{"title": "Retinogeniculate Development: The Role of Competition and Correlated Retinal Activity", "book": "Advances in Neural Information Processing Systems", "page_first": 91, "page_last": 97, "abstract": null, "full_text": "Retinogeniculate Development: \n\nThe Role of Competition and Correlated Retinal \n\nActivity \n\nRon Keesing* \nDept. of Physiology \nU.C. San Francisco \nSan Francisco, CA 94143 \nkeesing@phy.ucsf.edu \n\nDavid G. Stork \n\n*Ricoh California Research Center \n\n2882 Sand Hill Rd., Suite 115 \n\nMenlo Park, CA 94025 \n\nstork@crc.ricoh.com \n\nCarla J. Shatz \nDept. of Neurobiology \nStanford University \nStanford, CA \n94305 \n\nAbstract \n\nDuring visual development, projections from retinal ganglion cells \n(RGCs) to the lateral geniculate nucleus (LGN) in cat are refined to \nproduce ocular dominance layering and precise topographic mapping. \nNormal development depends upon activity in RGCs, suggesting a key \nrole for activity-dependent synaptic plasticity. Recent experiments on \nprenatal retina show that during early development, \"waves\" of activity \npass across RGCs (Meister, et aI., 1991). We provide the first \nsimulations to demonstrate that such retinal waves, in conjunction with \nHebbian synaptic competition and early arrival of contralateral axons, \ncan account for observed patterns of retinogeniculate projections in \nnormal and experimentally-treated animals. \n\n1 INTRODUCTION \nDuring the development of the mammalian visual system, initially diffuse axonal inputs \nare refined to produce the precise and orderly projections seen in the adult. In the lateral \ngeniculate nucleus (LGN) of the cat, projections arriving from retinal ganglion cells \n(RGCs) of both eyes are initially intermixed, and they gradually segregate before birth to \nform alternating layers containing axons from only one eye. At the same time, the \nbranching patterns of individual axons are refined, with increased growth in \ntopographically correct locations. Axonal segregation and refinement depends upon \n\n91 \n\n\f92 \n\nKeesing, Stork, and Schatz \n\npresynaptic activity - blocking such activity disrupts nonnal development (Sretavan, et \nal., 1988; Shatz & Stryker, 1988). These and fmdings in other vertebrates (Cline, et a1., \n1987) suggest that synaptic plasticity may be an essential factor in segregation and \nmodification of RGC axons (Shatz, 1990). \n\nPrevious models of visual development based on synaptic plasticity (Miller. et aI .\u2022 1989; \nWhitelaw & Cowan. 1981) required an assumption of spatial correlations in RGC activity \nfor nonnal development This assumption may have been justified for geniculocortical \ndevelopment. since much of this occurs postnatally: visual stimulation provides the \ncorrelations. Th~ assumption was more difficult to justify for retinogeniculate \ndevelopment. since this occurs prenatally - before any optical stimulation. \nThe first strong evidence for correlated activity before birth has recently emerged in the \nretinogenculate system: wave-like patterns of synchronized activity pass across the \nprenatal retina. generating correlations between neighboring cells' activity (Meister, et aI .\u2022 \n1991). We believe our model is the first to incorporate these important results. \n\nWe propose that during visual development. projections from both eyes compete to \ninnervate LGN neurons. Contralateral projections. which reach the LGN earlier. may \nhave a slight advantage in competing to innervate cells of the LGN located farthest from \nthe optic tract. Retinal waves of activity could reinforce this segregation and improve the \nprecision of topographic mapping by causing weight changes within the same eye -\nand \nparticularly within the same region of the same eye -\nto be highly correlated. Unlike \nsimilar models of cortical development. our model does not require lateral interactions \nbetween post-synaptic cells -\navailable evidence suggests that lateral inhibition is not \npresent during early development (Shotwell. et at, 1986). Our model also incorporates \naxon growth -\nan essential aspect of retinogeniculate development. since the growth and \nbranching ofaxons toward their ultimate targets occurs simultaneously with synaptic \ncompetition. Moreover. synaptic competition may provide cues for growth (Shatz & \nStryker, 1988). We consider the possibility that diffusing intracellular signals indicating \nlocal synaptic strength guide axon growth. \n\nBelow we present simulations which show that this model can account for development \nin nonnal and experimentally-treated animals. We also predict the outcomes of novel \nexperiments currently underway. \n2 SIMULATIONS \n\nAlthough the LGN is. of course, three-dimensional, in our model we represent just a \nsingle two-dimensional LGN slice, ten cells wide and eight cells high. The retina is then \none-dimensional: 50 cells long in our simulations. (This ratio of widths, 50/10, is \nroughly that found in the developing cat.) In order to eliminate edge effects, we \"wrap\" \nthe retina into a ring; likewise we wrap the LGN into a cylinder. \n\nDevelopment of projections to the LGN is modelled in the following way: projections \nfrom all fifty RGCs of the contralateral eye arrive at the base of the LGN before those of \nthe ipsilateral eye. A very rough topographic map is imposed, corresponding to coarse \ntopography which might be supplied by chemical gradients (Wolpert, 1978). \nDevelopment is then modelled as a series of growth steps, each separated by a period of \nHebb-style synaptic competition (Wig strom & Gustafson. 1985). During competition. \nsynapses are strengthened when pre- and post-synaptic activity are sufficiently correlated, \n\n\fRetinogeniculate Development: The Role of Competition and Correlated Retinal Activity \n\n93 \n\nand they are weakened otherwise. More specifically. for a given ROC cell i with activity \n~. the strength of synapse Wij to LON cell j is changed according to: \n\n~w,,=e(a, -a)(a.-~) \n\n1J \n\n1 \n\nJ \n\n[1] \n\nwhere (l and ~ are threshholds and e a learning rate. If a \"wave\" of retinal activity is \npresent. the activity of ROC cells is detennined as a probability of firing based on a \nOaussian function of the distance from the center of the wavefront. LON cell activity is \nequal to the sum of weighted inputs from ROC cells. \n\nMter each wave. the total synaptic weight supported by each ROC cell i is renormalized \nlinearly: \n\nw .,(t+l)= ~ \n\n1J \n\nw .. (t) \n\n1J \n\nk W, (t) \nk \n\n1k \n\n[2] \n\n[4] \n\nThe weights supported by each LON cell are also renonnalized. gradually driving them \ntoward some target value T: \n\nw .. (t+l)=w .. (t)+[T-Lw k,(t)] \n\nk \n\nJ \n\n1J \n\n1J \n\n[3] \n\nRenonnalization reflects the notion that there is a limited amount of synaptic weight \nwhich can be supported by any neuron. \nDuring growth steps, connections are modified based on the strength of neighboring \nsynapses from the same ROC cell. After normalization. connections grow or retract \naccording to: \n\nw .. (t+I)=w .. (t)+'Y L w\u00b7k(t) \n\n1J \n\n1J \n\nneighbors \n\n1 \n\nwhere 'Y is a constant term. Equation 4 shows that weights in areas of high synaptic \nstrength will increase more than those elsewhere. \n3 RESULTS \nSynaptic competition. in conjunction with waves of pre-synaptic activity and early arrival \nof contralateral axons. can account for pattens of growth seen in normal and \nexperimentally-treated animals. In the presence of synaptic competition. modelled axons \nfrom each eye segregate to occupy discrete layers of the LGN - precisely what is seen in \nnonnal development. In the absence of competition, as in treatment with the drug TTX. \naxons arborize throughout the LON (Figure I). \n\nThe segregation and refinement of retinal inputs to the LON is best illustrated by the \nfonnation of ocular dominance patterns and topographic ordering. In simulations of \nnormal development, where retinal waves are combined with early arrival of contalateral \ninputs, strong ocular dominance layers are formed: LON neurons farthest from the optic \ntract receive synaptic inputs solely from the contralateral eye and those closer receive only \nipsilateral inputs (Figure2, Competition). The development of these ocular dominance \npatterns is gradual: early in development, a majority of LON neurons receive inputs from \nboth eyes. When synaptic competition is eliminated, there is no segregation into eye(cid:173)\nspecific layers - LON neurons receive significant synaptic inputs from both eyes. These \nresults are consistent with labelling studies of cat development (Shatz & Stryker. 