 2001, DOI: 10.1038/nn771 Here we introduce a strategy in which pharmacology is used to induce the effects of recessive mutations. For example, mice heterozygous for a null mutation of the K-ras gene (K-ras+/–) show normal hippocampal mitogen-activated protein kinase (MAPK) activation, long-term potentiation (LTP) and contextual conditioning. However, a dose of a mitogen-activated/extracellular-signal-regulated kinase (MEK) inhibitor, ineffective in wild-type controls, blocks MAPK activation, LTP and contextual learning in K-ras+/– mutants. These indicate that K-Ras/MEK/MAPK signaling is critical in synaptic and behavioral plasticity. A subthreshold dose of NMDA receptor antagonists triggered a contextual learning deficit in mice heterozygous for a point mutation (T286A) in the αCaMKII gene, but not in K-ras+/– mutants, demonstrating the specificity of the synergistic interaction between the MEK inhibitor and the K-ras+/– mutation. This pharmacogenetic approach combines the high temporal specificity that pharmacological manipulations offer, with the molecular specificity of genetic disruptions.

It is well established that inter-individual differences in the efficacy and toxicity of many medications are related to genetic diversity. Over the past 50 years, there has been significant progress in understanding the clinical relevance of genetic determinants of drug responses1,2. Here we used pharmacogenetic interactions to induce the phenotypes of recessive mutations. Gene targeting in embryonic stem cells has been used to determine the role of biochemical pathways, such as second-messenger signaling cascades, in a variety of biological processes3,4. For example, gene knockout studies in mice have implicated a variety of neuronal proteins including neurotransmitter receptors, protein kinases, phosphatases and transcription factors in learning and memory5,6. However, this methodology does not allow for temporal control of these mutations, thus limiting both the design and the interpretation of the experiments. Although inducible genetic systems7,8 promise to circumvent some of these problems, the usefulness of these systems remains limited in contrast to the great number of conventional gene targeting mutants available. It is estimated that there are more than 4,000 targeted mutants already derived. In the heterozygote state, most of these mutations are recessive, and therefore, could be used with the pharmacogenetic approach introduced here. We used this approach to induce the effects of two recessive mutations in mice. This approach takes advantage of synergism between pharmacological and genetic manipulations. For example, the K-ras+/– mutation studied here did not affect hippocampal MAPK activation, LTP or learning. Similarly, the dose of the MAPK kinase (MEK) inhibitor that we used in the pharmacogenetic studies also did not affect these phenomena in wild-type mice. However, the combination of these genetic and pharmacological manipulations had a profound effect on hippocampal MAPK activation, LTP and learning. The studies 1238

presented here show the importance of K-Ras/MEK/MAPK signaling in synaptic and behavioral plasticity. Similarly, we also used this pharmacogenetic method to show that N-methyl-D-aspartate (NMDA) receptor-dependent autophosphorylation at T286 of α-calcium/calmodulin protein-dependent kinase II (αCaMKII) is required for contextual learning. These results indicate that this pharmacogenetic procedure may be widely applicable.

RESULTS MEK inhibition in K-ras+/– mice: biochemistry The Ras/MAPK cascade is important in many biological functions9–11. The K-ras gene is essential for mouse embryogenesis, as K-Ras-deficient embryos die12,13. However, K-ras heterozygous null mutants (K-ras+/–) grow and develop normally12,13. MEK is downstream of Ras, and activates MAPK in hippocampal neurons9,14. Therefore, we applied a pharmacogenetic approach that combined K-ras +/– mutation and a MEK inhibitor (SL327)15,16, to explore the function of K-Ras/MEK/MAPK signaling in the adult brain. Phorbol esters (for example, phorbol diacetate, or PDA) can activate the MAPK cascade in hippocampal neurons14,17–19. Exposure to PDA (10 µM) produced similar levels of p42 MAPK activation in the hippocampus of K-ras+/– mutants and wild-type littermates (Fig. 1a and b). However, application of the MEK inhibitor SL327 (1 µM) attenuated MAPK activation produced by PDA in the hippocampus of K-ras+/– but not in wild-type littermates (F3,24 = 13.2, p < 0.05; Fig. 1a and b). Neither 1 µM SL327 nor the K-ras+/– mutation alone affected MAPK phosphorylation in response to PDA. The baseline levels of phosphorylated p42 MAPK were also not altered by exposure to 1 µM SL327 in wild-type or K-ras+/– slices (F3,24 = 1.6, p > 0.05). Additionalnature neuroscience • volume 4 no 12 • december 2001


a

phospho-MAPK

total MAPK p44 p42

200

K-ras+/–

150

100 –10

0

fEPSP (% of baseline)


b

200

150

100

50 –10 0

10 20 30 40 50 60 70 80

20

Time (min)

Wild type DMSO 1 µ M SL327 10 µ M SL327

200

150

100 –10

0

10 20 30 40 50 60 70 80



250

K-Ras+/– DMSO 1 µ M SL327 150

100 –10

0

Time (min)

K-ras

+ –

– +

+/–

+ + wt

– + K-ras

+ + +/–

400 Control 300

PDA

200

* 100

0

* wt

K-ras

+/–

wt + SL327

+/–

K-ras + SL327

(F2,28 = 9.6, p < 0.05) without affecting baseline synaptic transmission at Schaffer collateral–CA1 synapses (Fig. 2b and c). Pretetanic application of 1 µM SL327 did not affect LTP in wild-type slices, but significantly inhibited the induction of hippocampal LTP in slices from K-ras+/– mice (F3,38 = 8.5, p < 0.05; Fig. 2c and d). Neither 1 µM SL327 nor K-ras+/– mutation alone affected LTP. Just as with the MAPK phosphorylation experiments, only the combination of these two treatments was effective. Post-tetanic application of 1 µM SL327 did not affect the expression or maintenance of established LTP in Wild type K-ras+/– slices (F1,12 = 0.5, p > 0.05; Fig. 2e), indicating 10 µM SL327/WT a temporally specific role of K-Ras/MEK/MAPK signaling in the induction but not in the expression or maintenance of LTP. These results are consistent with the observation that MAPK is activated 2 minutes after LTP induction 50 80 110 140 and returns to baseline levels by 45 minutes17,20. Time (min)

250

200

– –

b

d

c

+ – wt

MEK inhibition in K-ras+/– mice: LTP MAPK signaling is important in synaptic plasticity; LTP-inducing stimuli elicit hippocampal MAPK activation, and the inhibition of MAPK activation blocks LTP induction17,20. Thus, we used the pharmacogenetic approach outlined above to test whether K-Ras is involved in LTP, a cellular model of learning and memory. K-ras +/– mutants exhibited normal hippocampal LTP (F1,19 = 0.02, p > 0.05; Fig. 2a), just as they showed normal MAPK activation. In wild-type slices, the MEK inhibitor SL327 prevented the induction of LTP in a concentration-dependent manner

Wild type

– –

PDA SL327

ly, total MAPK levels were not different among the eight groups examined (Fig. 1a). These data show that a subthreshold dose of the MEK inhibitor triggers a MAPK phosphorylation deficit in K-ras+/– mutants. This suggests that K-Ras signaling is important in regulating MAPK activation in the hippocampus. The synergistic interaction between the subthreshold dose of the MEK inhibitor SL327 and the K-ras+/– mutation is not unlike previously described epistatic interactions between different mutations.

a 250

p44 p42

p42 MAPK activation (%)

Fig. 1. A subthreshold dose of a MEK inhibitor induces less MAPK activation in K-ras+/– mutants. (a) Representative western blots indicating protein bands visualized with antibodies to dually phosphorylated p42/p44 MAPK and total p42/p44 MAPK. Hippocampal slices from K-ras+/– and wild-type mice were exposed to 10 µM PDA or normal ACSF (control) under the presence or absence of 1 µM SL327. The total MAPK levels indicate equal protein loading. (b) Average results (±s.e.m.) of phospho-p42 MAPK levels normalized with the values of the wild-type control group. The SL327-treated K-ras+/– group showed a significantly smaller increase in phospho-p42 MAPK levels induced by PDA than each the other three groups (*p < 0.05), whereas the other groups did not differ each other.

10 20 30 40 50 60 70 80

MEK inhibition in K-ras+/– mice: behavior Contextual fear conditioning is a robust and long-lasting form of learning, which is sensitive to hippocampal manipulations21–24. In this test, mice learn to associate a distinct context (conditioned stimulus; CS) with an aversive stimulus such as foot shock (unconditioned stimulus; US). When placed back in the same training context, the mice exhibit a range of conditioned fear responses, including freezing25,26. Contextual conditioning specifically activates p42 MAPK signaling in the hippocampus15, and MEK inhibitors disrupt contextual condition-

Time (min)

e fEPSP (% of baseline)


articles

K-Ras+/–

250

DMSO 1 µ M SL327

200 150 100 –10

0

10 20 30 40 50 60 70 80 Time (min)

Fig. 2. A subthreshold dose of a MEK inhibitor induces an LTP impairment in K-ras+/– mutants. Each point indicates the field EPSP slope (mean ± s.e.m.) normalized to the average baseline response before the tetanus (delivered at time 0). (a) K-ras+/– mice showed normal hippocampal LTP. (b) Application of SL327 (indicated by the bar) had no effect on baseline synaptic transmission at CA1 synapses. (c, d) Hippocampal slices from wild-type and K-ras+/– mice were exposed to pre-tetanic application of SL327. SL327 (1 µM) triggered LTP deficits in K-ras+/– slices without affecting LTP induction in wild-type slices. Traces in panel (c) (wild-type control, left; wild-type + 1 µM SL327, right) and in panel (d) (K-ras+/– control, left; K-ras+/– + 1 µM SL327, right) are the average of fEPSPs recorded during baseline and 70–80 min after tetanization. Scale bars, 10 ms, 0.5 mV. (e) Post-tetanic SL327 had no effect on hippocampal LTP in slices from K-ras+/– mutants.


1239


articles

a 60

b

70 60

30 20

50 40 30 20

10

10

0

0

* 3

c

d

70

10

30

20

Wild-type

70 Percent freezing

50 40

*

30 20 10

20 DMSO

SL327

mg/kg i.p.

50 40 30 20

0

20 DMSO

mg/kg i.p.

SL327

ing15,16,27. Therefore, we used the pharmacogenetic approach outlined above to explore the role of K-Ras/MEK/MAPK signaling in contextual learning and memory. K-ras+/– mice tested 24 hours after training showed normal contextual conditioning (F1,18 = 0.3, p > 0.05; Fig. 3a). Pre-training administration of the MEK inhibitor SL327 impaired contextual conditioning in wild-type mice (24-h test) in a dose-dependent manner (F5,63 = 4.4, p < 0.05; Fig. 3b). Administration of 20 mg/kg of SL327 did not affect contextual conditioning in wild-type littermates, but significantly impaired it in K-ras +/– mutants (F1,16 = 7.8, p < 0.05; Fig. 3c). Thus, only the combination of the MEK inhibitor SL327 (20 mg/kg) and the K-ras+/– mutation disrupted contextual conditioning; neither manipulation alone had an effect. The unconditioned response (UR) to shock was not affected by SL327 in either K-ras+/– mutants or wild-type mice (data not shown), suggesting that this pharmacogenetic manipulation did not disrupt the mouse’s ability to perceive the foot shock. Post-training (2 h) administration of 20 mg/kg of SL327 did not affect contextual conditioning in K-ras+/– mutants (F1,18 = 1.4, p > 0.05; Fig. 3d), just as post-tetanic application of this compound failed to affect established LTP. These results indicate that K-Ras/MEK/MAPK signaling at the time of training, but not some time afterwards, is critical for formation of contextual memory. A weak training protocol (30-s placement-to-shock) was unable to trigger a contextual conditioning deficit in K-ras+/– mutants. K-ras +/– mice exhibited contextual freezing (23.5 ± 4.4%) comparable to that of wild-type littermates (21.9 ± 4.4%; F1,38 = 0.07, p > 0.05), suggesting that the K-ras+/– mutation alone does not weaken contextual conditioning. Whereas contextual learning is normal in K-ras+/– mutants, more complex hippocampus-based learning such as spatial learning in water maze is affected in these mice (data not shown). Fig. 4. A subthreshold dose of an NMDA receptor antagonist does not induce a contextual conditioning impairment in K-ras+/– mutants. Each column represents the percentage of freezing (mean ± s.e.m.) tested 24 h after training. Pre-training administration of CPP affected contextual learning in K-ras+/– mutants and wild-type littermates similarly.

1240

Specificity of pharmacogenetic interactions Previous studies suggested that NMDA receptordependent activation of hippocampal MAPK is critical for contextual fear conditioning15. However, it is unknown how NMDA receptor function is coupled to MAPK activation during learning. Therefore, we next tested whether a subthreshold dose of an NMDA receptor antagonist could induce a contextual conditioning deficit in K-ras+/– mice. Pretraining administration of 10 mg/kg but not 5 mg/kg CPP disrupted contextual conditioning (24-h test) in both wild-type (F 2,34 = 8.4, p < 0.05) and K-ras +/– mice (F2,32 = 8.8, p < 0.05; Fig. 4). These two doses of CPP affected wild-type and K-ras+/– mice in the same manner, demonstrating that decreases in NMDA receptor signaling do not affect K-ras+/– mice more severely than their wild-type littermates. There is much evidence for a close structural and functional link between the NMDA receptor and αCaMKII28–32. NMDAreceptor-dependent autophosphorylation of αCaMKII at T286 is required for the activation of this kinase32–35. Our previous studies demonstrated that mice homozygous for a point mutation at T286 that blocks the autophosphorylation of αCaMKII (αCaMKIIT286A–/–) have normal NMDA function, but have severe impairments in hippocampal plasticity and hippocampus-dependent learning and memory36. Similarly, this homozygous mutation completely blocked contextual fear conditioning (data not shown). In contrast, the heterozygous αCaMKIIT286A+/– mutants showed normal levels of contextual conditioning tested 24 h after training (Fig. 5b and d). In wild-type mice, contextual fear conditioning was disrupted by pre-training administration of NMDA receptor antagonists CPP (F 3,62 = 3.7, p < 0.05) or MK-801 (F4,48 = 3.8, p < 0.05) in a dose-dependent manner (Fig. 5a). Administration of doses of CPP (5 mg/kg) and MK-801 (0.01 mg/kg), which were behaviorally ineffective in wild-type K-ras

60

10

0

mg/kg i.p.

90 80

60

100

SL327

+/–

60 Percent freezing

Percent freezing

40

DMSO

Percent freezing


Percent freezing

50

Fig. 3. A subthreshold dose of a MEK inhibitor induces a contextual conditioning impairment in K-ras+/– mutants. Each column represents the percentage of freezing (mean ± s.e.m.) tested 24 h after training. (a) K-ras+/– mice showed normal levels of contextual freezing. (b) Pretraining administration of SL327 decreased contextual freezing in wild-type mice in a dose-dependent manner. *p < 0.05 versus DMSO-treated group. (c) K-ras+/– mice were administered DMSO or 20 mg/kg of SL327 1 h before training, which had no effect on freezing in wildtype littermates. SL327 (20 mg/kg) significantly decreased contextual freezing in K-ras+/– mice (*p < 0.05). (d) SL327 (20 mg/kg) administered 2 h post-training had no effect on contextual freezing in K-ras+/– mice.

Wild-type

50

K-ras

+/–

40 30 20

*

*

10 0

5 Saline

10 CPP

10

5 Saline

mg/kg i.p.

CPP



articles

a

b

40

40

Wild-type +/–

T286A

20

10

*

* 5 Saline

c

Percent freezing

30

0

10

20

CPP

0.01 0.03

0.1

0.3

*

20

10

0

mg/kg i.p.

MK-801

d

50

5 Saline

CPP

Saline

CPP

5 Saline

CPP

Saline

CPP

mg/kg i.p.

50

40

40 Percent freezing

Percent freezing


Percent freezing

30

30

*

20 10 0

0.01 Saline

MK801

0.01 Saline

mg/kg i.p.

K-801

30 20 10 0

5

5

mg/kg i.p.

Fig. 5. Subthreshold doses of NMDA receptor antagonists induce a contextual conditioning impairment in αCaMKIIT286A heterozygotes, but not in K-ras+/– mutants. Each column represents the percentage of freezing (mean ± s.e.m.) tested 24 h after training. (a) Pre-training administration of CPP and MK-801 decreased contextual freezing in wild-type mice in a dose-dependent manner. *p < 0.05 versus saline-injected group. (b, c) αCaMKIIT286A+/– mice and wild-type littermates were administered saline, 5 mg/kg CPP or 0.01 mg/kg MK-801 30 min before training. The CPPor MK-801-injected αCaMKIIT286A+/– group showed significantly smaller freezing responses than each the other three groups (*p < 0.05), whereas the other groups did not differ from each other. (d) CPP (5 mg/kg) administered 2 h post-training had no effect on contextual freezing in αCaMKIIT286A+/– mice or wild-type littermates.

mice, disrupted contextual fear conditioning in αCaMKIIT286A+/– mutants (CPP, F3,115 = 3.9, p < 0.05, Fig. 5b; MK-801, F3,56 = 6.4, p < 0.05, Fig. 5c). Therefore, 5 mg/kg of CPP triggered a contextual conditioning deficit in αCaMKIIT286A+/– mutants, but not in K-ras+/– mice. Therefore, the effects of this dose of CPP are specific to the αCaMKIIT286A+/– mutants, reflecting the close interaction between NMDA receptors and αCaMKII. Similarly, our data also suggest that K-Ras is critical in MEK-dependent activation of MAPK. These findings attest to the specificity of the synergistic interactions at the heart of this pharmacogenetic approach. The conditioning deficits in αCaMKIIT286A+/– mutants treated with either CPP or MK-801 were not due to changes in nociception, as neither of these pharmacogenetic manipulations affected unconditioned responses to foot shock (data not shown). Post-training injection of 5 mg/kg CPP also had no effect on contextual learning in αCaMKII T286A+/– mutants (F 3,36 = 0.6, p > 0.05; Fig. 5d), indicating that contextual conditioning requires NMDA-receptor-dependent autophosphorylation of αCaMKII, specifically around the time of training.

