Up
Previous
Next

Pooling of rod inputs and limits to sensitivity in primate retinal ganglion cells

Valerie Uzzell,1 Greg Field,2 Fred Rieke,2 and E.J. Chichilnisky1

1Salk Institute and UCSD
2University of Washington


Dark-adapted humans can detect flashes producing only 5-20 photoisomerizations across a pool of hundreds or thousands of rod photoreceptors. How can the visual system detect whether and when a sparse signal occurs in an array of many noisy detectors? We examined this question by comparing the detection and temporal sensitivity of retinal ganglion cells (RGCs) to that of their pooled rod inputs.

Primate rod and RGC detection and temporal sensitivity were characterized using a two-alternative forced choice linear discrimination of responses to dim flashes delivered at different times. Responses from isolated rods were obtained using suction electrode recordings. Multielectrode extracellular recordings from RGCs were obtained from isolated retinas maintained in vitro. Detection and temporal sensitivities of RGCs were compared to those of their linearly pooled rod inputs. For this comparison, simulated rod signals were created whose signal and noise characteristics were matched to real rod signals. These simulated rod signals were then weighted by a RGC receptive field and summed. A linear discriminant was applied to the time course of the summed rod signal to determine whether and when a flash occurred.

Most ON cells showed detection sensitivity 1-3 times lower than the physical limit imposed by spontaneous isomerizations of rhodopsin, indicating that rod signals may be processed nearly optimally in the primate retina. ON cell detection sensitivity was 2-4 times higher than predicted from linear pooling of rod inputs, indicating that the observed sensitivity requires nonlinear processing of rod signals. ON cell temporal sensitivity ranged from 20-50 ms, finer than the rod integration time of 200 ms. OFF cell detection and temporal sensitivity were lower. These results show that primate RGCs efficiently detect and discriminate the timing of sparse signals in thousands of noisy rod inputs.