Retinal adaptive mechanisms confer selectivity to homogeneous objects in natural scenes

Adaptive mechanisms in sensory neurons are crucial to transmit information in different contexts. In the retina, it is assumed that their role is to normalize neuronal responses to input statistics like mean and variance. However, this role has mostly been characterized with simple, artificial stimuli, and remains unclear for natural stimuli. Here we show that during their response to natural scenes, adaptive mechanisms reshape the feature selectivity of ganglion cells, the retinal output. We recorded retinal ganglion cell responses to rapid sequences of natural images in mice. Including a bio-inspired adaptive mechanism in an artificial neural network model was necessary to predict cell responses to new sequences of natural images. This adaptive mechanism tuned specific cell types to selectively respond to homogeneous regions situated within cluttered visual surrounds, a feature suited for detecting threats. Adaptive mechanisms do not merely normalize responses but actively enable new feature selectivity in the early visual system.