DISCUSSION Here we have introduced a way to manipulate molecular function in the central nervous system. This approach takes advantage of synergistic interactions between pharmacological and genetic manipulations to alter the function of specific signaling pathways in a temporally controlled manner. For example, the K-ras+/– mutation alone did not affect hippocampal MAPK activation. However, a subthreshold dose of an inhibitor of MEK, a component of the Ras/MEK/MAPK pathway9, triggered a comnature neuroscience • volume 4 no 12 • december 2001

plete block of hippocampal MAPK activation in the K-ras+/– mice. The same dose of this inhibitor did not affect MAPK activation in wild-type littermate controls. MEK and MAPK have previously been shown to be involved in LTP17,20 and learning15,16,27,37. Our pharmacogenetic approach indicated that K-Ras signaling is critical for the activation of the MAPK cascade during synaptic and behavioral plasticity. Although the K-ras+/– mutants showed normal hippocampal LTP and contextual conditioning, a dose of the MEK inhibitor ineffective in wild types induced both LTP and learning deficits in these mutants. Previous studies indicated that a H-ras mutation enhanced hippocampal NMDA currents and LTP38. However, it is unclear whether this enhancement in LTP is a compensatory reaction to the homozygous loss of H-ras during development. The advantage of our pharmacogenetic approach is that it circumvents developmental confounds. Unlike H-ras homozygous null mutants, the K-ras+/– mice show normal LTP; their deficits are observed only after drug treatment. The pharmacological induction of the K-ras+/– phenotype showed temporal specificity, as pre-training but not posttraining (2 h) administration of the MEK inhibitor triggered the contextual learning deficit in this mutant. Similarly, the MEK inhibitor is only effective when applied before the induction of LTP. Indeed, p42 MAPK activation is significantly increased in the hippocampus one hour after training, but returns to baseline levels within two hours of contextual conditioning 15 . Taken together, these findings indicate that K-Ras/MEK/MAPK signaling is critical in both the induction of LTP andhippocampus-dependent learning. 1241



articles

Although we did not find any evidence linking NMDA receptor signaling and K-Ras-dependent plasticity and learning, a pharmacogenetic approach like the one described here could be used to explore the proposed connection between NMDA signaling and different Ras proteins. For example, H-Ras is thought to regulate NMDA currents38. Additionally, NMDA signals may activate specific Ras proteins through Ras-GRF (a Ras guanine nucleotide releasing factor) thought to have a role in plasticity and learning39–41. NMDA receptor activation may also amplify certain Ras signals by inhibiting SynGAP, a synaptic Ras-GTPase activating protein42,43. Our present findings are not inconsistent with a possible link between NMDA signals and other Ras isoforms (such as H-Ras and N-Ras) signaling to the MAPK cascade. Although a subthreshold dose of an NMDA receptor blocker did not trigger deficits in the K-ras+/– mutants, it induced learning and LTP deficits (data not shown) in heterozygous αCaMK-IIT286A mice. Mice homozygous for this point mutation show profound deficits in spatial36 and contextual conditioning learning that could be conceivably due to developmental deficits. Our present data discount this possibility. We found that although heterozygous αCaMKIIT286A mice did not show a contextual deficit, a dose of NMDA receptor antagonists ineffective in wildtype mice induced a contextual conditioning deficit in these mutants. Importantly, this deficit was only induced when the drug was administered before training. Therefore, these data demonstrate that the NMDA receptor-dependent autophosphorylation of αCaMKII specifically during training is critical for learning32. The subsequent phosphorylation of key cellular substrates, such as glutamate receptors44, is thought to be required for early stages of memory formation45,46. Because the pharmacogenetic approach introduced here uses drugs at concentrations that are ineffective in wild types, the nonspecific effects of these drugs should be reduced. The results presented here demonstrate that pharmacological manipulations can be used to induce the phenotype of recessive mutations in mice. Although neither the αCaMKIIT286A+/– nor K-ras+/– mice showed contextual learning deficits, partial pharmacological disruption of specific signaling components upstream or downstream from those genetically targeted molecules triggered learning deficits in the mutants. Importantly, these partial pharmacological disruptions only affected contextual conditioning in the presence of the mutations. Thus, epistatic-like interactions between pharmacological and genetic manipulations can be used to induce the effects of mutations in mice in a temporally controlled manner, and to identify novel functional relations between signaling pathways. This approach combines the high temporal specificity that pharmacological manipulations offer, with the molecular specificity of genetic disruptions. Spatial control of these effects could also be accomplished by neuroanatomically guided injections of compounds of interest. Importantly, this pharmacogenetic approach will be applicable not only to neuroscience questions, but also to other biological problems and to other genetic systems (such as Drosophila, Caenorhabditis elegans and yeast).

METHODS

Mice. We used K-ras mutants (K-ras+/–)13 that were F1 progeny derived from a cross between mice heterozygous for a null mutation in the K-ras gene (129T2/SvEmsJ background) and C57Bl/6N mice. Starting with the αCaMKIIT286A+/– chimeras (contributing to 129 background), this mutation was backcrossed into the C57Bl/6 genetic background 36. The αCaMKIIT286A+/– mice used in the experiments were heterozygotes derived after 8–9 backcrosses into C57Bl/6N. At 4–5 weeks postnatally, the mice were weaned and their genotypes were determined with polymerase chain 1242

reaction analysis of tail DNA samples. All experiments were done with mice 3–7 months old, and a similar number of males and females were used. The mice were housed in groups and kept on a 12 h light/dark cycle, and the experiments were always conducted during the light phase of the cycle. With the exception of testing times, the mice had ad lib access to food and water. All the procedures used were approved by UCLA’s Animal Research Committee. Animals were maintained in accordance with the applicable portions of the Animal Welfare Act and the Department of Health and Human Services Guide to the Care and Use of Laboratory Animals. Drugs. SL327 (provided by DuPont Pharmaceuticals, Wilmington, Delaware) and phorbol 12,13-diacetate (PDA; Sigma, St. Louis, Missouri) were dissolved in 100% DMSO. [±]-3-[2-Carboxypiperazin4-yl]propanephosphonic acid (CPP; Sigma) and (+)-MK-801 hydrogen maleate (RBI, Natick, Massachusetts) were dissolved in saline or artificial cerebrospinal fluid (ACSF). All biochemical, electrophysiological and behavioral experiments were conducted with the experimenter blind to the drug treatments as well as the genotype of mice. Electrophysiology. Transverse hippocampal slices (400 µm thick) were maintained in a submerged recording chamber perfused with ACSF equilibrated with 95% O 2 and 5% CO 2 at 30°C. The ACSF contained 120 mM NaCl, 3.5 mM KCl, 2.5 mM CaCl2, 1.3 mM MgSO4, 1.25 mM NaH2PO4, 26 mM NaHCO3 and 10 mM D-glucose. Extracellular field EPSPs were recorded with a metallic electrode from the stratum radiatum layer of the area CA1, and the Schaffer collaterals were stimulated with a bipolar electrode. The intensity of stimulation (100-µs duration) was adjusted to give field EPSP approximately 33% of maximum. LTP was induced by a tetanic stimulation (100 Hz, 1 s) delivered at the test intensity. After the responses were monitored at least for 20 min to ensure a stable baseline, drugs were applied to hippocampal slices. SL327 (maximal final DMSO concentration, 0.1%) was applied for 1 h before tetanization, and was maintained in the bath throughout the recording period. In some slices, SL327 was applied post-tetanically (20 min after tetanization). To determine whether the magnitude of LTP differed significantly between the groups, responses from the last 10-min block of recordings (70–80 min) were compared statistically. Western blotting. Hippocampal slices prepared from K-ras+/– mice and wild-type littermates were transferred to the same chamber as used for electrophysiology, and then exposed to PDA (10 µM) or normal ACSF for 10 min. The slices were preincubated with SL327 for 70 min. In the last 10 min of preincubation, slices were exposed to PDA or ACSF. After pharmacological activation with PDA, the CA1 subregions of hippocampal slices were dissected out on ice and homogenized in an ice-cold lysis buffer (containing 25 mM HEPES (pH 7.5), 150 mM NaCl, 2 mM EDTA, 10% glycerol, 1% Triton X-100, 100 µg/ml phenylmethylsulfonyl fluoride (PMSF), 1 µg/ml leupeptin, 1 mM sodium orthovanadate (Na3VO4), 25 mM β-Na glycerophosphate and 10 mM NaF). After insoluble material was removed by centrifugation (13,000 g for 5 min), protein concentration of the supernatant was determined by using a PIERCE BCA Protein Assay Kit (Pierce, Rockford, Illinois). Samples containing equivalent amount of protein (20 µg) were separated by 12% SDS-PAGE gels and transferred onto nitrocellulose membranes (BioRad, Hercules, California). Membranes were blocked in TBS with 0.1% Tween-20 and 5% dry milk overnight at 4°C. Then the blots were incubated for 2 h at room temperature with the primary antibody which selectively recognizes p42/p44 MAPK dually phosphorylated at Thr202 and Tyr204 (1:1,000; New England Biolabs, Beverly, Massachusetts). Another antibody that detects both phosphorylated and unphosphorylated forms of p42/p44 MAPK (1:1,000; New England Biolabs) was used to measure total MAPK levels. After membranes were washed, the blots were incubated with HRP-conjugated secondary antibody (1:2,000; BioRad). Protein signals were visualized by enhanced chemiluminescence (ECL Western Blotting Analysis system, Amersham, Arlington Heights, Illinois). Densitometric analysis for quantification of the signals was performed using a desktop scanner and ImageQuaNT software (Molecular Dynamics, Sunnyvale, California). For each experiment, both phosphorylated and total MAPK levels were normalized to those observed in the control group of wild-type mice. nature neuroscience • volume 4 no 12 • december 2001



articles

Contextual fear conditioning. The contextual conditioning experiments were performed as described previously47. During training, mice were placed in the conditioning chamber for 2.5 min, and then αCaMKIIT286A+/– and Kras+/– mice were exposed to a 0.75 and 0.50 mA foot shock for 2 s, respectively. In a separate experiment, we assessed contextual fear conditioning in mice receiving a single foot shock 30 s following placement in the conditioning chamber during training. After shock delivery, mice were left in the chamber for another 30 s, and then returned to their home cage. CPP and MK-801 were administered intraperitoneally (i.p.) in a volume of 10 ml/kg 30 min before or 2 h after training. Mice received an i.p. injection of SL327 in a volume of 2 ml/kg 1 h before or 2 h after training. The mice were tested for contextual conditioning 24 h after training. Conditioning was determined by scoring freezing behavior (absence of all but respiratory movement) with automated procedures described previously47 when the mice were placed back into the conditioning chamber. Data analysis. The significance of differences between the groups was determined by a one-way analysis of variance followed by post-hoc Newman–Keuls test when F ratios reached significance (p < 0.05).

ACKNOWLEDGEMENTS We thank S.A. Josselyn, N.B. Fedorov and K.P. Giese for discussions, and R. Chen and M. Lacuesta for help with genotyping. We also thank J.M. Trzaskos (DuPont Pharmaceuticals Research Laboratories) and T. Jacks (Department of Biology, MIT) for donating SL327 and K-ras+/– mutants, respectively. This work was funded by grants from the McKnight Foundation, Merck Foundation and the NIH (P01HD33098 and AG13622) to A.J.S. M.O. was partially supported by a research fellowship from the Uehara Memorial Foundation for Life Sciences.

RECEIVED 20 AUGUST; ACCEPTED 25 OCTOBER 2001 1. Steimer, W., Muller, B., Leucht, S. & Kissling, W. Pharmacogenetics: a new diagnostic tool in the management of antidepressive drug therapy. Clin. Chim. Acta 308, 33–41 (2001). 2. McLeod, H. L. & Evans, W. E. Pharmacogenomics: unlocking the human genome for better drug therapy. Annu. Rev. Pharmacol. Toxicol. 41, 101–121 (2001). 3. Brandon, E. P., Idzerda, R. L. & McKnight, G. S. Knockouts. Targeting the mouse genome: a compendium of knockouts (Part I). Curr. Biol. 5, 625–634 (1995). 4. Brandon, E. P., Idzerda, R. L. & McKnight, G. S. Targeting the mouse genome: a compendium of knockouts (Part II). Curr. Biol. 5, 758–765 (1995). 5. Silva, A. J., Smith, A. M. & Giese, K. P. Gene targeting and the biology of learning and memory. Annu. Rev. Genet. 31, 527–546 (1997). 6. Chen, C. & Tonegawa, S. Molecular genetic analysis of synaptic plasticity, activity-dependent neural development, learning, and memory in the mammalian brain. Annu. Rev. Neurosci. 20, 157–184 (1997). 7. Furth, P. A. et al. Temporal control of gene expression in transgenic mice by a tetracycline-responsive promoter. Proc. Natl. Acad. Sci. USA 91, 9302–9306 (1994). 8. Ray, P. et al. Regulated overexpression of interleukin 11 in the lung. Use to dissociate development-dependent and -independent phenotypes. J. Clin. Invest. 100, 2501–2511 (1997). 9. Derkinderen, P., Enslen, H. & Girault, J. A. The ERK/MAP-kinases cascade in the nervous system. Neuroreport 10, R24–34 (1999). 10. Fukunaga, K. & Miyamoto, E. Role of MAP kinase in neurons. Mol. Neurobiol. 16, 79–95 (1998). 11. Sweatt, J. D. The neuronal MAP kinase cascade: a biochemical signal integration system subserving synaptic plasticity and memory. J. Neurochem. 76, 1–10 (2001). 12. Koera, K. et al. K-ras is essential for the development of the mouse embryo. Oncogene 15, 1151–1159 (1997). 13. Johnson, L. et al. K-ras is an essential gene in the mouse with partial functional overlap with N-ras. (published erratum, Genes Dev. 11, 3277, 1997) Genes Dev. 11, 2468–2481 (1997). 14. Roberson, E. D. et al. The mitogen-activated protein kinase cascade couples PKA and PKC to cAMP response element binding protein phosphorylation in area CA1 of hippocampus. J. Neurosci. 19, 4337–4348 (1999). 15. Atkins, C. M., Selcher, J. C., Petraitis, J., Trzaskos, J. & Sweatt, J. The MAPK cascade is required for mammalian associative learning. Nat. Neurosci. 1, 602–609 (1998). 16. Selcher, J. C., Atkins, C. M., Trzaskos, J. M., Paylor, R. & Sweatt, J. D. A necessity for MAP kinase activation in mammalian spatial learning. Learn. Mem. 6, 478–490 (1999). 17. English, J. D. & Sweatt, J. D. A requirement for the mitogen-activated protein kinase cascade in hippocampal long term potentiation. J. Biol. Chem. 272, 19103–19106 (1997).


18. Ebinu, J. O. et al. RasGRP, a Ras guanyl nucleotide- releasing protein with calcium- and diacylglycerol-binding motifs. Science 280, 1082–1086 (1998). 19. Tognon, C. E. et al. Regulation of RasGRP via a phorbol ester-responsive C1 domain. Mol. Cell Biol. 18, 6995–7008 (1998). 20. English, J. D. & Sweatt, J. D. Activation of p42 mitogen-activated protein kinase in hippocampal long term potentiation. J. Biol. Chem. 271, 24329–24332 (1996). 21. Frankland, P. W., Cestari, V., Filipkowski, R. K., McDonald, R. J. & Silva, A. J. The dorsal hippocampus is essential for context discrimination but not for contextual conditioning. Behav. Neurosci. 112, 863–874 (1998). 22. Fanselow, M. S. Contextual fear, gestalt memories, and the hippocampus. Behav. Brain Res. 110, 73–81 (2000). 23. Rampon, C. et al. Enrichment induces structural changes and recovery from nonspatial memory deficits in CA1 NMDAR1-knockout mice. Nat. Neurosci. 3, 238–244 (2000). 24. Shimizu, E., Tang, Y. P., Rampon, C. & Tsien, J. Z. NMDA receptor-dependent synaptic reinforcement as a crucial process for memory consolidation. Science 290, 1170–1174 (2000). 25. Fanselow, M. S. Factors governing one-trial contextual conditioning. Anim. Learn. Behav. 18, 264–270 (1990). 26. Phillips, R. G. & LeDoux, J. E. Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav. Neurosci. 106, 274–285 (1992). 27. Schafe, G. E., Nadel, N. V., Sullivan, G. M., Harris, A. & LeDoux, J. E. Memory consolidation for contextual and auditory fear conditioning is dependent on protein synthesis, PKA, and MAP kinase. Learn. Mem. 6, 97–110 (1999). 28. Husi, H., Ward, M. A., Choudhary, J. S., Blackstock, W. P. & Grant, S. G. Proteomic analysis of NMDA receptor-adhesion protein signaling complexes. Nat. Neurosci. 3, 661–669 (2000). 29. Kennedy, M. B. Signal-processing machines at the postsynaptic density. Science 290, 750–754 (2000). 30. 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 Nmethyl-D-aspartate receptor. Proc. Natl. Acad. Sci. USA 96, 3239–3244 (1999). 31. Silva, A. J. & Giese, K. P. in Neurobiology of Learning and Memory (eds. Martinez, J. & Kesner, R.) 89–142 (Academic, San Diego, California, 1998). 32. Lisman, J. The CaM kinase II hypothesis for the storage of synaptic memory. Trends Neurosci. 17, 406–412 (1994). 33. Fukunaga, K., Soderling, T. R. & Miyamoto, E. Activation of Ca2+/calmodulin-dependent protein kinase II and protein kinase C by glutamate in cultured rat hippocampal neurons. J. Biol. Chem. 267, 22527–22533 (1992). 34. 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). 35. Hanson, P. I. & Schulman, H. Neuronal Ca2+/calmodulin-dependent protein kinases. Annu. Rev. Biochem. 61, 559–601 (1992). 36. Giese, K. P., Fedorov, N. B., Filipkowski, R. K. & Silva, A. J. Autophosphorylation at Thr286 of the α calcium-calmodulin kinase II in LTP and learning. Science 279, 870–873 (1998). 37. Berman, D. E. & Dudai, Y. Memory extinction, learning anew, and learning the new: dissociations in the molecular machinery of learning in cortex. Science 291, 2417–2419 (2001). 38. Manabe, T. et al. Regulation of long-term potentiation by H-Ras through NMDA receptor phosphorylation. J. Neurosci. 20, 2504–2511 (2000). 39. Farnsworth, C. L. et al. Calcium activation of Ras mediated by neuronal exchange factor Ras-GRF. Nature 376, 524–527 (1995). 40. Giese, K. et. al. Hippocampus-dependent learning and memory is impaired in mice lacking the Ras-guanine-nucleotide releasing factor 1 (RAS-GRF1). Neuropharmacology 41, 791–800 (2001). 41. Brambilla, R. et al. A role for the Ras signalling pathway in synaptic transmission and long- term memory. Nature 390, 281–286 (1997). 42. 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). 43. Chen, H. J., Rojas-Soto, M., Oguni, A. & Kennedy, M. B. A synaptic RasGTPase activating protein (p135 SynGAP) inhibited by CaM kinase II. Neuron 20, 895–904 (1998). 44. Barria, A., Muller, D., Derkach, V., Griffith, L. C. & Soderling, T. R. Regulatory phosphorylation of AMPA-type glutamate receptors by CaMK-II during long-term potentiation. Science 276, 2042–2045 (1997). 45. Tan, S. E. & Liang, K. C. Spatial learning alters hippocampal calcium/calmodulin-dependent protein kinase II activity in rats. Brain Res. 711, 234–240 (1996). 46. Cammarota, M., Bernabeu, R., Levi De Stein, M., Izquierdo, I. & Medina, J. H. Learning-specific, time-dependent increases in hippocampal Ca2+/calmodulin-dependent protein kinase II activity and AMPA GluR1 subunit immunoreactivity. Eur. J. Neurosci. 10, 2669–2676 (1998). 47. Anagnostaras, S. G., Josselyn, S. A., Frankland, P. W. & Silva, A. J. Computerassisted behavioral assessment of Pavlovian fear conditioning in mice. Learn. Mem. 7, 58–72 (2000).