1988). \n\n\f94 \n\nKeesing, Stork, and Schatz \n\ncontralateral \n\nipsilateral \n\nc:::: \n\nQ .\u2022 -.\u2022 -8-e Q \n\nCJ \n\n= Q .\u2022 -Q= =8-e \n\n8 \n\nFigure 1: Retinogeniculate projections in vivo (adapted from Sretavan, et at, 1988.) \n(left), and simulation results (right). In the presence of competition (top), arbors are \nnarrow and spatially localized, confined to the appropriate ocular dominance layer. In the \nabsence of such competition (bottom), contralateral and ipsilateral projections are diffuse; \nthere is no discernible ocular dominance pattern. During simulations, projections are \nrepresented by synapses throughout the LGN slice, shown as squares; the particular \narborization patterns shown above are inferred from related simulations. \n\nCompetition \n\nNo Competition \n\nSimultaneous \n\n8 \n\n6 \n\n4 \n\n2 \n\n0 \n\n. . \n\n8 \n\n6 \n\n4 \n\n2 \n\n0 \n\n0 \n\n2 4 6 8 10 \n\n0 2 4 6 8 10 \n\n4 \n\n2IH-+-+-+-+-+-+-+-+-II \n\no 2 4 6 8 10 \n\nFigure 2: Ocular dominance at the end of development. Dark color indicates strongest \nsynapses from the contralateral eye, light indicates strongest synapses from ipsilateral, \nand gray indicates significant synapses from both eyes. In the presence of competition, \nLGN cells segregate into eye-specific layers, with the contralateral eye dominating cells \nwhich are farthest from the optic tract (base). When competition is eliminated (No \nCompetition), as in the addition of the drug TTX, there is no segregation into layers and \nLGN cells receive significant inputs from both eyes. These simulations reproduced \nresults from cat development. When inputs from both retinae arrive simultaneously \n(Simultaneous), ocular dominance \"patches\" are established, similar to those observed in \nnormal cortical development. \n\n\fRetinogeniculate Development: The Role of Competition and Correlated Retinal Activity \n\n95 \n\nRetinal waves cause the activity of neighboring ROCs to be highly correlated. When \ncombined with synaptic competition, these waves lead to a refinement of topographic \nordering of retinogeniculate projections. During development, the coarse topography \nimposed as ROC axons enter the LON is refined to produce an accurate, precise mapping \nof retinal inputs (Figure 3, Competition). Without competition, there is no refinement \nof topography, and the coarse initial mapping remains. \n\nCompetition \n\n\u2022 . . \n\n. . \u2022 . . \n\n. . . . . . . . . . ,. \n\n. \u2022 . \n\n. \u2022 \u2022 a [lO' I \u2022 I la \u2022 \u2022 . . . ,. \n\n. \u2022 a 0 cCCIl:J co\u00b7 \n. \n\n. \n\n. \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n. . . . . . . . . . . . .\u2022 aORb'\u00b0COI \u2022 . \n. . \n\u2022 \n\u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 a cOO I 00.\u00b7 . . . . . . . . \n\u2022 \u2022 \u2022 0 CCCDJco \u2022 \u2022 . . . . . . . . . . . . \n\n\u2022 D Ccl __ _ [J a -\n\n\u2022 -\n\n. \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022\u2022 \n\u00b7 . . . . . . . . . . . \u2022 \u2022 a C! I I [j a \u2022\u2022 \n\u2022 \u2022 \u2022 . . \n\u2022 \n. \u2022 a cCO I po a\u00b7 \u2022 . . . . . . . . . . \n\nI a cc[[]Jc [J a -\n. \n. \n\nto I 10 \u2022\u2022 . \u2022 \u2022 . \n\u00b7 \n\n. \u2022 . \n\n. \u2022 \n\n. \n\n. \n\n. \n\n. \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \u2022 ooC \n\nNo Competition \n\n.\u2022..\u2022 a\" \n\naC- a\u00b7 a a a\u00b7 \u2022 aC\u00b7 . -\nCa. \u2022 \u00b7\u00b7.c\u00b7 .. a\u00b7 . . \na\u00b7 \n. a \u00b7 \u00b7 \u00b7 \u00b7 \u00b7 \u00b7 \u00b7 \n. . . . a\" \n\u2022 . . 