Synaptotagmin: A Calcium Sensor on the Synaptic Vesicle Surface Author(s): Nils Brose, Alexander G. Petrenko, Thomas C. Südhof and Reinhard Jahn Source: Science, New Series, Vol. 256, No. 5059 (May 15, 1992), pp. 1021-1025 Published by: American Association for the Advancement of Science Stable URL: http://www.jstor.org/stable/2877129 Accessed: 01-10-2015 19:18 UTC Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. American Association for the Advancement of Science is collaborating with JSTOR to digitize, preserve and extend access to Science. http://www.jstor.org This content downloaded from 132.239.70.252 on Thu, 01 Oct 2015 19:18:21 UTC All use subject to JSTOR Terms and Conditions . REPORTS that does not require calculation of derivatives is placed by two dots (situated at the endpoints). Weinshall, Biol. Cybern. 64, 209 (1991). to look for random changes (controlled in appro- The network learned this task, as it did previously 19. S. Edelman, D. Reisfeld, Y. Yeshurun, CS-TR priate ways) in the parameter values that reduce in the line vernier and the bisection cases. The 91-20 (Department of Applied Mathematics and the error. In the simulations described in this better performance of the HyperBF module in the Computer Science, Weizmann Institute, Rehovot, report the model was endowed with a dual, incre- dot vernier task for small offsets parallels a recent Israel, 1991); R. Brunelli and T. Poggio, in Pro- mental learning mechanism. First, when the mod- surprising finding with human subjects (M. Fahle, ceedings of the 12th International Joint Confer- el's performance on a new input was markedly unpublished observations). ence on Artificial Intelligence, International Joint inadequate (in comparison with recent history), 13. In a recent study, R. Bennett and G. Westheimer Conference on ArtificialIntelligence, Inc., Sydney, that input was adjoined to the model as an addi- [Percept. Psychophys. 49, 541 (1991)] found sur- Australia, 24 to 30 August 1991 (Kaufmann, tional center (prototype). This happened mainly in prisingly little learning of thresholds in three-dot Mountain View, CA, 1991); S. Edelman and T. the initial trials, with the number of centers even- alignment and grating discrimination. Their exper- Poggio, A.I. Memo No. 1181 (ArtificialIntelligence tually reaching an asymptote that depended on iments used transfer of training across the stimu- Laboratory, Massachusetts Institute of Technolo- the nature of the task and on the parameters that lus range to probe for learning, hiding possible gy, Cambridge, MA, 1990); R. Brunelli and T. affected the decision to add new centers. The effects of fast learning that may have happened in Poggio, I.R.S. T. TechReport 9110-04 (Instituto per performance of the model during these first trials the baseline session (p. 544). Interestingly, the la Ricerca Scientifica e Tecnologica, Trento, Italy, improved quickly, then stabilized as the number lack of transfer across the stimulus range in these 1991). of centers approached the asymptote. Second, experiments is consistent with our notion of expe- 20. We are grateful to H. Buelthoff, F. Crick, F. Girosi, further gradual improvement in the performance rience-based learning. R. Held, A. Hurlbert,Y. Weiss, and G. Westheimer was obtained by letting the model carry out a local 14. A. Fiorentini and N. Berardi, Nature 287, 43 for useful discussions and suggestions. Support- random search in the space of existing HyperBF (1980). ed by a grant from the Office of Naval Research, center coordinates. This search was guided by 15. A. Karniand D. Sagi, Proc. Natl. Acad. Sci. U.S.A. Cognitive and Neural Sciences Division, by the feedback given to the model (that is, by indicating 88, 4966 (1991). Artificial Intelligence Center of Hughes Aircraft whether the response at each trial was correct). 16. K. Ball and R. Sekuler, Science 218, 697 (1982); Corporation, and by the Deutsche Forschungsge- Details of the learning algorithms, including an S. P. McKee and G. Westheimer, Percept. Psy- meinschaft (Heisenberg-Programme). Support for extension of the incremental learning algorithm to chophys. 24, 258 (1978); V. S. Ramachandran the Artificial Intelligence Laboratory's artificial in- a situation in which no explicit feedback is avail- and 0. Braddick, Perception 2, 371 (1973). telligence research is provided by the Advanced able, can be found in (6, 7). 