Retinal receptive-field substructure: scaffolding for coding and computation

The center-surround receptive field of retinal ganglion cells represents a fundamental concept for how the retina processes and encodes visual information. Yet, traditional approaches of using the receptive field as a linear filter to integrate light intensity over space often do not capture the responses of a ganglion cell to complex visual stimuli. Thus, models with local nonlinearities in subunits of the receptive field or with local temporal dynamics are emerging to better reflect relevant aspects of retinal circuitry and capture stimulus encoding. Here, we review recent efforts to identify such receptive-field substructure and evaluate its role in visual stimulus encoding. The concomitant development of new computational tools may pave the way toward a model-based, functional approach to retinal circuit analysis. Visual stimulus encoding by the retina is not fully captured by the center-surround receptive fields of retinal ganglion cells. Mounting evidence of nonlinear spatial signal integration under natural stimuli and of specific visual functions solved by distinct ganglion cell types indicates the need to better understand receptive-field substructure.Nonlinear spatial integration can be captured by subdividing ganglion cell receptive fields into subunits, which are thought to correspond to presynaptic bipolar cells. Several statistical and model-based methods have recently been developed to identify the subunit layout from spiking responses of ganglion cells to visual stimuli.The subunits provide scaffolding for retinal computations, which may act through local adaptation and inhibition to shape responses to dynamic stimulation and to extract specific visual features.