'C'\" a - \u00b7 \u00b7 -C\u00b7aaC\u00b7aC\u00b7 aa\u00b7 a-\u00b7 \u00b7 a . \u00b7 a\u00b7 \u00b7 a \u00b7 a \u00b7 \u00b7 \u00b7 \u00b7 \u00b7 \u00b7 . \u00b7 \u00b7 \u00b7 \u00b7 \u00b7 \n\u00b7 . -\n. -\n\u00b7 .\u2022\u2022\u2022\u2022 \u2022 .. C\u00b7 . C\u00b7 \u2022 \u2022 \u2022\u2022 , . . . . . . . \u2022 . . \u2022 . \n\u00b7 .Ca. a a\u00b7 . D D\u00b7 a \u00b7C \u2022\u2022.... a\u00b7\u00b7 ... a \u2022. C\u00b7 \n\u2022 a a C . CC\u00b7 . \u2022 Ca. . a . . . . . . C . . 0 \n\n. - .\u2022 a\u00b7 . - . . aC\u00b7 - a 0 a\u00b7 0 \u2022 \u2022 CaCa 0 aaCC- aC \n\u2022 a\u00b7 . \u2022 . 0 a \u00b7 . \u00b7 a\u00b7 . c\u00b7 c\u00b7 .\u2022. a\u00b7 aC. c. a a a\u00b7 a \n\u00b7C.\u00b7\u00b7. \u00b7C. a\u00b7\u00b7.\u00b7 a\u00b7 aD. a\u00b7\u00b7\u00b7\u00b7 aaa\u00b7\u00b7 \u00b7C\u00b7\u00b7 \n. a' \n.\u2022. \n. - . . . . . . . . . . . . \n,. . \u2022 . . . . \u2022 ,. .. \u2022 ,. \n....\u2022 a \u2022... a \u2022\u2022. a\u00b7 \n\u2022 C D D . C a \n\n. a\u00b7 .. - a\u00b7 . a a -[JCa. - a . - C\u00b7 - . . . - -\n\n.\u2022 C. \n\u00b7C\u00b7\u00b7\u00b7\u00b7 D' \u2022\u2022 a.Ca \u00b7C\u00b7 .CD\u00b7 'C' 'Co aCaC\u00b7\u00b7\u00b7\u00b7 a\" \n\n.\u2022 aa\u00b7\u00b7 a' aaC\" a\u00b7. a\u00b7CaCa\u00b7 \n\n\u2022 Ca\u00b7 D. . \u2022 . . . . a Ca. 0 \n\n. \n\n\u2022 \n\nFigure 3: Topographic mapping with and without competition. The vertical axis \nrepresents ten LON cells within one section, and the horizontal axis 50 ROC cells. The \nsize of each box indicates strength of the synapse connecting corresponding ROC and \nLON cells. If the system is topographically ordered, this connection matrix should \ncontain only connections forming a diagonal from lower left to upper right, as is found in \nnormal development in our model (Competition). When competition is eliminated, the \ntopographic map is coarse and non-contiguous. \n\n4 PREDICTIONS \nIn addition to replicating current experimental findings, our model makes several \ninteresting predictions about the outcome of novel experiments. If inputs from each eye \narrive simultaneously, so that contralateral projections have no advantage in competing to \ninnervate specific regions of the LON, synaptic competition and retinal waves lead to a \npattern of ocular dominance \"patches\" similar to that observed in visual cortex (Figure 2, \nSimultaneous). Topography is refined, but in this case a continuous map is formed \nbetween the two eyes (Figure 4) -\n\nagain similar to patterns observed in visual cortex. \n\n\f96 \n\nKeesing, Stork, and Schatz \n\n.. , .... \n. . -. . -......... . . \n\n\u2022 \u2022 . \u2022 . \u2022 ' \u2022 - DQ.J 10 [JCD \u2022 . \n\n. ..\u2022 \n\n. \n\n. \n1.1 , , LiiP -Ii ~ ~ : \n\n. . . . . . . . . . . . . . \n\n\u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 D CD , , , DC D \u2022 \n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \n\n\u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \n\u2022 \u2022 \u2022 \u2022 \u2022 _. _ ...... \"r\"T\"\"T\"\"'1.:.. ,.:,11 CU I I I DC II \u2022 \n. ... \n\n~....,_II.' \n\nII \n\n\u2022 \n\n\u2022 \u2022 \u2022 \u2022 \u2022 \u2022 \n\n. . .. , \n..\u2022 -CD \n. . . .. . ..... \n\nFigure 4: Topographic mapping with synchronous arrival of projections from both eyes. \nLight boxes represent contralateral inputs, dark boxes represent ipsilateral. Synaptic \ncompetition and retinal waves cause ocular segregation and topo:sraphic refinement, but in \nthis case the continuous map is formed using both eyes rather than a single eye. \nthe distribution of activity around \nOur model predicts that the width of retinal waves -\nis an essential factor in determining both the rate of ocular \nthe moving wavefront -\nsegregation and topographic refinement. Wide waves, which cause many RGes within \nthe same eye to be active, will lead to most rapid ocular segregation as a result of \ncompetition. However, wide waves can lead to poor topography: RGes in distant \nregions of the retina are just as likely to be simultaneously active as neighboring RGes \n(Figure 5). \n\nSegregation \n\n100 \n\n-\n\n(IIJ \n\n(IIJ \n\n~QC \n\n60 \n\n-; ~ 80 \nBQ \n.... \n0 \n--\"0 \ntil \nu .... \n.... ~ \n(IIJ \no QI \n... \n~~ \n\n20 \n\n40 \n\nQI QI \n\nrJ:J \n\n5 \n\n0 \n\n4 \n\n... \n... \n... \n3 ~ \nu QC \n.