1243


articles


Inferotemporal neurons represent low-dimensional configurations of parameterized shapes Hans Op de Beeck1,2, Johan Wagemans2 and Rufin Vogels1 1 Laboratorium voor Neuro- en Psychofysiologie, K.U. Leuven, Campus Gasthuisberg, Herestraat 49, B-3000 Leuven, Belgium 2 Laboratory of Experimental Psychology, K.U. Leuven, Tiensetraat 102, B-3000 Leuven, Belgium

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

Published online: 19 November 2001, DOI: 10.1038/nn767 Behavioral studies with parameterized shapes have shown that the similarities among these complex stimuli can be represented using a low number of dimensions. Using psychophysical measurements and single-cell recordings in macaque inferotemporal (IT) cortex, we found an agreement between low-dimensional parametric configurations of shapes and the representation of shape similarity at the behavioral and neuronal level. The shape configurations, computed from both the perceived and neuron-based similarities, revealed a low number of dimensions and contained the same stimulus order as the parametric configurations. However, at a metric level, the behavioral and neural representations deviated consistently from the parametric configurations. These findings suggest an ordinally faithful but metrically biased representation of shape similarity in IT.

The capacity to categorize stimuli is fundamental to all living organisms1,2. Theories of categorization agree upon the importance of the similarity between stimuli to account for many aspects of categorization performance 3–5. However, it is not straightforward to compute the degree of similarity between stimuli that can vary across a high number of dimensions, like complex shapes. Fortunately, the similarities among a set of complex stimuli can often be described in a more compact way6–8. Indeed, stimuli from many behaviorally relevant sets can be represented in a low-dimensional representation space in which the proximity between stimuli is related to their similarity. For example, by presenting the randomly ordered shapes of Fig. 1d in a particular order (Fig. 1a–c), the similarities can be easily described by a twodimensional square-like configuration. Several behavioral studies that have varied complex shape differences parametrically revealed that primates are able to represent the similarities between shapes in a low-dimensional representation space without ever seeing these stimuli in their parametric configuration9–12. Here we aim to study directly the neural basis of these lowdimensional representation spaces. Object recognition and categorization in macaques is thought to depend on the inferotemporal cortex (IT)13,14. Single IT neurons are selective for moderately complex object features15, but several studies have found little relationship between the similarities between complex objects and the responses of single IT neurons16,17. However, one needs to manipulate shape similarity parametrically to investigate how the responses of IT neurons to complex stimuli are related to the proximity of these stimuli in a low-dimensional space. Thus, we investigated whether the response pattern across a population of IT neurons can reveal a low-dimensional and faithful representation of shape similarity using parameterized shapes. 1244

As the analysis of the visual input in the visual system is highly nonlinear, the neuronal representation space could deviate from the configurations in parameter space in several ways. Previous psychophysical studies found an ordinal agreement between parametric configurations and their perceptual representation (the same number of dimensions and the same stimulus order)9–12, so we also expected this result at the neuronal level. However, configurations that fit perfectly on an ordinal level can differ on a metric level. Thus, apart from investigating ordinal relationships, we looked for consistent metric anomalies in perceptual and neuronal representation spaces with respect to the parametric similarities. In this study, we varied complex shape dimensions, that is, radial frequency components (RFCs), that have been used previously with human subjects10. These parametric variations affect a high number of perceptually salient dimensions10,18, but are not coded directly in the macaque visual system19. Nevertheless, in agreement with computational work11, we found that human and monkey subjects were able to represent the similarities between such shapes in low-dimensional representation spaces that agree well with pixel-based configurations (Fig. 2a–e). Moreover, the perceived similarities of the monkeys corresponded better with the stimulus similarities when these were computed using IT neuron responses (Fig. 2f) than with the pixel-based similarities. Further behavioral experiments in monkeys showed that the low-dimensional stimulus configuration predicts categorization performance.

RESULTS We investigated the underlying representation space of 24 shapes that were divided into 3 groups (Fig. 1). The within-group connature neuroscience • volume 4 no 12 • december 2001


articles


a

b

c

d

figurations formed two-dimensional squares in the parametric space, and similar but not identical configurations are present when using pixel-based similarities (Fig. 2a). Perceptual representation space: human data We verified the perceptual representation of the similarities between shapes in two human subjects performing similarity ratings. First, we applied hierarchical cluster analysis to determine whether these similarity ratings reflected the expected clustering of stimuli into the three groups. As expected, all stimuli from a given stimulus group were assigned to the same cluster before they were clustered with stimuli from the other two groups. A substantial proportion of variance (85%) in the similarity judgments was accounted for when reducing the 24 stimuli to these 3 clusters. Next, we analyzed the similarities among the eight shapes from each group. To determine whether the perceived similarities converge with the pixel-based similarities, we computed the congruence C between these two sets of similarities. For all groups of stimuli and for both subjects, significantly high congruences were found (Table 1). We used multidimensional scaling (MDS) to determine whether the shape similarities could be captured by the distance between the stimuli in a low-dimensional space. Two-dimensional MDS-derived configurations were computed for each subject and each group of stimuli separately (Fig. 2b and c). The more similar the two stimuli, the smaller the distance between the points representing each stimulus. Even with as few as two dimensions, these configurations accounted for most of the variance in the similarity ratings. All within-group similarities were represented in low-dimensional configurations that strongly resembled the expected configurations. First, no qualitative deviations, such as the reversal of a pair of shapes, were seen. We quantified this ordinal agreement between perceived and expected configurations by computing the Spearman rank order correlation (r s) of the polar angles of the stimulus points with respect to the center of each configuration. If one takes stimulus 1 as the reference with polar angle zero and proceeds clockwise, then one would expect stimulus 2 to have the next-smallest polar angle, followed by stimulus 3. The expected order of the stimuli was preserved perfectly in all stimulus groups (rs = 1.0, p < 0.01). Second, details of the expected configurations, such as their ‘squareness,’ were maintained. In a square, the angle between two lines connecting a point with its two neighboring nature neuroscience • volume 4 no 12 • december 2001

Fig. 1. Visual stimuli. (a–c) Three groups of eight shapes were used. Within each stimulus group, the parameter–space configuration of the stimuli is represented by the square arrangement of the stimuli. The top-left stimulus in each square has a low amplitude value for both manipulated radial frequency components. (d) Same 24 stimuli, but the 8 shapes from each stimulus group are presented in a random order in a row.

points is 90° at corner points, and 180° at points along the sides. Other configurations with the same stimulus order, such as eight equidistant points on a circle, do not possess this property. A chi-square test was used to assess the dependence of angle size (smaller or larger than 135°) and type of point (corner or side point in the parametric configuration) for each subject. Angles were greater at side points compared to corner points for each subject (χ2 = 10.74, p < 0.01 for subject 1; χ2 = 20.17, p < 0.01 for subject 2). Thus, the configurations of each subject differentiate between configurations of the same stimulus order, such as a square or a circle. To determine whether remaining metric deviations from the expected configurations were systematic rather than due to measurement noise, we determined the inter-subject consistency of deviations of the perceived similarities with respect to the pixelbased similarities. For group A and B stimuli, C was lower when the perceived similarity ratings were compared between the two subjects (Table 1, C(human 1, human 2)) than when each subject’s data were compared with the pixel-based similarities (C(pixel, human 1) and C(pixel, human 2)). This suggests that the deviations of the perceived similarities were likely due to measurement noise. However, there was a striking consistency in the way the perceived similarities of the two subjects deviated from the pixel-based similarities for group C stimuli (compare length and direction of red lines in Fig. 2b and c). Perceptual representation space: monkey data Two monkeys (Y and E) were trained in a same-different task. Following previous studies in macaques12,20, the proportion of correct ‘different’ responses for a particular stimulus pair was taken as measure of the dissimilarity of that pair. This measure can be subjected to the same analyses as performed on the human data. First, hierarchical cluster analysis showed that the most similar stimuli belong to the same stimulus group. All stimuli from a given stimulus group were assigned to the same cluster, before they were clustered with stimuli from the other two groups. However, the clustering in monkeys was less than in human subjects. The proportion of variance accounted for by reducing the 24 stimuli to 3 clusters was smaller for the monkey data (averaged across monkeys, 47% of variance explained; humans, 85%). Second, analysis of the similarity data for each stimulus group separately showed that the congruence with the pixel-based similarities was always highly significant (Table 1). Two-dimensional MDS-derived configurations (Fig. 2d and e) explained most of the variance in the similarity ratings (averaged across monkeys, 96%, 88% and 97% for group A, B and C, respectively). Overall, the within-group similarities were represented in lowdimensional configurations that captured most aspects of the expected configurations. First, the stimulus order matched the expected order (rs = 1, p < 0.01) for all stimulus groups in each monkey, except for group B in monkey Y (reversal of stimuli 6 1245



articles

Fig. 2. MDS-derived two-dimensional configurations of stimuli. (a) Configurations based on the pixel-based similarity between stimuli. The numbers refer to the stimuli as labeled in Fig. 1. The other panels show a comparison between the configurations (blue squares) of stimuli in (a) and the configurations (green diamonds) found for the perceived similarities of the first (b) and second (c) human subject, the perceived similarities of monkey Y (d) and monkey E (e), and the neuron-based similarities (f). The amount of variance (r2) of the similarity values that was explained by the two-dimensional configuration is denoted below each panel, as is the range ([minimum; maximum]) of similarity values. These ranges are expressed in different scales for each dataset and can only be compared among stimulus groups within each dataset (pixels, Euclidean distance; human, mean similarity rankings; monkey, proportion different responses; neuronal data, distance in 124-dimensional neuronal space). To aid a visual comparison of the pixel-based with the other configurations, the latter were Procrustes transformed (combination of translation, scaling, rotation and reflection)11. Red lines connect corresponding points in both configurations. As the configurations have arbitrary origin, scale and orientation, the labeling and scale on both axes in each graph are omitted.

and 8 (rs = 0.93, p < 0.01)). The latter deviation was not a consequence of a poor reliability, as it was replicated. Second, there was a clear difference in the MDS-derived configurations between corner and side points as defined in the expected configurations, with larger angles at side points compared to corner points (χ 2 = 8.17, p < 0.01 for monkey Y; χ 2 = 10.74, p < 0.01 for monkey E). We determined whether the remaining deviations of the perceived from the pixel-based similarities were consistent between the two monkeys. There was a significantly high congruence between the data of monkeys Y and E for stimulus groups A and C, although no such inter-subject consistency was present for group B (Table 1, C(monkey Y, monkey E)). Each monkey tended to represent stimulus 3 of group A more toward the center of the configuration than expected from the pixel-based configurations (Fig. 2d and e). For group C, the perceived shape configurations showed a bias towards the vertical parametric dimension (Fig. 1c), a trend also present in human subjects (Fig. 2b and c). This consistency between humans and monkeys for group C was assessed using the averaged human and monkey data, and was significant (Table 1, C(humans, monkeys); Supplementary Fig. 1, see supplementary information page of Nature Neuroscience on line for the application of MDS on these averaged data).

a

b

c

d

e

f

Neuronal representation space in IT cortex We recorded from 124 single neurons in area TE of monkeys F (n = 51) and M (n = 73) while they were learning to categorize the 24 stimuli in 2 classes. Recordings were performed in the lower bank of the superior temporal sulcus and lateral to the anterior middle temporal sulcus. Most neurons (82%) responded twice as strongly to their preferred shape than to the least preferred shape. Many of these neurons combined high within-group selectivity with lower between-group selectivity. For example, the 1246

neuron in Fig. 3a responded differently to similar shapes within each stimulus group (for example, shapes C1 and C3), whereas it responded with similar strength to shapes that look dissimilar to human and monkey observers (for example, shapes C1 and A6). Many neurons were very selective within stimulus groups, but stimuli from different groups often elicited similar responses (Fig. 3b). Indeed, the median of the distribution of the selectivity index (SE, see Methods; Fig. 4a) was 2.4, indicating that most neurons do not respond more strongly to all stimuli of one group compared to the responses to stimuli of the other groups (as can nature neuroscience • volume 4 no 12 • december 2001


articles

shape C1 and was still responsive for the neighboring shapes C2 Group A Group B Group C and C8, but no comparable Congruence between pixel-based similarities responses were elicited at any and empirical similarities point further in stimulus space. C (pixels, human 1) 0.973+ 0.987+ 0.960+ As a consequence, polar plots for C (pixels, human 2) 0.989+ 0.987+ 0.976+ each stimulus group revealed reg+ + + C (pixels, humans) 0.991 0.995 0.974 ular and unimodal tuning curves C (pixels, monkey Y) 0.979+ 0.984+ 0.979+ (Fig. 3b). The ratio index (RR) C (pixels, monkey E) 0.988+ 0.983+ 0.976+ captures the deviation of this tuning curve with respect to an uniC (pixels, monkeys) 0.989+ 0.990+ 0.979+ modal sinusoidal modulation C (pixels, neurons) 0.996+ 0.993+ 0.991+ (see Methods). The three neurons in Fig. 3b illustrate tuning curves Congruence between two sets across the whole distribution of of empirical data RR values (Fig. 4b). For all cases C (human 1, human 2) 0.967 0.976 0.981* with a good within-category C (monkey Y, monkey E) 0.985* 0.972 0.990* selectivity (that is, a response difC (humans, monkeys) 0.987 0.988 0.992* ference between preferred and C (monkeys, neurons) 0.990* 0.991* 0.984* least preferred shape above 50%, +Probability of random configuration, p < 0.001. *Probability that deviations from pixel-based similarities are n = 121), the median RR index random, p < 0.05. was 1.9. Thus, the responses of most IT neurons were closely related to the parametric variation of shape similarity. be illustrated by sorting the stimuli according to response We used the responses of the 124 neurons to compute the neustrength, Supplementary Figs. 1 and 2). ron-based similarity between each pair of stimuli. NotwithstandThe within-category selectivity of these neurons was regular, ing the low between-group selectivity of many neurons, hierarchical meaning that their responses decreased monotonically with cluster analysis of the neuron-based similarities revealed the expectincreasing parametric distance from the preferred shape of a stimed clustering at the population level: all stimuli from a group were ulus group. For example, the neuron in Fig. 3a responded well to


Table 1. Congruence between configurations for each group of stimuli.

a

b

Fig. 3. Responses of single IT neurons. (a) Peristimulus time histograms of a single IT neuron from monkey M. The ordering and numbering of the panels is the same as in Fig. 1. Stimulus presentation (300 ms) is indicated by the bar underneath each histogram. The histograms have a bin width of 25 ms and the height of each bin is normalized to the maximum bin across all histograms (94 spikes/s). (b) Polar plots of the within-group response pattern, making use of the radial position of each stimulus with respect to the center of the parametric square configuration (numbering of shapes is the same as in a). The black tuning curves represent the responses of the neuron of (a), whereas the responses of two other single neurons (monkey F) are shown in red and blue. Responses across all three panels are normalized to the maximum response for each neuron separately (maximum responses were 27, 59 and 34 spikes/s for the black, blue and red curves, respectively). The standard error of the mean for the maximum of each tuning curve is indicated. The RR indices for stimulus groups A, B and C were 1.8, 3.6 and 2.8, respectively, for the neuron indicated by black lines (SE = 27); 5.2, 3.0 and 4.8 for the neuron indicated by blue lines (SE = 2.3); and 1.6, 0.68 and 0.33 for the neuron indicated by red lines (SE = 3.8). nature neuroscience • volume 4 no 12 • december 2001

1247


articles

ties for all stimulus groups (Table 1; C(monkeys, neurons)). The neuron-based similarities were a better predictor of the averaged perceived similarities than were the pixel-based similarities (the same result was obtained with the behavioral data considered for each monkey separately). The neural as well as the perceptual representation of stimulus A3 tended more toward the center of the configuration than expected from the pixel-based configurations. For group B, both the averaged perceived similarities and the neuron-based Fig 4. Distribution of SE and RR indices. (a) The SE index is calculated for all neurons responding configurations tend to differentiate the with at least twice as many spikes to their preferred shape than to the least preferred shape stimuli on the right of the parametric con(n = 102). (b) The RR index is computed for all cases with a good within-category selectivity (that figuration more strongly than those on the is, a response difference between preferred and least preferred shape above 50%, n = 121). All val- left (Fig. 1b; although the behavior-based ues above five are collapsed in one column (maximum was 38 and 14 for SE and RR, respectively). representation of the stimuli on the left is The median of each distribution is indicated by an arrow. not consistent between monkeys, each animal was more sensitive for differences between stimuli on the right compared to stimuli on the left, as was the population of neurons). The results assigned to the same cluster, before they were clustered with stimfor group C indicate that the neurons tend to be more sensitive uli from other groups. Because 45% of the variance was explained for differences along the vertical parametric dimension, as were by assigning the stimuli to 3 groups, the degree of clustering at the the animals and human subjects at the behavioral level. neuronal level is comparable to that found in the monkey behavioral data, but less so than in human subjects. For each stimulus group, the congruence between the neuEffects of categorization on neuronal representation ron-based similarities and the pixel-based similarities was highDuring the recordings, the monkeys learned to categorize the ly significant (Table 1). The proportion of variance explained by shapes from each stimulus group into two classes (Fig. 6). In 2-dimensional MDS-derived configurations (Fig. 2f) exceeded some stimulus groups, the division of the stimulus space into 99% for each group. As found for the perceived similarities, the specific regions followed a simple linear decision rule (group C positions of the eight shapes in these configurations captured and A in monkeys F and M, respectively). In other groups, howmost aspects of the expected configurations. First, we found the ever, the categorization rule required the stimulus space to be same stimulus order in the neuron-based configurations comparceled into specific quadrants (group B in monkey M) or into pared to the expected configurations for stimulus groups A and arbitrary decision regions (groups A and B in monkey F and C (r s = 1, p < 0.01), and only a small deviation for group B group C in monkey M). In this last category rule, highly similar shapes require a different response. (rs = 0.98, p < 0.01). Second, the MDS-derived configurations A possible effect of categorizing the images during the were square-like, with larger angles at side points compared to recordings on the neural responses was examined by comparing corner points (χ2 = 8.17, p < 0.01). the neuronal data between monkeys for each stimulus group Reasonable fits could already be obtained with smaller samples separately. Indeed, each monkey learned a different rule for a of neurons. The mean congruence between neuron- and pixelparticular group, allowing a comparison between the neural based similarities was significant with as few as eight neurons (Fig. 5a). The congruence depends also on the tuning properties of the neurons in b addition to the effect of sample size (Fig. a 5b). First, congruence was lower for a sample of neurons with an irregular tuning (RR < median RR) compared to neurons with a regular tuning (RR > median RR). Second, restricting the range of preferred shapes (tuning optima) reduced congruence when combining at least four neurons. Thus, both the regularity of the tuning of single neurons and the range of their preferred shapes contribute to the efficiency of the coding of shape similarity at the population level. Fig. 5. Congruence (C) between pixel-based and neuron-based similarities as a function of number The neuron-based similarities seem to and type of neurons. Each data point represents the mean congruence of 150 random samples faithfully represent many metric aspects of (whiskers indicate the standard error of the mean). (a) Congruence as a function of sample size. the perceived similarities of the monkey. The ordinate scale starts from the expected value when the pixel-based similarities are compared Indeed, the congruence between the behav- with the similarities in random configurations (C = 0.86). The congruence of an individual sample is significantly different from this expected value for all values above the dashed line (C = 0.96). (b) ioral and neuronal data was greater than Difference between the congruence from (a) and the congruence obtained when neurons are would be expected based on their respective selected from subsets of the population. A positive difference corresponds to a higher congruence congruences with the pixel-based similari- relative to that shown in (a). RR, ratio index.


