Temporal pattern recognition in retinal ganglion cells is mediated by dynamical inhibitory synapses

The efficient coding theory postulates that sensory neurons compress relevant information before sending it to the rest of the brain. To this end, these neurons are thought not to signal all input features, but only surprising events, e.g. mismatches between expected and perceived inputs. In the retina, it has been shown that ganglion cells, the retinal output, strongly respond when a periodic sequence of flashes suddenly ends. Strikingly, the latency of this response shifts by the same amount as the period of the flash sequence, as if ganglion cells responded to the omitted flash with constant latency. This suggests that the retina has a temporal expectation of when the omitted stimulus should have occurred, and is able to signal the violation of this expectation. The mechanism behind this predictive latency shift remains unclear. Several models have been proposed, but have not been corroborated by data.
Here we combined modeling and experiments to show that this Omitted Stimulus Response is due to inhibitory neurons with depressing synapses. When blocking glycinergic inhibitory transmission in the retina, we found that the response remained, but the predictive latency shift was lost. Based on this result, we developed a model of the retinal circuit where the response to the omitted stimulus arises from an interplay between excitatory and delayed inhibitory inputs. Inhibition delays the response at the end of the sequence, while the depressing synapse is necessary to shift this delay as a function of the frequency of the flash sequence. Our circuit model reproduced our experimental findings and generated new predictions that we confirmed by further experiments. Since depressing inhibitory synapses are ubiquitous in sensory circuits, our results suggest they could be a key component to generate predictive responses that have been observed in several brain areas.