= (IIJ \n-~ \n2 =-Q \n....... \n(IIJ \n\n~(IIJ \n0 \n\n1 =-\n0 \nE-\n\nO \n\n0 \n0.0 0.2 0.4 0.6 0.8 1.0 \n\nAverage Activity \nin \nNeighboring RGCs \n\nFigure 5: The width of retinal \"waves\" determines ocular dominance and topography in \nnormal development in our model. Width of retinal waves is represented by the average \nactivity in RGC cells adjacent to the Gaussian wavefront: high activity indicates wide \nwaves. Topographic error (scale at right) represents the average distance from an RGCs \ntarget position multiplied by the strength of the synaptic connection. LGN cells are \nconsidered ocularly segregated when they receive .9 or more of their total synaptic input \nfrom one eye. Wide waves lead to rapid ocular segregation - many RGes within the \nsame retina are simulaneously active. An intermediate width, however, leads to lower \ntopographic error - wide waves cause spurious correlations, while narrow waves don't \nprovide enough information about neighboring RGCs to significantly refine topography. \n\n\fRetinogeniculate Development: The Role of Competition and Correlated Retinal Activity \n\n97 \n\n5 SUMMARY \nOur biological model differs from more developed models of cortical development in its \ninclusion of 1) differences in the time of arrival of RGC axons from the two eyes, 2) lack \nof intra-target (LGN) inhibitory connections, 3) absence of visual stimulation, and 4) \ninclusion of a growth rule. The model can account for the development of topography \nand ocular dominance layering in studies of normal and experimental-treated cats, and \nmakes predictions concerning the role of retinal waves in both segregation and \ntopography. These neurobiological experiments are currently underway. \n\nAcknowledgements \n\nThanks to Michael Stryker for helpful suggestions and to Steven Lisberger for his \ngenerous support of this work. \n\nReferences \n\nCline, H.T., Debski, E.A., & Constantine-Paton, M .. (1987) \"N-methyl-D-aspartate \nreceptor antagonist desegregates eye-specific stripes.\" PNAS 84: 4342-4345. \n\nMeister, M., Wong, R., Baylor, D., & Shatz, C. (1991) \"Synchronous Bursts of Action \nPotentials in Ganglion Cells of the Developing Mammalian Retina.\" Science. 252: 939-\n943. \n\nMiller, K.D., Keller, J.B., & Stryker, M.P. (1989) \"Ocular Dominance Column \nDevelopment: Analysis and Simulation.\" Science. 245: 605-615. \n\nShatz, C.J. (1990) \"Competitive Interactions between Retinal Ganglion Cells during \nPrenatal Development.\" 1. Neurobio. 21(1): 197-211. \n\nShatz, C.J., & Stryker, M.P. (1988) \"Prenatal Tetrodotoxin Infusion Blocks Segregation \nof Retinogeniculation Afferents.\" Science. 242: 87-89. \n\nShotwell, S.L., Shatz, C.J., & Luskin, M.B. (1986) \"Development of Glutamic Acid \nDecarboxylase Immunoreactivity in the cat's lateral geniculate nucleus.\" 1. Neurosci. 6(5) \n1410-1423. \n\nSretavan, D.W., Shatz, C.J., & Stryker, M.P. (1988) \"Modification of Retinal Ganglion \nCell Morphology by Prenatal Infusion of Tetrodotoxin.\" Nature. 336: 468-471. \n\nWhitelaw, V.A., & Cowan, J.D. (1981) \"Specificity and plasticity of retinotectal \nconnections: a computational model.\" 1. Neurosci. 1(12) 1369-1387. \n\nWigstrom, H., & Gustafsson, B. (1985) \"Presynaptic and postsynaptic interactions in the \ncontrol of hippocampal long-term potentiation.\" in P.W. Landfield & S.A. Deadwyler \n(Eds.) Longer-term potentiation: from biophysics to behavior (pp. 73-107). New York: \nAlan R. Liss. \n\nWolpert, L. (1978) \"Pattern Formation in Biological Development.\" Sci. Amer. 239(4): \n154-164. \n\n\f\fPART II \n\nNEURO-DYNAMI CS \n\n\f\f", "award": [], "sourceid": 565, "authors": [{"given_name": "Ron", "family_name": "Keesing", "institution": null}, {"given_name": "David", "family_name": "Stork", "institution": null}, {"given_name": "Carla", "family_name": "Shatz", "institution": null}]}