Simulating the effects of temporal correlation between signals of primate ganglion cells

A physiologically plausible neural net, simulating ganglion cell signals in the primate retina was shown earlier to support hyperacuity performance in humans (Wachtler et al. 1996, J. computational Neuroscience 3:73-82). In this model the signals of ganglion cells are assumed to be statistically independent, except for stimulus induced components, even in immediately adjacent cells. Ganglion cell signals in the retina of salamanders show temporal correlations depending on mutual distance (Meister et al. 1995, Science 270:1207-1210). Moreover, by using the method of reverse correlation, receptive fields (RFs) can be characterized not only due to spikes measured in individual cells but also due to correlated spikes in adjacent cells. Remarkably the size of RFs determined the latter way was much smaller than the linear superposition of the original RFs. This implies a mechanism which can transmit additional information using a given array of cells by picking up correlated signals. We attempted to analyze the consequences of such a mechanism for correlations possibly present in the primate retina. The model proposed by Wachtler et al. was modified in several respects to allow the introduction of temporal correlations between the spikes fired by ganglion cells. As reported for the salamander retina, correlation strength is assumed to decrease with increasing distance. In analogy to the experiments of Meister et al. performance was analyzed with the method of reversed correlation. When correlation is symmetric, ganglion cells are coupled not only at the local but also at the global level. Spontaneaous and induced activity in primate ganglion cells found in the primate retina is typically about 20 times higher than that in salamanders. With these activity levels and for coupling strengths as high as those in salamanders the network rapidly becomes unstable. Decreasing coupling strength diminishes instability. For very low coupling strengths the network is stable and a slight increase in signal/noise ratio is found compared to the non-coupled condition. The RFs constructed from the correlated signals of adjacent cells were found to correspond to the linear superpositions of the RFs of the two original cells. In addition, the performance of the coupled network does not differ from the non-coupled network with respect to the support of hyperacuity. In summary, due to the vastly different activity levels in retinal ganglion cells of salamanders and primates, no direct generalization between these species seems possible with respect to the use of temporal correlations to improve spatial resolution