a

1248

b



articles


a

b

Fig. 6. The division of each stimulus group into two response categories for each monkey. Filled and open squares refer to stimuli that were associated with a leftward or a rightward eye movement, respectively. The stimulus order is the same as in Fig. 1.

similarity spaces for different category rules. We observed no systematic effects of categorization. First, the within-category selectivity of single neurons was regular with respect to the parametric similarity between stimuli and it did not follow the category rule. For example, when a monkey had learned an arbitrary category rule, no neurons responded stronger to all stimuli from one category compared to the responses to the other category. Second, arbitrary rules did not change the dimensionality or the order of shapes within the MDS-derived configurations. Third, even more subtle metric changes, like expansions/contractions of dimensions or parts of the stimulus space, were not induced by the categorization task. The effect of similarity on categorization performance If stimulus similarity determines categorization performance, then several predictions can be made regarding the difficulty of learning the different category rules. First, the more tightly stimuli from a given category are clustered (that is, the larger the inter-category distance relative to the intra-category distance),

a

the easier it will be to categorize any given stimulus. Second, the more closely a stimulus is located to a boundary between two categories, the more difficult it will be to assign it to one or the other of those categories. The errors the monkeys made while learning these categorization problems confirmed both predictions. First, the more tightly that shapes having to be categorized into the same class are clustered, the more easily a particular categorization was learned (Fig. 7a). Linear rules were learned more easily than a quadrant rule, but even the latter was more easily learned than the arbitrary rules. Second, in the case of a linear rule, the category assignments of shapes that were close to a category boundary were learned more slowly compared to more distantly located shapes (Fig. 7b). Nevertheless, performance on the stimuli close to a linear category border was above 90% after one week of training. This shows that the monkeys were able to discriminate neighboring stimuli in the parametric space. So, the difficulties with the arbitrary rules (performance after several weeks of training lower than 85%, Fig. 7a) were not merely due to stimulus discriminability, but indeed reflect the clustering of the stimuli within the representation space.

DISCUSSION Our data reveal a rather faithful representation of the physical similarities among high-dimensional stimuli in monkeys at the neuronal level. The neuronal representation of a configuration of shapes preserves the dimensionality and stimulus order of the configuration, and even some metric properties like the difference between a square and circle configuration. The response patterns of single neurons were related to the distance between shapes within these lowdimensional representation spaces, a property that contributes to the efficient coding of similarity at the population level. Notwithstanding the close agreement between physical and neuron-based similarities, our results revealed that the neuron-based similarities were a better predictor of perceived similarities than were the physical similarities. Finally, the low-dimensional representation of shape similarity determined the difficulty of learning specific categorization rules, but the reverse was not the case: applying a categorization rule did not change the representation of similarity in IT.

b

Fig. 7. The performance of the two monkeys in the categorization task for each stimulus group. The stimulus–response associations for each group and monkey are shown in Fig. 4. (a) Performance averaged across all stimuli from each stimulus group for different sessions: day 1 and 2 (performance at the end of the first and second session, respectively), and rec 1 and 2 (behavioral performance during the first and second half of the recording sessions, respectively). Point style represents the type of category rule, whereas the line style refers to the different stimulus groups (group A, dotted line; B, dashed line; C, solid line). Vertical bars indicate 95% confidence intervals of one line plot (the number of observations is equal for the other stimulus groups). (b) Performance for the linear categorization problems in the first two sessions as a function of the distance between each stimulus and the optimal decision boundary in parametric space. Pairwise close stimuli are C1, C2, and C5, C6 for monkey F, and A1, A2, and A5, A6 for monkey M. Vertical bars indicate 95% confidence intervals for performance with close stimuli in each monkey. nature neuroscience • volume 4 no 12 • december 2001

1249


articles


Fig. 8. Possible responses of a hypothetical neuron. The stimuli to which the neuron responds are indicated with open squares; black circles indicate no response. (a) A representation of three stimuli within two dimensions. (b) Including a parametric variation of the similarity between stimuli 1 and 3 requires the addition of a third dimension for all similarities between stimuli to be captured.

Previous behavioral studies have already shown that humans and monkeys represent parameterized two- and threedimensional shapes in a manner that preserves relative similarities among the stimuli in parameter space (refs. 9–12 and Sigala et al., Soc. Neurosci. Abstr. 26, 448.10, 2000). Computational work21 has revealed that such representation of similarity can be simulated by a network composed of units that encode shapes by computing their similarity to reference shapes (‘radial basis functions’22,23). Our data indicate that shape similarity is implemented at the neuronal level in a similar way. First, we noted responses of single neurons within a single stimulus group that follow the pattern expected of radial basis functions: the responses to other shapes than the most optimal shape of a group (the ‘reference shape’) decreased gradually with increasing distance from the position of the reference shape in parameter space. The responses of single neurons across different stimulus groups seemed to contradict this conclusion, insofar as many neurons responded with similar strength to dissimilar shapes from different stimulus groups (Fig. 3). The same observation has been made in previous studies that revealed that images capable of activating an IT neuron need not be similar to one another17,18,24. It is tempting to relate this observation to the fact that the concept of radial basis function networks has been introduced to explain the representation of similar but not distant objects, and that objects that are highly dissimilar need not be embedded into the same low-dimensional space6. However, the previous difficulties with finding a correlation between shape similarity and the responses of IT neurons could also be related to the lack of proper stimulus parameterization. Indeed, the highdimensional nature of these complex stimuli makes it difficult to ascertain what relationship the shapes may have with regard to one another as long as similarity is not controlled for in a parametric way. For instance, consider a realistic pattern of responses to three dissimilar stimuli whose physical similarities can be described within a two-dimensional space (data points 1–3 in Fig. 8a). Within this stimulus set, a neuron responding strongly to stimuli 1 and 3 but not to stimulus 2 would exemplify a response pattern bearing no relationship to the similarity between the stimuli. However, including additional stimuli in the stimulus set (Fig. 8b) might reveal a more regular response pattern with a systematic unimodal tuning for a third stimulus dimension. Thus, we need a parametric control of similarity to arrive at a more principled understanding of the tuning characteristics of these neurons. Indeed, using parameterized stimuli, we found regular tuning curves of IT neurons for combinations of complex shape dimensions. A second point of agreement between our data and the computational efforts is the demonstration that a population of neurons tuned to a set of reference shapes can underlie a low-dimensional and ordinally faithful representation of shape configurations. In addition, our results confirm that both assumptions of the model, a regular tuning on one hand, and different optimal shapes for different neurons on the other hand, contribute to the efficiency of the coding of shape similarity. 1250

a

b

However, by examining consistent metric deviations from the expected representation spaces we found that the representation of similarity at the behavioral and neuronal level is not completely faithful at the metric level. Indeed, the neuron-based similarities were always a better predictor of the monkey-perceived similarities than were the pixel-based similarities. With the idea of a network using radial basis functions, one can account for most aspects of our data, but cannot predict these metric biases (although they could be explained ad hoc). We applied measures of physical similarity that make no assumptions about how the image is analyzed in the visual system (in contrast to, for example, a wavelet analysis with oriented filters). Of course, physical similarity can be quantified by an almost infinite number of measures beyond the limited set we have used, and some of these alternative measures could do better. But, even in the latter case, we need a model that can explain why some measures are better than others although they all agree in an ordinal sense. Many theoretical models have tried to characterize the nonlinear processing steps in the hierarchically organized visual system (for example, see refs. 25–28). They make different predictions about the occurrence of systematic biases toward a higher sensitivity for some stimulus differences than for others, but our study was not designed to differentiate between these models. In agreement with categorization models using the concept of similarity3–5, we showed that the representation of similarity has a profound influence on categorization performance. However, computing similarity is only a first step. Learning a particular categorization rule implies that the low-dimensional representation space is parceled into regions that require the same response. Neuropsychological theories of categorization localize the representation of stimulus similarity within extrastriate cortex, but the stimulus–response mapping is presumed to be done in other areas such as prefrontal cortex, hippocampus and basal ganglia 29–33 . In line with this strict segregation of stimulus–response associations from the representation of stimulus similarity, the representation spaces we found at the neuronal level in visual cortex were not altered by learning a specific categorization rule. Further research is needed to compare the visual responses in different brain regions using parameterized stimuli to see how these visual representation spaces are transformed into category or ‘response spaces.’

METHODS Subjects. Four monkeys (Macaca mulatta) and two naive humans were subjects. All procedures34,35 were approved by the K.U. Leuven Ethical Committee for animal experiments and followed NIH guidelines. Stimuli. The stimuli were 24 closed contours defined by 7 RFCs36, and were divided into 3 groups with distinct RFCs. Within each group, differences between shapes were induced by independently varying the amplitude of two RFCs, creating eight shapes arranged in a square-like manner in the two-dimensional amplitude space (Fig. 1). The shapes (maximum size, 6°) were presented on a gray background (luminance, nature neuroscience • volume 4 no 12 • december 2001


articles


17 candelas/m2 (cd/m2)) and filled with noise (pixel luminances, either 0.05 or 34 cd/m2). The mean luminance of the shapes and background was equal. Shape position was randomized within a square region ranging from 2° to 4° centered on the fixation spot, except during the singlecell recordings where all shapes were presented foveally. As a first physical dissimilarity measure, we computed the distance between two stimuli in a two-dimensional configuration equal to the parametric space. Second, the Euclidean and the city-block distance between two shapes, i and j, within the space of 256 × 256 pixels11,37, was computed: (Σx=1:256Σy=1:256 |pixi(x, y) – pixj(x, y)|r)1/r

(1)

For shape and background pixels, pix(x, y) is 1 and 0, respectively; for city-block and Euclidean metrics, r is 1 and 2, respectively. Finally, we calculated the average information that a pixel of one image provides about the corresponding pixel in another image 38. The congruence between these pixel-based dissimilarity measures was high. The Euclidean distances provided the best fit with the behavioral and neuron-based similarity measures and are used in the Results. Behavioral tasks. Monkeys Y and E were trained in a same-different task with fixation control39. After 700-ms fixation (window, 2°), two shapes were shown successively for 300 ms (interstimulus interval, 500 ms). A leftward and rightward saccade was the correct response in trials with identical (50% of the trials) or different shapes, respectively. Aborted trials (for example, responses before the end of the second stimulus) were not included in the analyses. Correct responses were rewarded by juice. Each monkey was trained for several months in this task using a wide variety of images. Their performance level for highly dissimilar novel stimuli was 95% correct. During the experiment, all pairs of shapes were presented for 16 trials each, whereas additional testing (72 trials/pair) was done for pairs of shapes within a stimulus group. One additional session was run for monkey Y with group B stimuli (34 trials/pair). The performance level was 79% (Y) and 88% (E) correct when all shapes were paired and 69% (Y) and 80% (E) correct when only within-group comparisons were shown, which is as good as in other studies12,20. The 2 other monkeys learned to categorize the 24 shapes into 2 response categories (Fig. 6). The procedure was the same as for monkey Y, except that only one shape was shown, after which the monkeys had to make either a leftward or a rightward eye movement. Each monkey had several months of experience in this task with other stimuli. Similarity judgments from the human subjects were obtained with a rating method. The rating and the same-different technique produce equivalent results11,40. The subjects were asked to fixate a spot while two stimuli were shown for 300 ms (interstimulus interval, 500 ms). They had to rate the similarity between the two stimuli on a scale from one (very similar) to nine (very dissimilar). All possible pairs of shapes were presented four times for each subject, and additional testing (10 trials/pair) was done for pairs of shapes within a stimulus group. Recordings. IT recordings34,35 were performed in monkeys F and M during the categorization task. Recording sites were verified using CT images with the guiding tube in situ in monkey M and post mortem in monkey F. We searched for responsive neurons by presenting all 24 shapes. Responsive neurons were investigated further by presenting all shapes for at least 6 trials (median 12 trials). All further analyses were done using the mean number of spikes in the 50–350-ms interval after stimulus onset after normalization to the maximum response. All neurons were shapeselective (one-way ANOVA41). The selectivity index (SE) compared the selectivity within the group to which the optimal stimulus belonged (maxR and minR(within) being the responses to the best and the worst stimulus within that group), with the maximum response elicited by a stimulus from the other groups (maxR(between)): SE = (maxR – minR(within))/(maxR – maxR(between))

(2)

SE ranges between 0 (no within-group selectivity) and infinity (no between-group selectivity). The within-group response pattern was represented on a polar plot using the radial position of each stimulus with respect to the center of nature neuroscience • volume 4 no 12 • december 2001

the parametric configuration, and the unimodality of the selectivity was determined by a fast Fourier transform of these plots42. Unimodal selectivities will be reflected in a Fourier spectrum dominated by the firstorder component. The ratio (RR) of the size of this first-order component to the largest of all other components was taken as a measure of the dominance of the first-order component. A high RR value reflects an unimodal tuning, but a low RR does not necessarily correspond to an irregular tuning. (Many factors contribute to a low RR, such as an asymmetrical but unimodal tuning.) So, the RR index underestimates the relationship between shape similarity and IT selectivity. To analyze the representation of the shape similarities at the population level, we computed the distance between a pair of stimuli i and j in the multidimensional space spanned by the responses of all neurons35,43,44: [(Σn = 1:124 |Respi(n) – Respj(n)|2)/124]1/2

(3)

Here, n is the cell number. The inverse of these distances measures the similarity of the neural representations of two shapes. The effect of tuning properties on the coding of similarity at the population level was assessed as follows. First, we drew a random sample of neurons (with replacement) from the population. The sample size ranged from 2 to 64, and 150 samples of each size were analyzed (50 for each stimulus group). For each stimulus group, only neurons that responded to at least one shape were included, giving a population size of 105, 109 and 117 for groups A, B and C, respectively. Because we found no consistent differences among stimulus groups, the results were pooled. Second, we selected neurons from four different subsets. For each subset, we selected 150 samples for each of 3 sizes (2, 4 or 8). In the first and second subset, we selected only neurons with RR lower and higher, respectively, than the median RR of the population. In the third subset, the neurons in a sample had similar preferred shapes. In each sample, the first neuron was randomly drawn but the other neurons were required to have tuning curves with an optimum at the same or a neighboring stimulus. In the fourth subset, the optima of the neurons were distributed (optima were 180° apart for two neurons, 90° apart for four neurons, and were at all positions for eight neurons). Analysis of similarity data. The different sets of similarity data were analyzed using Statistica software (StatSoft, Tulsa, Oklahoma). First, we checked whether the 24 shapes showed the expected clustering in three groups by performing hierarchical cluster analysis45. This algorithm starts from a configuration with as many clusters as stimuli, and groups similar stimuli in several steps (starting with the most similar stimuli) until all stimuli form one cluster. We describe the results obtained by using a weighted pair-group average, but other rules such as single and complete linkage revealed the same pattern of results. We have summarized the results by describing at which level stimuli are clustered with stimuli from the same and different stimulus groups. The proportion of variance in the original data that could be accounted for by three clusters of stimuli measures the degree of clustering. The dissimilarity between stimuli within a cluster was set to zero, whereas the ‘linkage distance’ at the level at which two clusters were grouped was taken as the dissimilarity between stimuli from different clusters. Second, we analyzed the similarity data from each stimulus group separately with nonmetric MultiDimensional Scaling (MDS)43,45. As a statistical criterion to decide whether the perceived and neuronbased similarities converged with the similarities from the parametric or pixel-based configurations, we computed the congruence coefficient C between different sets A and B of similarity data46: C = (Σn = 1:28 dA(n) × dB(n))/(ΣndA2(n) × Σn dB2(n))1/2

(4)

The index n provides a summation across the 28 pairs of distances. The closer C approximates 1, the better the fit. The significance of a C value was evaluated using Monte Carlo simulations (n = 1000) to determine the expected congruence when a square configuration is compared with a random two-dimensional configuration of eight points. This distribution approximated a normal distribution with mean 0.865 and a standard deviation of 0.029, meaning that a value above 0.960 is highly significant (p < 0.001), which is similar to the result found for a range 1251



articles

of configurations11. As this index is skewed to unity, small differences in C in the range of 0.95–1 are meaningful. Measurement noise would introduce deviations of a measured configuration with respect to the actual configuration that are unrelated from one set of measurements to the other. However, if the deviations are consistent across datasets, then one can conclude that the actual configuration is distorted with respect to the expected configuration. This is analogous to determining whether a correlation between two variables (the two measured configurations) can be explained by their correlations with a third variable (the expected configuration).To simulate the effect of measurement noise, we added uniform noise to the similarities expected in a square configuration. We used several noise magnitudes to cover the range of congruences found in our study. The fit between two noisy datasets was almost never higher than the best of both fits found by comparing each noisy dataset with the perfect square similarities (p < 0.001). Even a fit between two noisy sets that is larger than the worst but lower than the best of these is significant (p < 0.05). If a configuration is measured twice, then the difference between the measurements will be larger than the difference between one measurement and the configuration. If the difference between the two measurements is smaller, then probability is high that a systematic factor determines the way in which the measurements deviate from the expected configuration (Table 1). Note: Supplementary information is available on the Nature Neuroscience web site (http://neuroscience.nature.com/web_specials).