17. Y. Fr6gnac, D. Shulz, S. Thorpe, E. Bienenstock, Research Projects Agency of the Department of 5. A description of multilayer perceptrons and the Nature 333, 367 (1988). Defense. T.P. is supported by the Uncas and back-propagation technique used for learning is 18. T. Poggio and S. Edelman, ibid. 343, 263 (1990); Helen Whitaker chair. in D. E. Rumelhart, G. E. Hinton, R. J. Williams, S. Edelman and T. Poggio, Int. J. Pattern Recog- Nature 323, 533 (1986). An overview of some of nit. Artif. Intell., in press; S. Edelman and D. 18 December 1991; accepted 24 March 1991 the classical techniques can be found in S. Omo- hundro [Complex Syst. 1, 273 (1987)] and in R. 0. Duda and P. E. Hart [Pattern Classification and Scene Analysis (Wiley, New York, 1973)]. Rela- tions between multilayer perceptrons and Hy- Synaptotagmin: A Calcium Sensor on the perBF networks are mentioned in (4) and studied in M. Maruyama, F. Girosi, T. Poggio, A.l. Memo Synaptic Vesicle Surface No. 1291 (Artificial Intelligence Laboratory, Mas- sachusetts Institute of Technology, Cambridge, MA, 1992). Nils Brose,* Alexander G. Petrenko, Thomas C. Sudhof, 6. T. Poggio, M. Fahle, S. Edelman, A.l. Memo No. ReinhardJahntt 1271 (Artificial Intelligence Laboratory, Massa- chusetts Institute of Technology, Cambridge, MA, Neurons release neurotransmitters by calcium-dependent exocytosis of synaptic 1991); S. Edelman, T. Poggio, M. Fahle, Comput. vesicles. Vision Graph. Image Process. B, in press. The However, the molecular steps transducing the calcium signal into membrane fusion are still simulation results were robust with respect to all an enigma. It is reported here that synaptotagmin, a highly conserved synaptic vesicle parameters, including the number of inputs. protein, binds calcium at physiological concentrations in a complex with negatively charged 7. Y. Weiss, S. Edelman, M. Fahle, T. Poggio, CS-TR 91-21 (Department of Applied Mathematics and phospholipids. This binding is specific for calcium and involves the cytoplasmic domain of Computer Science, Weizmann Institute, Rehovot, synaptotagmin. Calcium binding is dependent on the intact oligomeric structure of syn- Israel, 1991). aptotagmin (it is abolished by proteclytic cleavage at a single site). These results suggest 8. We have also experimented with a different ver- that synaptotagmin sion of the HyperBF model, in which orientation- acts as a cooperative calcium receptor in exocytosis. selective receptive fields similar to those of simple cells in Vi played the role of the basis functions. See (7). This version of the model replicated the absolute values and the time course of the im- provement of the thresholds found in human psy- Calcium-dependentexocytosis of synaptic entryand the releaseof transmitteris in the chophysical data, in addition to replicating the vesicles is the central step in the sequence range of 200 ,us. This implies that a com- data concerning the percentage of correct re- of events from the arrival of an action plex betweensynaptic vesicles and the plas- sponses. 9. Hyperacuity-level performance was independent potential to the release of neurotransmit- ma membranemust exist in the restingstate of the precise location of the receptors. At the ters. It is generally accepted that Ca2+ because the time after Ca2+ entry is too same time, different quasi-random receptor mo- enters the nerve terminalvia voltage-gated short to allow for vesicle docking before saics yielded different thresholds, sometimes by Ca2+ as much as a factor of 2. A similar range of channels in the presynapticplasma fusion. Furthermore,the dependence of hyperacuity thresholds is observed in human sub- membrane.Intracellular recordings in mod- transmitterrelease on the intraterminal jects, even at full acuity and with perfectly normal el synapsessuch as the squidgiant synapse Ca2+ concentrationis nonlinearand highly eyes. have shown that the latencybetween Ca2+ cooperative(1). 10. The model also exhibited learning on a longer time scale (4, 7), similar to the slow long-term The Ca2+ receptorprotein for exocytosis learning component found in human subjects (M. N. Brose and R. Jahn, Department of Neurochemistry, has not been identified. However, certain Max-Planck-Institute for Fahle and S. Edelman, in preparation). Psychiatry, D-8033 Martins- predictions about its properties can be 11. R. Watt and F. W. Campbell, Spat. Vision 1, 31 ried, Germany. A. G. Petrenko and T. C. SOdhof, Howard Hughes (1985). made. Becauseof the shortlatency between Medical Institute and Department of Molecular Genet- 12. The stimulus in the bisection task consists of three Ca2+ influx and exocytosis,it is likely that ics, University of Texas Southwestern Medical Center, dots, arranged in a vertical line, at an approxi- Dallas, TX 75235. the Ca2+ receptoris part of the complex mately even spacing. The subject has to deter- formed mine whether the middle dot is above or below the *Present address: Salk Institute, Molecular Neurobiol- between the plasmamembrane and midpoint of the segment formed by the other two ogy Laboratory, 10010 North Torrey Pines Road, La the synapticvesicle and is probablylocated dots. The HyperBF module learned this hypera- Jolla, CA 92037. on one of these membranecompartments. cuitytask just as easily as it did in the linevernier tPresent address: Howard Hughes Medical Institute In addition,Ca2+ must induce a change in case (6).