Understanding how the retina responds to natural visual stimuli is a daunting challenge, both because the statistics of natural scenes are not fully known [1,2] and because the output of the retina can be highly non-linear [3] and history-dependent [4] even when driven by simple stimuli. Here, we focus on assessing the variety of spiking patterns produced by retinal ganglion cells in response to repeated presentations of natural scenes.

Ganglion cells from isolated retinae of larval tiger salamanders and rabbits were recorded extracellularly with a multi-electrode array. Natural movies were videotaped with a Camcorder, digitized to 8-bit resolution, displayed using a Macintosh computer and monitor, and projected onto the retina. Movies were taken of woodland and river scenes in the daytime with the camera mounted on a tripod and occasionally panned slowly. As a result, image motion was either slow or intrinsic to the scene. This is a reasonable approximation for the salamander, an animal that moves infrequently and slowly, but is not as appropriate for the rabbit. Each movie clip lasted 30 sec and was repeated 60 times.

Under stimulation from three different natural movies, ganglion cells in both the salamander and rabbit responded with discrete episodes of firing separated by intervals of complete silence. These firing events had a rapid onset - the temporal jitter of the first spike of an event was typically 10 to 20 msec - and reached peak firing rates of ~100 Hz. Events contained up to 15 spikes with high precision - the variance in the number of spikes per trial was significantly below the average spike number. Furthermore, ganglion cells were silent the vast majority of the time - the post-stimulus time histogram fell below a criterion response (5% of its peak firing rate) for ~0.85 of the total stimulus time in the salamander.

These patterns of firing shared many qualitative similarities with those found under random flicker stimulation, a simpler stimulus ensemble [5]. A notable difference, however, was the greater variety of timecourses for firing events under natural stimulation. Maybe one can tailor these synthetic stimuli to match various statistical properties of Nature and reproduce the observed variety of firing patterns. Such stimuli would provide a useful testbed for the stimulus-response relationship under natural conditions.

[1] D. J. Field, Neural Computation 6, 559-601 (1994).
[2] D. L. Ruderman and W. Bialek, Phys. Rev. Lett. 73, 814-817 (1994).
[3] R. M. Shapley and J. D. Victor, J. Physiol. 285, 275-298 (1978).
[4] S. M. Smirnakis, M. J. Berry, D. K. Warland, W. Bialek, and M. Meister, Nature 386, 69-73 (1997).
[5] M.J. Berry, D.K. Warland, and M. Meister, PNAS 94, 5411-5146 (1997).