ACKNOWLEDGEMENTS This work was supported by the Geneeskundige Stichting Koningin Elizabeth, Interuniversitaire Attractiepolen (IUAP P4/22) and the University Research Council (IDO/98/002). We thank M. De Paep, P. Kayenbergh, G. Meulemans, G. Vanparrys, S. Van Wetter and G. Kayaert for technical assistance, and S. Raiguel and F. Wichmann for their comments on the manuscript. H.O.d.B. is a research assistant of the Fund for Scientific Research (FWO) Flanders.

RECEIVED 6 SEPTEMBER; ACCEPTED 25 OCTOBER 2001 1. Edelman, G. M. Neural Darwinism (Basic Books, New York, New York, 1987). 2. Smith, E. E. & Medin, D. L. Categories and Concepts (Harvard Univ. Press, Cambridge, Massachusetts, 1981). 3. Nosofsky, R. M. Choice, similarity, and the context theory of classification. J. Exp. Psychol. Learn. Mem. Cogn. 10, 104–114 (1984). 4. Ashby, F. G. & Perrin, N. A. Toward a unified theory of similarity and recognition. Psychol. Rev. 95, 124–150 (1988). 5. Shepard, R. N. Toward a universal law of generalization for psychological science. Science 237, 1317–1323 (1987). 6. Edelman, S. Representation and Recognition in Vision (MIT Press, Cambridge, Massachusetts, 1999). 7. Roweis, S. T. & Saul, L. K. Nonlinear dimensionality reduction by locally linear embedding. Science 290, 2323–2326 (2000). 8. Tenenbaum, J. B., de Silva, V. & Langford, J. C. A global geometric framework for nonlinear dimensionality reduction. Science 290, 2319–2323 (2000). 9. Shepard, R. N. & Cermak, G. W. Perceptual-cognitive explorations of a toroidal set of free-form stimuli. Cogn. Psychol. 4, 351–377 (1973). 10. Cortese, J. M. & Dyre, B. P. Perceived similarity of shapes generated from fourier descriptors. J. Exp. Psychol. Hum. Percept. Perform. 22, 133–143 (1996). 11. Cutzu, F. & Edelman, S. Representation of object similarity in human vision: psychophysics and a computational model. Vision Res. 38, 2229–2257 (1998). 12. Sugihara, T., Edelman, S. & Tanaka, K. Representation of objective similarity among three-dimensional shapes in the monkey. Biol. Cybern. 78, 1–7 (1998). 13. Dean, P. Effects of inferotemporal lesions on the behavior of monkeys. Psychol. Bull. 83, 41–71 (1976). 14. Logothetis, N. K. & Sheinberg, D. L. Visual object recognition. Annu. Rev. Neurosci. 19, 577–621 (1996). 15. Tanaka, K. Inferotemporal cortex and object vision. Annu. Rev. Neurosci. 19, 109–139 (1996).

1252

16. Desimone, R., Albright, T. D., Gross, C. G. & Bruce, C. Stimulus-selective properties of inferior temporal neurons in the macaque. J. Neurosci. 4, 2051–2062 (1984). 17. Gochin, P. M., Colombo, M., Dorfman, G. A., Gerstein, G. L. & Gross, C. G. Neural ensemble coding in inferior temporal cortex. J. Neurophysiol. 71, 2325–2337 (1994). 18. Wilkinson, F., Wilson, H. R. & Habak, C. Detection and recognition of radial frequency patterns. Vision Res. 38, 3555–3568 (1998). 19. Albright, T. D. & Gross, C. G. Do inferior temporal cortex neurons encode shape by acting as Fourier Descriptor filters? Proc. Int. Conf. Fuzzy Logic & Neural Networks, Izuka, Japan, 375–378 (1990). 20. Sands, S. F., Lincoln, C. E. & Wright, A. A. Pictorial similarity judgements and the organization of visual memory in the rhesus monkey. J. Exp. Psychol. 111, 369–389 (1982). 21. Edelman, S. Representation is representation of similarities. Behav. Brain Sci. 21, 449–467 (1998). 22. Poggio, T. & Girosi, F. Regularization algorithms for learning that are equivalent to multilayer networks. Science 247, 978–982 (1990). 23. Poggio, T. & Edelman, S. A network that learns to recognize threedimensional objects. Nature 343, 263–266 (1990). 24. Vogels, R. Categorization of complex visual images by rhesus monkeys. Part 2: single-cell study. Eur. J. Neurosci. 11, 1239–1255 (1999). 25. Marr, D. Vision: A Computational Investigation into the Human Representation and Processing of Visual Information (Freeman, San Francisco, California, 1982). 26. Riesenhuber, M. & Poggio, T. Models of object recognition. Nat. Neurosci. 3, 1199–1204 (2000). 27. Biederman, I. Recognition-by-components: a theory of human image understanding. Psychol Rev. 94, 115–147 (1987). 28. Wallis, G. & Rolls, E. T. Invariant face and object recognition in the visual system. Prog. Neurobiol. 51, 167–194 (1997). 29. Ashby, F. G., Alfonso-Reese, L. A., Turken, A. U. & Waldron, E. M. A neuropsychological theory of multiple systems in category learning. Psychol. Rev. 105, 441–481 (1998). 30. Murray, E. A., Bussey, T. J. & Wise, S. P. Role of prefrontal cortex in a network for arbitrary visuomotor mapping. Exp. Brain Res. 133, 114–129 (2000). 31. Wise, S. P. & Murray, E. A. Role of the hippocampal system in conditional motor learning: mapping antecedents to action. Hippocampus 9, 101–117 (1999). 32. Asaad, W. F., Rainer, G. & Miller, E. K. Neural activity in the primate prefrontal cortex during associative learning. Neuron 21, 1399–1407 (2000). 33. Freedman, D. J., Riesenhuber, M., Poggio, T. & Miller, E. K. Categorical representation of visual stimuli in the primate prefrontal cortex. Science 291, 312–316 (2001). 34. Op de Beeck, H. & Vogels, R. Spatial sensitivity of macaque inferior temporal neurons. J. Comp. Neurol. 426, 505–518 (2000). 35. Vogels, R., Biederman, I., Bar, M. & Lorincz, A. Inferior temporal neurons show greater sensitivity to nonaccidental than to metric shape differences. J. Cogn. Neurosci. 13, 444–453 (2001). 36. Zahn, C. T. & Roskies, R. Z. Fourier descriptors for plane closed curves. IEEE Trans. Comp. 21, 269–281 (1972). 37. Grill-Spector, K. et al. Differential processing of objects under various viewing conditions in the human lateral occipital complex. Neuron 24, 187–203 (1999). 38. Rieke, F., Warland, D., de Ruyter van Steveninck, R. R. & Bialek, W. Spikes: Exploring the Neural Code Vision (MIT Press, Cambridge, Massachusetts, 1997). 39. Robinson, D. A. A method of measuring eye movements using a scleral search coil in a magnetic field. IEEE Trans. Biomed. Eng. 101, 131–145 (1963). 40. Cortese, J. M. Perceptual Similarity of Closed Contours. Thesis, Univ. of Illinois (1992). 41. Kirk, R. E. Experimental Design: Procedure for the Behavioral Sciences (Brooks-Cole, Belmont, California, 1968). 42. Worgotter, F. & Eysel, U. T. Quantitative determination of orientational and directional components in the response of visual cortical cells to moving stimuli. Biol. Cybern. 57, 349–355 (1987). 43. Young, M. P. & Yamane, S. Sparse population coding of faces in the inferotemporal cortex. Science 256, 1327–1331 (1992). 44. Kobatake, E., Wang, G. & Tanaka, K. Effects of shape-discrimination training on the selectivity of inferotemporal cells in adult monkeys. J. Neurophysiol. 80, 324–330 (1998). 45. Shepard, R. N. Multidimensional scaling, tree-fitting, and clustering. Science 210, 390–398 (1980). 46. Borg, I. & Leutner, D. Measuring the similarity of MDS configurations. Multivar. Behav. Res. 20, 325–334 (1985).



articles


Genetic influences on brain structure Paul M. Thompson1, Tyrone D. Cannon2, Katherine L. Narr1, Theo van Erp2, Veli-Pekka Poutanen3, Matti Huttunen4, Jouko Lönnqvist4, Carl-Gustaf Standertskjöld-Nordenstam3, Jaakko Kaprio5, Mohammad Khaledy1, Rajneesh Dail1, Chris I. Zoumalan1 and Arthur W. Toga1 1 Laboratory of Neuro Imaging and Brain Mapping Division (Rm. 4238, Reed Neurological Research Center), Department of Neurology,

UCLA School of Medicine, 710 Westwood Plaza, Los Angeles, California 90095-1769, USA 2 Departments of Psychology, Psychiatry, and Human Genetics, UCLA School of Medicine, 1285 Franz Hall, 405 Hilgard Avenue,

Los Angeles, California 90095-1563 3 Department of Radiology, University of Helsinki Central Hospital, Meilahti Clinics, FIN-00290 Helsinki, Finland 4 Department of Mental Health and Alcohol Research, National Public Health Institute of Finland, Mannerheimintie 166, SF-00300 Helsinki, Finland 5 Department of Public Health, Universities of Helsinki and Oulu, P.O. Box 41, Mannerheimintie 172, University of Helsinki,

FIN-00014 Helsinki, Finland Correspondence should be addressed to P.T. ([email protected])

Published online: 5 November 2001, DOI: 10.1038/nn758 Here we report on detailed three-dimensional maps revealing how brain structure is influenced by individual genetic differences. A genetic continuum was detected in which brain structure was increasingly similar in subjects with increasing genetic affinity. Genetic factors significantly influenced cortical structure in Broca’s and Wernicke’s language areas, as well as frontal brain regions (r2MZ > 0.8, p < 0.05). Preliminary correlations were performed suggesting that frontal gray matter differences may be linked to Spearman’s g, which measures successful test performance across multiple cognitive domains (p < 0.05). These genetic brain maps reveal how genes determine individual differences, and may shed light on the heritability of cognitive and linguistic skills, as well as genetic liability for diseases that affect the human cortex.

The degree to which genes and environment determine brain structure and function is of fundamental importance. Largescale neuroimaging and genetic studies are beginning to uncover normal and disease-specific patterns of gene and brain function in large human populations1,2. Yet, little is known about the genetic control of human brain structure, and how much individual genotype accounts for the wide variations among individual brains. Recent reports show that many cognitive skills are surprisingly heritable, with strong genetic influences on IQ3,4, verbal and spatial abilities, perceptual speed5 and even some personality qualities, including emotional reactions to stress6. These genetic relationships persist even after statistical adjustments are made for shared family environments, which tend to make members of the same family more similar. Given that genetic and environmental factors, in utero and throughout lifetime, shape the physical development of the brain, we aimed to map patterns of brain structure that are under significant genetic control, and determine whether these structural features are linked with measurable differences in cognitive function. The few existing studies of brain structure in twins suggest that the overall volume of the brain itself7 and some brain structures, including the corpus callosum8,9 and ventricles, are somewhat genetically influenced, whereas gyral patterns, observed qualitatively10 or by comparing their twodimensional projections, are much less heritable11. To make the transition from volumes of structures to detailed maps of genetic influences, advances in brain mapping technology have allowed the detailed mapping of structural features of the human cortex, including gray matter distribution, gyral patnature neuroscience • volume 4 no 12 • december 2001

terning, and brain asymmetry. These features each vary with age, gender, handedness, hemispheric dominance and cognitive performance in both health and disease. Composite maps of these features, generated for large populations, can reveal patterns not observable in an individual 12 . Such patterns include statistical maps that show whether heredity and nongenetic factors are involved in determining specific aspects of brain structure. Among the structural features that are genetically regulated and have implications for cortical function is the distribution of gray matter across the cortex. This varies widely across normal individuals, with developmental waves of gray matter gain and loss subsiding by adulthood13, and complex deficit patterns observed in Alzheimer’s disease, schizophrenia, and healthy subjects at genetic risk for these disorders. In this study, we began by comparing the average differences in gray matter (Fig. 1) in groups of unrelated subjects, dizygotic (DZ) and monozygotic (MZ) twins (see Methods). Although both types of twins share gestational and postgestational rearing environments, DZ twins share, on average, half their segregating genes, whereas MZ twins are normally genetically identical (with rare exceptions due to somatic mutations). We found that brain structure is under significant genetic control, in a broad anatomical region that includes frontal and language-related cortices. The quantity of frontal gray matter, in particular, was most similar in individuals who were genetically alike; intriguingly, these individual differences in brain structure were tightly linked with individual differences in IQ (intelligence quotient). The resulting genetic brain maps reveal a strong relationship between genes, brain structure and behavior, suggest1253



articles

Fig. 1. Genetic continuum of similarity in brain structure. Differences in the quantity of gray matter at each region of cortex were computed for identical and fraternal twins, averaged and compared with the average differences that would be found between pairs of randomly selected, unrelated individuals (blue, left). Color-coded maps show the percentage reduction in intra-pair variance for each cortical region. Fraternal twins exhibit only 30% of the normal inter-subject differences (red, middle), and these affinities are largely restricted to perisylvian language and spatial association cortices. Genetically identical twins display only 10–30% of normal differences (red and pink) in a large anatomical band spanning frontal (F), sensorimotor (S/M) and Wernicke’s (W) language cortices, suggesting strong genetic control of brain structure in these regions, but not others (blue; the significance of these effects is shown on the same color scale).

ing that highly heritable aspects of brain structure may be fundamental in determining individual differences in cognition.

RESULTS MZ within-pair gray matter differences were almost zero (intraclass r ≈ 0.9 and higher, p < 0.0001 corrected; Fig. 1, right column) in a broad anatomical band encompassing frontal, sensorimotor and linguistic cortices, including Broca’s speech and Wernicke’s language comprehension areas. Because MZ twins are genetically identical, any regional differences would be interpreted as being attributable to environmental effects or gene–environment interactions. Meanwhile, sensorimotor and parietal occipital but not frontal territory was significantly more similar in DZ twins than random pairs (Figs. 1 and 2). Affinity was greatest in the MZ pairs, suggesting a genetic continuum in the determination of structure. A genetic continuum In population genetics, a feature is heritable if it shows a genetic cascade in which within-pair correlations (Fig. 2) are highest for MZ twins, lower for DZ twin pairs and lowest of all for unrelated subjects. As we expected specific regions of cortex to be more heritable than others, we plotted these correlations across the cortex (Fig. 2) and assessed their statistical significance (see Methods). This uncovered a successively increasing influence of common genetics. A 95–100% correlation was revealed between MZ twins in frontal, linguistic and parietooccipital association cortices, suggesting individual differences in these regions can be largely attributed to genetic factors. DZ twins, who share half their genes on average, were still nearidentical in the supramarginal component of Wernicke’s lan1254

guage area (r2 = 0.7–0.8; p < 0.0001) and highly similar in parieto-occipital association areas (60–70% correlation; p < 0.001). They also showed significantly less affinity (p < 0.05) in a sharply defined region that included the frontal cortices (p > 0.05). The resulting pattern of twin correlations suggests substantial genetic influences in this region. Mapping genetic correlations With a sample size of only 40 twins, heritability coefficients cannot be estimated precisely, and limited statistical power precludes the detection of differences in heritability between individual regions of cortex. Preliminary comparisons of MZ and DZ correlations suggested that frontal, sensorimotor and anterior temporal cortices were under significant genetic control (p < 0.05, rejecting the hypothesis that heritability (h2) = 0; one-tailed). Preliminary estimates suggested that discrete middle frontal regions, near Brodmann areas 9 and 46 (ref. 27), displayed a 90–95% genetic determination of structure (that is, h2 ≈ 0.90–0.95). Many regions are under tight genetic control (bilateral frontal and sensorimotor regions, p < 0.0001; Fig. 3). Due to small sample sizes, any provisional heritability estimates should be interpreted with caution, but were comparable with twin-based estimates for the most highly genetically determined human traits, including fingerprint ridge count (h2 = 0.98), height (h2 = 0.66) and systolic blood pressure (h2 = 0.57)28. Genetic influences here are far higher than for the most environmentally influenced characters (such as social maturity, for which h2 = 0.16; ref. 29). Language asymmetry Given the high heritability of reading skills and performance on linguistic tasks30, we were interested in whether the structure of nature neuroscience • volume 4 no 12 • december 2001



articles

Fig. 2. Correlation between twins in gray matter distribution. Genetically identical twins are almost perfectly correlated in their gray matter distribution, with nearidentity in frontal (F), sensorimotor (S/M) and perisylvian language cortices. Fraternal twins are significantly less alike in frontal cortices, but are 90–100% correlated for gray matter in perisylvian language-related cortex, including supramarginal and angular territories and Wernicke’s language area (W). The significance of these increased similarities, visualized in color, is related to the local intra-class correlation coefficents (r).

language cortices would also be heritable, and if so, whether heritability would be higher in the left hemisphere, which is dominant for language in most (right-handed) subjects. The heritability of brain size does not vary markedly by hemisphere31, but one MZ twin study of cortical surface areas32 suggested the possibility of differing left/right genetic influences. Intriguingly, when a threedimensional map was made subtracting the heritability of the structures in one hemisphere from their counterparts in the other, differences in Wernicke’s language area were highly significant, even after hemispheric differences in gyral patterning were directly accommodated. Heritability was significantly greater on the left (p < 0.05, corrected). Although no other regions displayed this lateralized effect, we cannot infer that there are no such asymmetries elsewhere, as we were underpowered, in a sample of 40, to make general comparisons of heritability among cortical regions. Nonetheless, the asymmetry in language-related cortex was significant, and was corroborated by the genetic correlation maps as well (Figs. 2 and 3), in that Wernicke’s and Broca’s speech area displayed highly significant heritability on the left (p < 0.0001) but not on the right (p > 0.05). Cognitive linkages To make a preliminary assessment of whether gray matter differences between subjects were significantly linked with differences in cognitive function, a cognitive measure termed ‘Spearman’s g’ was assessed for all 40 twins. Like IQ, this widely used measure isolates a component of intellectual function common to multiple cognitive tests, and has been shown to be highly heritable across many studies, even more so than specific cognitive abilities (h 2 = 0.62 (ref. 4, compare with ref. 24);

h 2 = 0.48 (ref. 33); h 2 = 0.6–0.8 (ref. 34, compare with refs. 35–38)). We found that differences in frontal gray matter were significantly linked with differences in intellectual function (Table 1; p < 0.0044; p < 0.0176 after correction for multiple tests) as quantified by g, which was itself also highly heritable (h2 = 0.70 ± 0.17 in this study). Although these preliminary correlations should be evaluated in a larger sample, a recent abstract also observed that differences in regional gray matter volume were significantly correlated with differences in IQ, in a sample of 28 pediatric MZ twin pairs (mean age, 12.1 years) studied volumetrically (E. Molloy et al., 7th Annual Meeting of the Organization for Human Brain Mapping, 447, Brighton, England, 2001). In frontal brain regions, a regionally specific linkage has previously been found39 between g and metabolic activity measured by positron emission tomography (PET), suggesting that general cognitive ability may in part derive from a specific frontal system important in controlling diverse forms of behavior. Frontal regions also show task-dependent activity in tests involving working (short-term) memory, divided and sustained attention, and response selection40. Genetic factors may therefore contribute to structural differences in the brain that are statistically linked with cognitive differences. This is especially noteworthy, as cognitive performance seems to be linked with brain structure in the very regions where structure is under greatest genetic control (Figs. 2 and 3). This emphasizes the pronounced contribution of genetic factors to structural and functional differences across individuals, as detected here in frontal brain regions. Fig. 3. Significance of genetic control of gray matter distribution. Brain regions for which cortical gray matter distribution is under significant genetic control are shown in red. Frontal (F) and lateral temporal (T) regions show significant heritability, consistent with their near-identity in identical twins (Fig. 2) and the weaker patterns of correlations observed in fraternal twins, who have less similar genotypes. Wernicke’s area shows significantly higher heritability in the left hemisphere (Wleft), which is generally dominant for language function (p < 0.05 for asymmetry).


