Responses of Different Retinal Ganglion Cell Types to Motion of 'Natural' Scenes

Scenes from nature were projected onto the mammalian retina to simulate motion: saccades, optic flow, object motion, and fixational eye movements. We studied 7 ganglion cell types: ON and OFF brisk–transient, ON and OFF brisk–sustained, ON and ON–OFF direction–selective (DS), and local–edge. Peak rate was the highest 5 ms time bin in the PSTH; firing fraction was the fraction of time bins during which a cell fired ≥ 5% of its peak. Temporal precision was measured by spike time jitter across stimulus repeats. Spike–count distributions were computed with sliding windows of 100–500ms. Information rates were estimated by the direct method. Efficiency was information rate normalized by the maximum possible information at the average spike rate. Results are compared by independent t–test.


ON and OFF brisk cells were similar and thus pooled. For all types of motion brisk–transient and brisk–sustained cells had the highest average spike rates (p<0.05). Brisk–transient cells had the highest peak rates (p<0.005), and local–edge cells had the lowest (p<0.01 except ON DS). Responses were sparse: most cells had firing fractions <0.5. Brisk–sustained cells had the highest fractions (p<0.05 except local–edge). Brisk–transient and ON–OFF DS types were more precise than brisk–sustained cells (p<0.005), and local–edge cells were by far the least precise (p<0.005). Spike–count distributions were approximately exponential and highly skewed toward zero spikes per time window. However, cells with the highest average spike rates had less skewed, dome–shaped distributions, with peaks greater than zero. Brisk cells showed the highest information rates (p<0.05). Yet for each type and stimulus, information rate was about 30% of the maximum that the cell could transmit at its average spike rate; efficiency for all cells was ∼ 0.30.