A fundamental issue in our understanding of the neural code of the retina is the reproducibility of retinal spike trains to repeated presentations of the same visual stimulus. If single spikes are deterministically elicited by the stimulus, the brain can attach individual visual messages to each spike; whereas if spike trains are stochastic, then the brain must average over many spikes to obtain a clear visual message. Here, we study the vertebrate retina, stimulating an isolated retina with a computer monitor and recording action potentials extracellularly with a multi-electrode array.
Our data indicate that for some classes of stimuli, such as random flicker with a broad frequency spectrum, retinal spike trains are highly deterministic with precisely reproduced spike trains on repeated trials. Single spikes occur with a temporal precision of as little as 5 msec, while the average spacing between spikes is 300 to 500 msec. Furthermore, spikes occur in discrete firing "events", and the variance in the number of spikes in an event over repeated stimulus presentations differs significantly from Poisson statistics. These results imply that the timing of single action potentials can convey discrete information about the visual stimulus, in contrast to the prevailing view that information is carried by a neuron's average firing rate.