1255


articles

Table 1. Random effects analysis regressing individual regional gray matter measures on the IQ measure, Spearman’s g (n = 40 subjects). (a) Random effects analysis


Measure Whole brain gray matter volume Frontal gray matter volume Temporal gray matter volume Parietal gray matter volume Occipital gray matter volume

Controlling for overall gray matter only

Regression coefficient (β) 0.0037 0.072 0.039 0.055 0.033

Effect size (t) 1.73 1.95 0.23 0.41 0.15

After controlling for other predictors

Significance

F1,33

Significance

0.046 0.029** 0.411 0.343 0.439

3.92 9.37 3.77 0.54 0.04

0.0561 0.0044** 0.0607 0.4690 0.8376

(b) Correlation analyses in independent samples First sample using twin 1; n = 20 Frontal gray matter volume Independent sample using twin 2; n = 20 Frontal gray matter volume

r (twin 1) 0.45343 r (twin 2) 0.37392

Significance 0.0256** Significance 0.0574*

A highly significant relationship (**p < 0.0044) exists between gray matter volume in the frontal cortex and Spearman’s g. The random effects analysis (a) is most powerful47–49. It uses all 40 subjects’ data and it explicitly models and controls for correlations between twins in both measures47–49. The first columns show regression coefficients, effect sizes (t) and significance values for each regional brain measure, in a step-wise regression where only the predictive effects of overall gray matter volume (t = 1.73, p < 0.046) on IQ are factored out. In the final two columns, a Type III (simultaneous) regression model48 was used, meaning that each predictor was tested controlling for all other model terms simultaneously. In this second analysis, correlations between brain regions are accounted for in assessing the significance of each regional effect. An F statistic and a significance value are shown assessing the fit of each model parameter. Although the power is substantially less, correlations were also significant if analyses were restricted to independent samples including one twin from each pair. In (b), correlations are measured in 20 twins (arbitrarily termed ‘twin 1’) selected randomly, one from each of the 20 twin pairs (n = 20). Correlations are repeated in ‘twin 2’ (n = 20; using the other subject from each of the 20 twin pairs). Pearson partial correlation coefficients (r), and their significance levels (onetailed) are shown, suggesting in each independent sample a positive relationship between (greater) frontal gray matter and better cognitive performance. Although this analysis is slightly less powerful (L.A. Kurdek, Technical Report, Wright State University, Department of Psychology, 2001, available at http://www.psych.wright.edu/lkurdek/analyze.htm) due to splitting the sample into halves, the cognitive relationship appears at trend level in one sample (*p < 0.0574*) and significantly in the other (**p < 0.0256).

DISCUSSION Genetic brain mapping Influences of nature and nurture in the determination of individual brain structure are not independent; genes necessarily operate through the environment, particularly if they concern susceptibilities to environmental stressors or hazards41. Nonetheless, twin designs can reveal the degree to which heredity is involved, and the extent to which individual differences can be attributed to genetic and environmental factors. Whereas genetic influences strongly determine aspects of intellect and its closely related traits, the extent to which genes shape brain structure is heterogeneous. The gene control of brain structure displays asymmetries that mirror asymmetries in the brain’s functional organization, and genes strongly control a broad anatomical band encompassing frontal, linguistic and sensorimotor cortex. As with any polygenic trait, multiple genes are likely to combine additively or interact at the same or different loci (dominance or epistasis) to structure the adult brain. Future studies mapping quantitative trait loci are likely to provide insight into the genes that determine brain structure42, and neurocognitive skills that in some cases depend on it43. The tight coupling of brain structure and genetics, particularly in frontal brain regions, may contribute to the genetic liability for diseases that affect the integrity of the cortex. Frontal gray matter deficits are found in both schizophrenia patients and their healthy first-degree relatives44–46, and there is a strong familial risk for many neurodegenerative diseases that affect the frontal 1256

cortex, including frontotemporal dementia and primary progressive aphasia. The genetic cascades implicated in these diseases may or may not overlap with those involved in cortical determination, but the genetic coupling of brain structure we report here may result in increased familial liability to cortical degenerative disease, specifically in highly genetically determined frontal regions. By controlling for nongenetic factors, twin studies may offer unique advantages in isolating disease-specific differences in these highly heritable brain regions. Genetic brain maps, such as those introduced in this study, may reveal how genes determine individual differences in brain structure and function. Additional linkages were observed between cortical differences and intellectual function, suggesting that genetic brain mapping may shed light on the heritability of cognitive and linguistic skills, as well as familial liability for diseases that affect the human cortex.

METHODS Subjects. 40 healthy normal subjects, consisting of 10 monozygotic (MZ) and 10 dizygotic (DZ) twin pairs, were drawn from a twin cohort consisting of all the same-sex twins born in Finland between 1940 and 1957, inclusive, in which both members of each pair were alive and residing in Finland as of 1967 (n = 9,562 pairs, 2,495 MZ; 5,378 DZ; 1,689 of unknown zygosity)14. Pairs were excluded if either member or any of their first-degree relatives had a history of hospitalization, medicine prescriptions, or work disability due to a psychiatric indication from 1969 to 1991. MZ pairs were matched with the DZ pairs for age (48.2 ± 3.4 years), gender, handedness, duration of cohabitation and parental social class. nature neuroscience • volume 4 no 12 • december 2001


articles


Each zygosity group included five male pairs and five female pairs. The study protocol was reviewed and approved by the institutional review boards of the University of California (Los Angeles) and the National Public Health Institute of Finland, and all subjects signed IRB-approved informed-consent forms. Cognitive testing. Each twin in a pair received a neuropsychological test battery15 from a different examiner blind to zygosity. All subjects received the same test battery in a fixed order. Seventeen different cognitive domains were assessed, including verbal and spatial working memory, selective and divided attention, verbal knowledge, motor speed and visuospatial ability. A measure of general cognitive ability, in the form of an overall IQ, was prorated from age-scaled scores on the Vocabulary, Similarities, Block Design, and Digit Symbol subtests of the Wechsler Adult Intelligence Scale—Revised (WAIS-R; D. Wechsler, WAIS-R Manual, Psychological Corporation, Cleveland). This measure exhibited 98% correlation with full-scale IQ based on all of the WAIS-R subtests. Zygosity. For all pairs, zygosity was determined by DNA analysis using the following markers: DIS80 (20 alleles), DI7S30 (13 alleles), apoB (20 alleles), COL2A1 (10 alleles), vWA (9 alleles) and HUMTH01 (6 alleles). Assuming an average heterozygosity rate of 70% per marker, this procedure will falsely classify a DZ pair as MZ in approximately 1/482 cases. Magnetic resonance imaging. Three-dimensional maps of gray matter and models of cortical surface anatomy were derived from high-resolution three-dimensional (2562 × 124 resolution) T1-weighted (MPRAGE) magnetic resonance images acquired from all 40 subjects on a 1.5 T Siemens scanner (New York, New York). Image processing and analysis. A radio-frequency bias field correction algorithm eliminated intensity drifts due to scanner field inhomogeneity. A supervised tissue classifier generated detailed maps of gray matter, white matter and cerebrospinal fluid (CSF). Briefly, 120 samples of each tissue class were interactively tagged to compute the parameters of a Gaussian mixture distribution that reflects statistical variability in the intensity of each tissue type16,17. A nearest-neighbor tissue classifier assigned each image voxel to a particular tissue class (gray, white or CSF), or to a background class. The inter/intra-rater reliability of this protocol, and its robustness to changes in image acquisition parameters, have been described previously17. The error variance, that is, the variation associated with map error and reproducibility, was further confirmed to be small by the extremely high intra-class correlations in the MZ pairs (around 1.0), which would not otherwise be obtainable (Fig. 2). Gray matter maps were retained for subsequent analysis. Three-dimensional cortical maps. To facilitate comparison and pooling of cortical data across subjects, a high-resolution surface model of the cortex was automatically extracted for each subject18. Thirty-eight gyral and sulcal boundaries, representing the primary gyral pattern of each subject, were digitized on the highly magnified three-dimensional surface models. Gyral patterns and cortical models were used to compute a three-dimensional vector deformation field, which reconfigures each subject’s anatomy to the average configuration of the entire group (n = 40), matching landmark points, surfaces and curved anatomic interfaces. Data were accordingly averaged or compared, to the maximum possible degree, across corresponding cortical regions12. Additional threedimensional vector deformation fields reconfigured one twin’s anatomy into the shape of the other, matching landmark points, surfaces and curved anatomic interfaces on the pair of three-dimensional image sets. Given that the deformation maps associate cortical locations with the same relation to the primary folding pattern across subjects, a local measurement of gray matter density was made in each subject and averaged across equivalent cortical locations. Gray matter mapping. To quantify local gray matter, we used a measure termed ‘gray matter density,’ which has been used in previous studies to compare the spatial distribution of gray matter across subjects19,20,12. This measures the proportion of tissue that segments as gray matter in a small region of fixed radius (15 mm) around each cortical point. Given the large nature neuroscience • volume 4 no 12 • december 2001

anatomic variability in some cortical regions, high-dimensional elastic matching of cortical patterns21, constrained by all 1520 (40 × 38) threedimensional sulcal models, associated measures of gray matter density from homologous cortical regions across subjects. Maps of intra-pair gray matter differences, generated within each MZ and DZ pair, were subsequently elastically realigned for averaging across the 10 pairs within each group, before inter-group comparisons. Mapping genetic correlations and asymmetry. Intra-class correlation between pairs of each zygosity was computed at each cortical point, after testing for heteroscedastic variance across each group. First, to assess whether it was significantly non-zero, broad-sense heritability was computed using Falconer’s method22 to determine all genic influences on the phenotype (with heritability, h2, defined as twice the difference between MZ and DZ intra-class correlation coefficients). Because nongenetic familial effects contribute to the resemblance between relatives, such effects were accommodated, if not entirely eliminated, by assuming the same common environmental variance for MZ and DZ pairs (compare with ref. 4). For this study, random field models were preferred as opposed to a full structural equation model (for example, ref. 23) given the low degrees of freedom per point available to estimate dominance and epistatic variance terms and reject simpler models based on the available database of 40 scans. Interaction and gene–environment covariance terms, as well as unique and shared environment factors, may be estimable with a more general familial design, an adoption design, or by using sample sizes much larger than available in the present study23 (compare with ref. 24). The significance of genetic effects was computed point-wise by reference to an analytical null distribution (F-test) and was confirmed separately by assembling an empirical null distribution using 1,000,000 random pairings to avoid assuming bivariate normality. All map-based inferences were corrected for multiple comparisons by permutation. We used permutations to make statistical inferences that were not based on any assumptions about the error covariances. To correct for the multiple comparisons implicit in our brain maps we established the null distribution of the largest statistic over the voxels analyzed. By adopting the critical threshold of this largest statistic, we could then maintain both strong and weak control over false-positive rates over the voxels analyzed. Asymmetric heritability was tested by computing a set of 40 flows driving each subject’s left hemisphere model onto the right, matching gyral patterns, and computing a field of heritability differences. This field was compared with its standard error (pooled across contralateral cortical points, after testing equality of variance across hemispheres (compare with ref. 26)). Regions of asymmetric heritability were detected and their significance was assessed by permuting the covariate vector coding for hemisphere. Maps of MZ intra-pair gray matter differences associated with intra-pair differences in the cognitive measure, Spearman’s g, were generated by elastically realigning three-dimensional maps for averaging across all MZ twin pairs and modeling g as a continuous covariate for linkage with local gray matter distribution. Maps identifying these linkages were computed point-wise across the cortex and assessed statistically by permutation by computing the area of the average cortex with statistics above a fixed threshold in the significance maps (p < 0.01). Null distributions were assembled from random pairings of unrelated subjects. We preferred this to an analytical null distribution to avoid assuming that the smoothness tensor of the residuals of the statistical model was stationary across the cortical surface26. In each case, the covariate vector was permuted 1,000,000 times on an SGI RealityMonster supercomputer with 32 internal R10000 processors. An algorithm was then developed to report the significance probability for each map as a whole12, so the significance of intra-pair variance reduction by zygosity, heritability, asymmetry and cognitively linked patterns of gray matter distribution could be assessed after the appropriate correction for multiple comparisons.

ACKNOWLEDGEMENTS Grant support was provided by a P41 Resource Grant from the National Center for Research Resources (P.T., A.W.T.; RR13642) and by a National Institute of Mental Health grant (T.D.C.). Additional support for algorithm development

1257


articles

was provided by the National Library of Medicine, NINDS, the National Science Foundation, and a Human Brain Project grant to the International Consortium for Brain Mapping, funded jointly by NIMH and NIDA. Special thanks go to U. Mustonen, A. Tanksanen, T. Pirkola, and A. Tuulio-Henriksson for their contributions to subject recruitment and assessment.


RECEIVED 4 JUNE; ACCEPTED 19 OCTOBER 2001 1. Collins, F. S. & McKusick, V. A. Implications of the Human Genome Project for medical science. JAMA 285, 540–544 (2001). 2. Huerta, M. F. & Koslow, S. H. Neuroinformatics: opportunities across disciplinary and national borders. Neuroimage 4, 4–6 (1996). 3. Plomin, R. & Loehlin, J. C. Direct and indirect IQ heritability estimates: a puzzle. Behav. Genet. 19, 331–342 (1989). 4. McClearn, G. E. et al. Substantial genetic influence on cognitive abilities in twins 80 or more years old. Science 276, 1560–1563 (1997). 5. Alarcón, M., Plomin, R., Fulker, D. W., Corley, R. & DeFries, J. C. Multivariate path analysis of specific cognitive abilities data at 12 years of age in the Colorado Adoption Project. Behav. Genet. 28, 255–264 (1998). 6. Eley, T. C. & Plomin, R. Genetic analyses of emotionality. Curr. Opin. Neurobiol. 7, 279–284 (1997). 7. Tramo, M. J. et al. Brain size, head size, and intelligence quotient in monozygotic twins. Neurology 50, 1246–1252 (1998). 8. Oppenheim, J. S., Skerry, J. E., Tramo, M. J. & Gazzaniga, M. S. Magnetic resonance imaging morphology of the corpus callosum in monozygotic twins. Ann. Neurol. 26, 100–104 (1989). 9. Pfefferbaum, A., Sullivan, E. V., Swan, G. E. & Carmelli, D. Brain structure in men remains highly heritable in the seventh and eighth decades of life. Neurobiol. Aging 21, 63–74 (2000). 10. Biondi, A. et al. Are the brains of monozygotic twins similar? A threedimensional MR study. Am. J. Neuroradiol. 19, 1361–1367 (1998). 11. Bartley, A. J., Jones, D. W. & Weinberger, D. R. Genetic variability of human brain size and cortical gyral patterns. Brain 120, 257–269 (1997). 12. Thompson, P. M. et al. Cortical change in Alzheimer’s disease detected with a disease-specific population-based brain atlas. Cereb. Cortex 11, 1–16 (2001). 13. Giedd, J. N. et al. Brain development during childhood and adolescence: a longitudinal MRI study. Nat. Neurosci. 2, 861–863 (1999). 14. Kaprio, J., Koskenvuo, M. & Rose, R. J. Change in cohabitation and intrapair similarity of monozygotic (MZ) cotwins for alcohol use, extraversion, and neuroticism. Behav. Genet. 20, 265–276 (1990). 15. Cannon, T. D. et al. The inheritance of neuropsychological dysfunction in twins discordant for schizophrenia. Am. J. Hum. Gen. 67, 369–382 (2000). 16. Zijdenbos, A. P. & Dawant, B. M. Brain segmentation and white matter lesion detection in MR images. Crit. Rev. Biomed. Eng. 22, 401–465 (1994). 17. Sowell, E. R. et al. In vivo evidence for post-adolescent brain maturation in frontal and striatal regions. Nat. Neurosci. 2, 859–861 (1999). 18. MacDonald, D., Avis, D. & Evans, A. C. Multiple surface identification and matching in magnetic resonance images. Proc. SPIE 2359, 160–169 (1994). 19. Wright, I. C. et al. A voxel-based method for the statistical analysis of gray and white matter density applied to schizophrenia. Neuroimage 2, 244–252 (1995). 20. Ashburner, J. & Friston, K. J. Voxel-based morphometry—the methods. Neuroimage 11, 805–821 (2000). 21. Thompson, P. M., Woods, R. P., Mega, M. S. & Toga, A. W. Mathematical/computational challenges in creating deformable and probabilistic atlases of the human brain. Hum. Brain Mapp. 9, 81–92 (2000). 22. Falconer, D. S. Introduction to Quantitative Genetics Edn. 3 (Longman, Essex, UK, 1989). 23. Neale, M. C. & Cardon, L. R. Methodology for Genetic Studies of Twins and Families (Kluwer Academic, Boston, 1992).

