Information exchange between pairs of spike trains in the mammalian visual system

We have studied the neural firing patterns of retinal ganglion cells (RGCs) and their target lateral geniculate nucleus (LGN) cells. Reliable information transmission coexists with fractal fluctuations which appear in RGC and LGN firing patterns. Unexpectedly, these fluctuations appear not to be independent across LGN cells, information is also shared among pairs of LGN spike trains. Over short time scales, we find that clusters of spikes in the RGC neural firing pattern appear at the LGN output essentially unchanged, while isolated RGC firing events are more likely to be eliminated; thus the LGN action-potential sequence is :not simply a randomly deleted version of the RGC spike train. Employing information-theoretic techniques, we estimate the information efficiency of the LGN neuronal output - the proportion of the variation in the LGN firing pattern that carries information about its associated RGC input - to be in the vicinity of 50% over counting windows below about 10 ms. We develop a new information-theoretic measure which helps determine at what time scale a neural spike train changes from a time code to a rate code.