1258

24. Feldman, M. W. & Otto, S. P. Twin studies, heritability, and intelligence. Science 278, 1383–1384 (1997). 25. Loehlin, J. C. & Nichols, R. C. Heredity, Environment and Personality (Univ. of Texas Press, Austin, Texas, 1976). 26. Thompson, P. M. et al. in Handbook on Medical Image Analysis (ed. Fitzpatrick, M.) 1066–1131 (SPIE, Bellingham, Washington, 2000). 27. Rajkowska, G. & Goldman-Rakic, P. S. Cytoarchitectonic definition of prefrontal areas in the normal human cortex: II. Variability in locations of areas 9 and 46 and relationship to the Talairach Coordinate System. Cereb. Cortex 5, 323–337 (1995). 28. Smith, C. in Textbook of Human Genetics (eds. Fraser, G. & Mayo, O. Blackwell, Oxford, 1975). 29. Huntley, R. M. C. in Genetic and Environmental Factors in Human Ability (eds. Meade, J. E. & Parkes, A. S., Oliver and Boyd, Edinburgh, Scotland, 1966). 30. Gayan, J. & Olson, R. K. Reading disability: evidence for a genetic etiology. Eur. Child Adolesc. Psychiatry 8, 52–55 (1999). 31. Pennington, B. F. et al. A twin MRI study of size variations in human brain. J. Cogn. Neurosci. 12, 223–232 (2000). 32. Tramo, M. J. et al. Surface area of human cerebral cortex and its gross morphological subdivisions. J. Cogn. Neurosci. 7, 292–301 (1995). 33. Devlin, B., Daniels, M. & Roeder, K. The heritability of IQ. Nature 388, 468–471 (1997). 34. Finkel, D. et al. Longitudinal and cross-sectional twin data on cognitive abilities in adulthood: the Swedish adoption/twin study of aging. Devel. Psychol. 34, 1400–1413 (1998). 35. Swan, G. E. et al. Heritability of cognitive performance in aging twins. The National Heart, Lung, and Blood Institute Twin Study. Arch. Neurol. 47, 259–262 (1990). 36. Loehlin, J. C. Partitioning environmental and genetic contributions to behavioral development. Am. Psychol. 44, 1285–1292 (1989). 37. Chipuer, H. M., Rovine, M. J., Plomin, R. LISREL Modeling: genetic and environmental influences on IQ revisited. Intelligence 14, 11–29 (1990). 38. Plomin, R. & Petrill, S.A. Genetics and intelligence: what’s new? Intelligence 24, 53–78 (1997). 39. Duncan, J. et al. A neural basis for general intelligence. Science 289, 457–460 (2000). 40. Funahashi, S., Bruce, C. J. & Goldman-Rakic, P. S. Mnemonic coding of visual space in the monkey’s dorsolateral prefrontal cortex. J. Neurophysiol. 61, 331–349 (1989). 41. Rutter, M., Silberg, J., O’Connor, T. & Simonoff, E. Genetics and child psychiatry: I. Advances in quantitative and molecular genetics. J. Child Psychol. Psychiatry 40, 3–18 (1999). 42. Williams, R. W., Strom, R. C., & Goldowitz, D. Natural variation in neuron number in mice is linked to a major quantitative trait locus on Chr 11. J. Neurosci. 18, 138–146 (1998). 43. Hill, L. et al. DNA pooling and dense marker maps: a systematic search for genes for cognitive ability. Neuroreport 10, 843–848 (1999). 44. Cannon, T. D. et al. Regional gray matter, white matter, and cerebrospinal fluid distributions in schizophrenic patients, their siblings, and controls. Arch. Gen. Psychiatry 51, 651–661 (1998). 45. Weinberger, D. R., DeLisi, L. E., Neophytides, A. N. & Wyatt, R. J. Familial aspects of CT scan abnormalities in chronic schizophrenic patients. Psychiatry Res. 4, 65–71 (1981). 46. Suddath, R. L. et al. Anatomical abnormalities in the brains of monozygotic twins discordant for schizophrenia. N. Engl. J. Med. 322, 789–794 (1990). 47. Kreft, I. & de Leeuw, J. Introducing Multilevel Modeling (Sage, London, 1998). 48. Littell, R. C., Milliken, G. A., Stroup, W. W. & Wolfinger, R. D. SAS System for Mixed Models (SAS Institute, Cary, North Carolina, 1996). 49. Hedeker, D., Gibbons, R. D. & Flay, B. R. Random-effects regression models for clustered data with an example from smoking prevention research. J. Consult. Clin. Psychol. 62, 757–765 (1994).



articles


Human memory formation is accompanied by rhinal–hippocampal coupling and decoupling Jürgen Fell1, Peter Klaver1, Klaus Lehnertz1, Thomas Grunwald1, Carlo Schaller2, Christian E. Elger1 and Guillén Fernández1 Departments of Epileptology1 and Neurosurgery2, University of Bonn, Sigmund-Freud Str. 25, D-53105 Bonn, Germany Correspondence should be addressed to J.F. ([email protected])

Published online: 5 November 2001, DOI: 10.1038/nn759 In humans, distinct processes within the hippocampus and rhinal cortex support declarative memory formation. But do these medial temporal lobe (MTL) substructures directly cooperate in encoding new memories? Phase synchronization of gamma-band electroencephalogram (EEG) oscillations (around 40 Hz) is a general mechanism of transiently connecting neural assemblies. We recorded depth-EEG from within the MTL of epilepsy patients performing a memorization task. Successful as opposed to unsuccessful memory formation was accompanied by an initial elevation of rhinal–hippocampal gamma synchronization followed by a later desynchronization, suggesting that effective declarative memory formation is accompanied by a direct and temporarily limited cooperation between both MTL substructures.

Lesions of certain MTL substructures, most notably of the hippocampus and the parahippocampal region, disturb the declarative memory system, the system that makes memory accessible to conscious recollection1–4. Neuroimaging studies suggest that both these MTL substructures participate in memory formation5–10. A process in the rhinal cortex, which consists of the histologically distinct entorhinal and perirhinal cortices and is part of the parahippocampal region, precedes a hippocampal process11. Anatomically, the perirhinal and parahippocampal cortices provide most of the neocortical input to the entorhinal cortex, which in turn provides the predominant cortical input to the hippocampus via the perforant path12–14. However, there is no stringent evidence for a direct interaction between rhinal cortex and hippocampus during declarative memory formation in humans15. Moreover, the exact time when such a presumed interaction might take place is unknown. Phase synchronization of gamma oscillations (electrical brain activity) of around 40 Hz is a general mechanism underlying transient functional coupling of neural assemblies16,17. This mechanism provides an explanation for the flexibility and specificity of functional associations between brain modules. Evidence has accumulated that phase coupling of induced (that is, not stimulus-locked) gamma activity is essential in object representation. Different object features processed by distinct neural assemblies are bound together to one coherent percept by synchronized induced activity in the gamma range18,19. Long-range association of cortical modules via induced gamma-band coupling subserves the integration of cognitive processes20–22. In contrast, the evoked (stimulus-locked) gamma response seems not to be involved in assembly coupling, and its functional role is still unclear19. Thus, we analyzed induced gamma synchronization of depth-EEG activity, which was recorded simultaneously nature neuroscience • volume 4 no 12 • december 2001

from rhinal and hippocampal electrodes in epilepsy patients during a word memorization task. We found successful declarative memory formation to be associated with a transient reduction of gamma power in rhinal and hippocampal recordings together with an initial enhancement of gamma synchronization between both MTL substructures, followed by a later desynchronization.

RESULTS We took EEG recordings from nine patients with unilateral temporal lobe epilepsy while they performed a single-trial word-list learning protocol with a free recall test after a distraction task. We implanted multicontact depth electrodes bilaterally along the longitudinal axis of each MTL during presurgical evaluation because the zone of seizure onset could not be determined unequivocally by noninvasive investigations23. Depth electrodes had a cylindrical surface area of 10 mm2. Sensitivity is maximal for field potentials generated within the adjacent brain tissue and, in general, decays with the inverse square of the distance24. For example, compared to adjacent brain tissue 0.1 mm away from the recording electrode (0.1 mm is the order of magnitude of the thickness of hippocampal layers), a source 1 cm away only contributes 0.01% to the recorded signal. The placement of electrode contacts within the hippocampus and the anterior parahippocampal gyrus, which is covered by rhinal cortex13, were ascertained by magnetic resonance images in each patient (Fig. 1). Only EEG recordings from the MTL contralateral to the zone of seizure origin were analyzed to reduce poorly controllable effects introduced by the epileptic process25. Moreover, none of the MTLs investigated here showed any pathology, such as hippocampal atrophy, on clinical MR scans performed during the presurgical work-up. Within these non-epileptic MTLs, we analyzed rhinal and hippocampal recordings with the best signal-to1259


articles

Rhinal cortex

b

Hippocampus


M

Coronal

Sagittal

noise ratio, assessed by amplitudes of event-related potentials (ERPs) recorded during the same task11. The average distance between the selected rhinal and hippocampal locations was 1.6 ± 0.25 cm (mean ± s.e.m.; range, 0.8–2.6 cm). Each patient participated in 20 study test blocks plus 2 training blocks immediately before the experiment, each containing 12 semantically unrelated German nouns. Patients were instructed to memorize each word presented sequentially on a computer monitor. To prevent ongoing rehearsal, a distraction task was conducted after each block. Thereafter, patients were asked to recall freely the previously displayed words in any order (mean recall rate, 29.7%; range, 20.0–54.6%). In accord with neuropsychological findings in large series of patients with temporal lobe epilepsy26, memory performance was poorer in the patients investigated here than in healthy subjects performing a similar task27. To compare successful and unsuccessful memory encoding, EEG was separated offline into segments for subsequently recalled and unrecalled items. To obtain an optimal time, as well as frequency resolution within the gamma band, we then applied a wavelet technique to the EEG instead of the traditional coherence method. EEG was wavelet-filtered in the gamma frequency range from 32 to 48 Hz (2 Hz steps). Because Fig. 2. Changes of phase synchronization between rhinal cortex and hippocampus (%) relative to prestimulus baseline for subsequently recalled versus unrecalled words. Synchronization values were averaged over all analyzed gamma frequencies (32–48 Hz); mean and s.e.m. are plotted. The x-axis depicts the time with respect to stimulus onset (word presentation). Time course of synchronization values was calculated from overlapping windows of 100 ms duration shifted in steps of 20 ms.

1260

Fig. 1. Localization of MTL depth electrodes. (a) Hippocampally adjusted axial MR image of a patient with bilateral depth electrodes in situ (Hi, hippocampus; R, right; L, left). Due to MR artifacts, the electrodes appear much larger than their actual size of 1 mm diameter. (b) Black dots (coronal) and bars (sagittal) in standardized drawings indicate the approximate location of MTL electrode contacts, which were used to record the EEG tracings analyzed in this study (Am, amygdala; Cs, collateral sulcus; Hi, hippocampus; Pg, parahippocampal gyrus; Rc, rhinal cortex).

we investigated induced gamma activity, that is, gamma activity occurring in a non-time-locked fashion in response to the stimuli, analysis was based on single-trial evaluations19. Phase synchronization values between electrode contacts within the rhinal cortex and the hippocampus were calculated from the individual wavelet-transformed EEG segments. The higher the synchronization value, the more constant is the phase difference between the two electrodes over all trials. Additionally, averaged power values were determined separately for rhinal and hippocampal recordings for subsequently recalled and unrecalled words. Finally, synchronization and power values were averaged for consecutive 100-ms time windows from –100 ms to 1500 ms relative to stimulus onset. The time course of phase synchronization between rhinal cortex and hippocampus averaged over all frequencies between 32 and 48 Hz shows a dissociation between subsequently recalled and unrecalled words starting within the first 100 ms after stimulus onset (Fig. 2; main effect of time, F14,1008 = 3.47, p < 0.001, ε = 0.60 and interaction of memory × time, F 14,1008 = 3.20, p < 0.01, ε = 0.50). Average gamma synchronization between rhinal and hippocampal recordings was increased by up to 16% for subsequently recalled as opposed to subsequently forgotten words from 100 up to 300 ms (main effects of memory, 100–200 ms, F1,72 = 9.39, p < 0.005; 200–300 ms, F1,72 = 3.94, p = 0.051). After the early enhancement in gamma synchronization, a second increase was detected from 500 to 600 ms (main effect of memory, F1,72 = 6.17, p < 0.02) and, finally, a decrease of synchronization was observed from 1000 to 1100 ms (main effect of memory, F1,72 = 6.62, p < 0.02; Fig. 2). The early synchronization effect is most pronounced in the frequency range from 36 to 40 Hz and for these frequencies reaches up to 30% (Fig. 3). The second synchronization increase (500–600 ms) and the later desynchronization are prominent in the frequency range from 32 Hz to 40 Hz. Phase lag distributions for both conditions (subsequently recalled and unrecalled words) have a Gaussian shape and are centered around zero (Fig. 4). The difference in synchronization for successful and unsuccessful encoding thus arises from a narrowing of the phase lag distribution caused by an increased amount of phase differences close to Percentage change of rhinal–hippocampal synchronization relative to baseline level (mean ± s.e.m.)

a

Time (s)




articles

Fig. 3. Differences of phase synchronization between rhinal cortex and hippocampus (%) relative to prestimulus baseline for subsequently recalled versus unrecalled words. Color-coded plot of S(recalled) – S(unrecalled), with S(recalled) = phase synchronization (%) for subsequently recalled words and S(unrecalled) = phase synchronization (%) for subsequently unrecalled words. The different gamma range frequencies (32–48 Hz) are represented in the y direction and time is depicted in the x direction. Synchronization/desynchronization is coded on a color scale: red areas show an enhancement; blue areas, a reduction of synchronization for subsequently recalled versus unrecalled words.

zero. This finding indicates that rhinal and hippocampal neurons oscillate together in a more synchronous rhythm when encoding leads to successful memory formation. To explore the possibility that the subsequent memory effects identified here could be related to general, ubiquitous effects found throughout the brain, we examined the synchronization between rhinal cortex and a temporolateral location (gyrus temporalis superior). For identification of the zone of seizure onset, EEG recordings were made from this location with subdural strip electrodes in seven of the nine patients28. The two remaining patients had no electrodes outside the MTL. The analysis of rhinal-temporolateral synchronization values revealed neither a memory effect (F1,54 = 1.20, p = 0.28), nor a memory × time interaction (F14,756 = 1.15, p = 0.33, ε = 0.63). Moreover, none of the individual time windows showed a statistically significant memory effect (each p > 0.05). Absolute gamma power values at hippocampal sites were about threefold larger than values from rhinal contacts (average over all frequencies, time windows and conditions, for hippocampus, 39.6 ± 39.8 µV2; for rhinal cortex, 14.2 ± 12.2 µV2; paired two-tailed t-tests for each individual frequency, condition and time window; all p < 0.05, T8 > 2.35). The time course of gamma power at rhinal cortex and hippocampus (Fig. 5) dissociated significantly between conditions (ANOVA effects, for rhinal cortex, time (F 14,1008 = 14.64, p < 10 –12 , ε = 0.41) and memory × time (F14,1008 = 4.64, p < 0.0001, ε = 0.54); for hippocampus, time (F14,1008 = 4.68, p < 0.0001, ε = 0.62) and memory × time (F14,1008 = 6.37, p < 10–7, ε = 0.59)). In the rhinal cortex, a gamma peak was observed for both conditions at around 250 ms. However, gamma power was reduced for subsequently recalled compared to unrecalled words (significant reductions from 600 to 800 ms and 1300 to 1400 ms). Similarly, in hipnature neuroscience • volume 4 no 12 • december 2001

pocampal recordings, gamma power was significantly diminished in EEG segments related to successful as opposed to unsuccessful memory formation. This difference was detected between 100 and 400 ms after stimulus onset. Finally, we compared absolute synchronization and power values for subsequently recalled and unrecalled words during the prestimulus baseline window (–100 to 0 ms) to examine whether EEG changes before word presentation are responsible for our findings, which would have suggested a slowly modulated encoding state rather than transient processes8. However, we did not find significant baseline differences between subsequently recalled and unrecalled trials for either rhinal–hippocampal synchronization values (F1,72 = 0.28, p = 0.60), or for gamma power from rhinal (F 1,72 = 1.85, p = 0.18) or hippocampal (F1,72 = 1.05, p = 0.31) recordings.

DISCUSSION Our results show EEG activity in the gamma frequency range in field recordings from within the human MTL during a memory task. A previous study29 revealed a generally higher gamma power in parahippocampal than neocortical recordings in humans. We extend this knowledge by showing that gamma power in hippocampal recordings is even threefold higher than in the parahippocampal region, suggesting that high-frequency oscillations of around 40 Hz have a prominent involvement in medial temporal and especially hippocampal information processing. Intracranial EEG recordings allow the reliable separation of synchronization and power effects. In view of the anatomical proximity of the inspected areas (mean distance, 1.6 cm), such a separation would be impossible with surface EEG recordings 30,31 . Previous ERP data indicate 11,32 that there is no detectable correlation between EEG recorded from within the 1261


articles

450

Subsequently recalled words Subsequently unrecalled words

350

300

250

200

150 -180

-90

0

90

Phase differences (degrees)

hippocampus and the rhinal cortex, even with electrode distances of less than 1 cm. The large anterior medial temporal lobe N400 component, for instance, which reflects word processing and can be recorded with an amplitude of up to 70 µV from rhinal cortex, is usually not observable in recordings from within the hippocampus11,27. On the other hand, hippocampal activity is shielded toward the outside by the radial cylindrical arrangement of hippocampal pyramidal layers33. Thus, it is highly unlikely that our results are biased by correlated EEG recordings. We found successful memory formation to be accompanied by two factors: an early increase and a later decrease in gamma synchronization between rhinal and hippocampal recording sites, and a transient reduction of gamma power at both locations partly within the same time window. The enhancement of gamma-band synchronization observed here between the rhinal cortex and the hippocampus occurs at zero-phase lag. Such a synchronization requires highly accurate timing of neural discharges within a time range of just a few milliseconds34. By achieving this, gamma synchronization enables a precise functional association between specific brain regions over short as well as longer distances16,17. Thus, gamma-band coupling between rhinal cortex and hippocampus is likely to establish a transient connection between both MTL structures initializing declarative memory formation. The time course of modulation of gamma synchronization found here is consistent with reports of altered firing rates of single anterior rhinal neurons within 200 ms after visual object presentation 35. However, it remains unclear whether semantic information provided by each stimulus is already available during the initiation of rhinal–hippocampal coupling or not. If not, directed attention might, in a first step, allocate specific connections necessary for memory formation before actual information transfer takes place. Attention-driven enhancement of gamma-band phase synchronization has been shown in several studies36–38. In principle, cortical regions like the prefrontal cortex and the superior temporal sulcus, which are anatomically connected with the MTL, could influence rhinal–hippocampal interaction. The early timing, however, suggests that the observed coupling might be initiated by a top-down process mediated directly by the thalamus39. The later decrease in synchronization (1000–1100 ms) may occur following information transfer from the parahippocampal region to

the hippocampus11 and terminate the communication between both structures. Such a functional decoupling has been found in a visual face perception task and has been termed active desynchronization21. The timing of rhinal–hippocampal coupling and 180 decoupling fits well with the sequence of processes as monitored by ERPs recorded separately from the rhinal cortex and hippocampus during the same task11. Rhinal ERPs in response to subsequently recalled words start to differ from ERPs in response to subsequently forgotten words about 300 ms after stimulus onset. This subsequent memory effect in the rhinal cortex is followed by a hippocampal effect some 200 ms later that lasts until about 2000 ms after stimulus onset. Assuming that rhinal–hippocampal information transfer occurs between the onset of the rhinal and the hippocampal ERP effects, the early beginning of gamma phase coupling revealed here would allow the preparation for and the actual transfer of information. The decoupling observed follows the end of the rhinal subsequent memory effect at about 900 ms after stimulus onset, the time point when information transfer to the hippocampus might be accomplished.

Percentage changes of gamma band power relative to baseline (mean ± s.e.m.)

Number of occurences


400

Fig. 4. Distribution of phase differences between rhinal cortex and hippocampus in the gamma frequency range (32–48 Hz) for subsequently recalled versus unrecalled words. Phase differences were evaluated from the time window between 100 and 200 ms after item presentation.

Time (s)

Fig. 5. Changes of EEG gamma power (%) averaged over all frequencies (32–48 Hz) relative to prestimulus baseline for subsequently recalled versus unrecalled words. Mean ± s.e.m. is plotted. The x-axis depicts the time with respect to stimulus onset (word presentation). Time (s)

1262




articles

Gamma oscillations induced in the CA1 field of hippocampal slices within the first second following stimulation are associated with a prolonged elevation of excitatory postsynaptic potentials 40,41 , suggesting a crucial involvement in synaptic plasticity, the synaptic correlate of memory formation 42 . These in vitro data and our findings underline the importance of medial temporal gamma activity in memory formation. A direct line linking these two sets of data, however, cannot be drawn, as we recorded macroscopic field potentials summing up hippocampal mass activity and were not able to distinguish oscillations generated specifically by distinct hippocampal subregions like the CA1 field. Our analysis of macroscopic field potentials revealed that efficient medial temporal information processing, leading to successful memory formation, correlates with reduced gamma power at both recording sites. The transient reduction of gamma oscillations might be explained by the necessity to suppress noise-like ambient gamma activity unrelated to specific study items43. One might speculate that in an event of unsuccessful encoding, ongoing background gamma activity interferes with item related activity and distorts the process of memory formation. Thus, reduced gamma power, as assessed here during successful encoding, might be a correlate of a higher specificity of local assembly activation. Another interpretation could be that gamma activity behaves like a hippocampal resting rhythm similar to the occipital alpha rhythm. In this picture, certain components of rhinal–hippocampal circuits might be shut off or reset during successful memory encoding. The results presented here suggest that gamma power reduction and appropriate neural coupling and decoupling interact in declarative memory formation within the human MTL. Our data do not exclude a third pacemaker site driving phase-locked gamma activity in rhinal cortex and hippocampus independently from each other. The strong anatomical connection between both structures13,14, however, supports the hypothesis of a direct rhinal–hippocampal interaction underlying gamma-band phase coupling and decoupling. Thus, our findings accord with models11,44, proposing that the formation of new declarative memories requires a direct cooperation between rhinal cortex and hippocampus.

METHODS Subjects. All 9 patients (6 women, 3 men; mean age, 34.1 ± 8.3 years) had pharmacoresistant unilateral temporal lobe epilepsies (mean duration of illness, 26.4 ± 9.1 years). They were native German speakers. At the time of the experiment, all patients received carbamazepine (plasma concentration 8 to 12 µg/ml) as the only centrally acting drug. During presurgical evaluation, at least three spontaneous seizures were recorded invasively using bilateral MTL depth electrodes in all patients and temporo-lateral strip electrodes in seven patients. In six patients, seizures originated exclusively from the right MTL; in three patients, exclusively from the left MTL. No seizure occurred within 24 hours before the investigation. After resection of the epileptic MTL, all patients remained free of seizures (follow-up, 6 to 15 months). In all patients, histopathological examination of tissue resected at the time of epilepsy surgery revealed hippocampal sclerosis. The EEG study was approved by the local medical ethics committee. Each patient gave written informed consent. Word memorization protocol. Words were presented in uppercase letters for 400 ms. Interstimulus intervals were randomized and ranged from 2.3 s to 2.7 s (mean 2.5 s). Word length ranged from 4 to 11 (mean 6) letters and word frequency ranged from 15 to 175 per million (mean 75 per million)45. Patients were asked to use a rote strategy to memorize each word, avoiding memory aids such as making rows, sentences, stories or pictures. During the distraction task, patients were instructed to count backward by threes, starting at a number between 81 and 99 displayed on screen. nature neuroscience • volume 4 no 12 • december 2001

EEG recording. Depth electroencephalograms were referenced to linked mastoids, bandpass-filtered (0.03 to 85 Hz, 6 dB/octave), and recorded with a sampling rate of 173 Hz (12-bit analog–digital conversion). To determine the anatomical positions of electrode contacts, MRI scans were acquired in sagittal and adjusted coronal planes, perpendicular to the longitudinal axis of the hippocampus. Electrode contacts were mapped by transferring their positions from MRI to standardized anatomical drawings46 (Fig. 1). EEG trials and corresponding power spectra were visually inspected for artifacts in the gamma frequency range and 4.9% of all trials were excluded from analysis. Measuring phase synchronization and power. EEG trials were filtered in the gamma frequency range from 32 Hz to 48 Hz (2-Hz steps) by wavelet transforms implementing Morlet wavelets of 7 cycles length. The filtered signals wj,k (j, time point within a trial; k, trial number) hereby result from the time convolution of original signals and the complex wavelet function47. From the wavelet transformed signals wj,k, the phases ϕj,k (ϕj,k = arctan (Im(wj,k)/Re(wj,k))) and the power values Pj,k (Pj,k = Re(wj,k)2 + Im(wj,k)2) were extracted for each time point j of each trial k. Power values were averaged separately for trials corresponding to subsequently recalled and unrecalled words. For each time point of each trial, phase differences ∆ϕj,k between hippocampal and rhinal electrode contacts were determined. Phase synchronization values Sj were calculated based on the definition of circular variance48.

Sj =

1 N

Σe N

i∆ ϕj , k

k=1

where N is the number of trials; Sj ∈ [0,1].

Different numbers of trials for subsequently recalled and unrecalled words would cause a bias in the absolute values of the synchronization measure. Therefore, trial numbers were adjusted between conditions using randomized trial lists for the condition with the originally larger trial number. Finally, power and phase synchronization values were averaged for non-overlapping successive time windows of 100 ms duration from –100 to 1500 ms (16 windows in total). Statistical analysis. Synchronization and power values were normalized with respect to prestimulus values (window 1) separately for each subject and each filter frequency. For statistical evaluation, we conducted threeway ANOVAs with time (15 windows) and memory (subsequently recalled versus unrecalled) as repeated measures, and frequency (9 levels) as independent variable. p-values were Huynh–Feldt corrected for inhomogeneities of covariance when necessary49. The notation of the F-values is Fx,y with x being the model degrees of freedom, that is, the number of adjustable parameters in the model, and y being the residual degrees of freedom, that is, the number of degrees of freedom that are not taken up by the model. In a subsidiary analysis, each time window was tested separately by two-way ANOVAs. Effects for time windows with p-values less than 0.05/15 (Bonferroni correction) and for doublets of neighboring time windows each with a p-value less than 0.05 (combined probability less than 0.052 × 14 = 0.035) were regarded as statistically significant.

ACKNOWLEDGEMENTS This research was supported by the Deutsche Forschungsgemeinschaft (DFG Fe479/4-1). We wish to thank H. Beck, W. Burr, A. Engel, C. Koch, M. Reuber and I. Tendolkar for comments on earlier drafts of the manuscript. We also thank H. Urbach for providing the MR images and I. Blümcke for providing the histopathological reports.

RECEIVED 12 JULY; ACCEPTED 10 OCTOBER 2001 1. Scoville, W. B. & Millner, B. J. Loss of recent memory after bilateral hippocampal lesions. Neurol. Neurosurg. Psychiatry 20, 11–21 (1957). 2. Eichenbaum, H. A cortical-hippocampal system for declarative memory. Nat. Rev. Neurosci. 1, 41–50 (2000). 3. Gabrieli, J. D. E. Cognitive neuroscience of human memory. Annu. Rev. Neurosci. 49, 87–115 (1998).

1263



articles

4. Zola-Morgan, S. & Squire, L. R. Neuroanatomy of memory. Annu. Rev. Neurosci. 16, 547–563 (1993). 5. Otten, L. J., Henson, R. N. A. & Rugg, M. D. Depth of processing effects on neural correlates of memory encoding. Brain 124, 399–412 (2001). 6. Kirchhoff, B. A., Wagner, A. D., Maril, A. & Stern, C. E. Prefrontal-temporal circuitry for episodic and subsequent memory. J. Neurosci. 20, 6173–6180 (2000). 7. Henke, K., Weber, B., Kneifel, S., Wieser, H. G. & Buck, A. Human hippocampus associates information in memory. Proc. Natl. Acad. Sci. USA 96, 5884–5889 (1999). 8. Fernández, G., Brewer, J. B., Zhao, Z., Glover, G. H. & Gabrieli, J. D. Level of sustained entorhinal activity at study correlates with subsequent cued-recall performance: a functional magnetic resonance imaging study with high acquisition rate. Hippocampus 9, 35–44 (1999). 9. Brewer, J. B., Zhao, Z., Desmond, J. E., Glover, G. H. & Gabrieli, J. D. E. Making memories: brain activity that predicts how well visual experience will be remembered. Science 281, 1185–1187 (1998). 10. Wagner, A. D. et al. Building memories: remembering and forgetting of verbal experiences as predicted by brain activity. Science 281, 1188–1191 (1998). 11. Fernández, G. et al. Real-time tracking of memory formation in the human rhinal cortex and hippocampus. Science 285, 1582–1585 (1999). 12. Lavenex, P. & Amaral, D. G. Hippocampal-neocortical interaction: a hierarchy of associativity. Hippocampus 10, 420–430 (2000). 13. Amaral, D. G. & Insausti, R. in The Human Nervous System (ed. Paxinos, G.) 711–755 (Academic, San Diego, 1990). 14. Witter, M. P., Groenewegen, H. J., Lopes da Silva, F. H. & Lohman, A. H. M. Functional organization of the extrinsic and intrinsic circuitry of the parahippocampal region. Prog. Neurobiol. 33, 161–253 (1989). 15. Brown, M. W. & Aggleton, J. P. Recognition memory: what are the roles of the perirhinal cortex and hippocampus? Nat. Rev. Neurosci. 2, 51–61 (2001). 16. Engel, A. K. & Singer, W. Temporal binding and the neural correlates of sensory awareness. Trends Cogn. Sci. 5, 16–25 (2001). 17. Varela, F. J., Lachaux, J.-P., Rodriguez, E. & Martinerie, J. The brainweb: phase synchronization and large-scale integration. Nat. Rev. Neurosci. 2, 229–239 (2000). 18. Sauvé, K. Gamma-band synchronous oscillations: recent evidence regarding their functional significance. Conscious. Cogn. 8, 213–224 (1999). 19. Tallon-Baudry, C. & Bertrand, O. Oscillatory gamma activity in humans and its role in object representation. Trends Cogn. Sci. 3, 151–162 (1999). 20. Singer, W. Striving for coherence. Nature 397, 391–393 (1999). 21. Rodriguez, E. et al. Perception’s shadow: long-distance synchronization of human brain activity. Nature 397, 430–433 (1999). 22. Miltner, W. H., Braun, C., Arnold, M., Witte, H. & Taub, E. Coherence of gamma-band EEG activity as a basis for associative learning. Nature 397, 434–436 (1999). 23. Van Roost, D., Solymosi, L., Schramm, J., Van Oosterwyck, B. & Elger, C. E. Depth electrode implantation in the length axis of the hippocampus for the presurgical evaluation of medial temporal lobe epilepsy: a computed tomography-based stereotactic insertion technique and its accuracy. Neurosurgery 43, 819–826 (1998). 24. Morris, H. H. & Luders, H. Electrodes. Electroencephalogr. Clin. Neurophysiol. Suppl. 37, 3–26 (1985). 25. Paller, K. A., McCarthy, G., Roessler, E., Allison, T. & Wood, C. C. Potentials evoked in human and monkey medial temporal lobe during auditory and visual oddball paradigms. Electroencephalogr. Clin. Neurophysiol. 84, 269–279 (1992). 26. Hermann, B. P., Seidenberg, M., Schoenfeld, J. & Davies, K. Neuropsychological characteristics of the syndrome of mesial temporal lobe epilepsy. Arch. Neurol. 54, 369–376 (1997). 27. Fernández, G. et al. Event-related potentials of verbal encoding into episodic

1264

28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40.

41.

42. 43. 44. 45. 46. 47. 48. 49.

memory: dissociation between the effects of subsequent memory performance and distinctiveness. Psychophysiology 35, 709–720 (1998). Elger, C. E. et al. Human temporal lobe potentials in verbal learning and memory processes. Neuropsychologia 35, 657–667 (1997). Hirai, N., Uchida, S., Maehara, T., Okubo, Y. & Shimizu, H. Enhanced gamma (30–150 Hz) frequency in the human medial temporal lobe. Neuroscience 90, 1149–1155 (1999). Menon, V. et al. Spatio-temporal correlations in human gamma band electrocorticograms. Electroencephalogr. Clin. Neurophysiol. 98, 89–102 (1996). Bullock, T. H. et al. EEG coherence has structure in the millimetre domain: subdural and hippocampal recordings from epileptic patients. Electroencephalogr. Clin. Neurophysiol. 95, 161–177 (1995). McCarthy, G., Nobre, A. C., Bentin, S. & Spencer, D. D. Language-related field potentials in the anterior-medial temporal lobe: I. Intracranial distribution and neural generators. J. Neurosci. 15, 1080–1089 (1995). Klee, M., & Rall, W. Computed potentials of cortically arranged populations of neurons. J. Neurophysiol. 40, 647–666 (1977). Traub, R. D., Whittington, M. A., Stanford, I. M. & Jefferys, J. G. R. A mechanism for generation of long-range synchronous fast oscillations in the cortex. Nature 383, 621–624 (1996). Kreiman, G., Koch, C. & Fried, I. Imagery neurons in the human brain. Nature 408, 357–361 (2000). Steinmetz, P. N. et al. Attention modulates synchronized neuronal firing in primate somatosensory cortex. Nature 404, 187–190 (2000). Desmedt, J. E. & Tomberg, C. Transient phase-locking of 40 Hz electrical oscillations in prefrontal and parietal human cortex reflects the process of conscious somatic perception. Neurosci. Lett. 168, 126–129 (1994). Müller, M. M., Gruber, T. & Keil, A. Modulation of induced gamma band activity in the human EEG by attention and visual information processing. Int. J. Psychophysiol. 38, 283–299 (2000). LaBerge, D. Attention, awareness, and the triangular circuit. Conscious. Cogn. 6, 149–181 (1997). Traub, R. D., Whittington, M. A., Buhl, E. H., Jefferys, J. G. & Faulkner, H. J. On the mechanism of the γ_β frequency shift in neuronal oscillations induced in rat hippocampal slices by tetanic stimulation. J. Neurosci. 19, 1088–1105 (1999). Whittington, M. A., Traub, R. D., Faulkner, H. J., Stanford, I. M. & Jefferys, J. G. Recurrent excitatory postsynaptic potentials induced by synchronized fast cortical oscillations. Proc. Natl. Acad. Sci. USA 94, 12198–12203 (1997). Bliss, T. V. P. & Lomo, T. Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. J. Physiol. (Lond.) 232, 331–356 (1973). Craik, F. I. M., Govoni, R., Naveh-Benjamin, M. & Anderson, N. D. The effects of divided attention on encoding and retrieval processes in human memory. J. Exp. Psychol. Gen. 125, 159–180 (1996). Buzsaki, G. The hippocampo-neocortical dialogue. Cereb. Cortex 6, 81–92 (1996). Baayen, R. H., Piepenbrock, R. & van Rijn, H. CELEX Lexical Database (Linguistic Data Consortium, University of Pennsylvania, Philadelphia, 1993). Jackson, G. D. & Duncan, J. S. MRI Neuroanatomy (Churchill Livingstone, New York, 1996). Daubechies, I. The wavelet transform, time-frequency localisation and signal analysis. IEEE Trans. Inform. Theory 36, 961–1005 (1990). Mardia, K. V. Probability and Mathematical Statistics: Statistics of Directional Data (Academic, London, 1972). Huynh, H. & Feldt, L. S. Estimation of the box correction for degrees of freedom from sample data in the randomized plot and split plot designs. J. Educ. Stat. 1, 69–82 (1976).



errata

Reversal of subjective temporal order due to arm crossing


Shinya Yamamoto and Shigeru Kitazawa Nat. Neurosci. 4, 759–765 (2001) Because of an error in proof corrections, a word was misprinted in the third line of the abstract. The correct abstract appears below.

How does the brain order successive events? Here we studied whether temporal order of two stimuli delivered in rapid succession, one to each hand, is determined before or after the stimuli are localized in space. When their arms were uncrossed, subjects could accurately report the temporal order, even when the interval between stimuli was as short as 70 ms. In most trials, subjects could also judge temporal order when their arms were crossed, but only if given adequate time (>1 s). At moderately short intervals (<300 ms), crossing the arms caused misreporting (that is, inverting) of the temporal order. Thus, at these intervals, the determining factor of temporal order was the spatial location of the hands. We suggest that it is not until the spatial locations of the hands are taken into account that the cutaneous signals from the respective hands are ordered in time.


1265

Nature Neuroscience December 2001

Recommend Documents