Characterization of macaque retinal ganglion cell responses using spike-triggered covariance

J W Pillow, E P Simoncelli and E J Chichilnisky

Published in Computational and Systems Neuroscience (CoSyNe), (109), Mar 2004.
Light responses of retinal ganglion cells (RGCs) exhibit several types of nonlinear behavior, including an accelerating contrast-response function, nonlinear pooling of spatial information, and dynamic gain adjustment. Such behaviors have been separately documented in a number of different experiments, each based on a different type of restricted stimuli. We describe a spike-triggered covariance (STC) analysis that aims to provide a complete characterization of RGC responses.

Multi-electrode recordings from macaque RGCs were obtained from isolated retinas stimulated with one-dimensional spatiotemporal white noise (i.e. flickering bars). Spike-triggered average (STA) analysis revealed center-sur-round spatial organization and biphasic temporal integration expected from RGCs. Eigenvector analysis of the STC revealed components with spatial and temporal structure distinct from the STA. Excitatory components (i.e. those associated with large eigenvalues) exhibited temporal structure similar to the STA, but finer spatial structure, consistent with receptive field subunits. Suppressive components exhibited spatial structure similar to the STA or to excitatory components, but temporal structure that was time-delayed relative to the STA, consistent with spike generation or contrast gain control nonlinearities.

The detailed spatial and temporal structure of these components was probed with a subspace-conditional analysis. STC components were examined in a subspace corresponding to either temporal or spatial stimulus features, obtained by filtering the stimulus with the spatial or temporal profile of the STA. The reduced stimulus dimensionality increased statistical power and revealed a larger number of significant STC components: multiple excitatory spatial components and one or two suppressive temporal components in most cells.

The STC components were then used to constrain a model of RGC light response. The model consisted of identical spatially shifted subunits whose outputs were combined nonlinearly, and a divisive temporal feedback signal. Subunit profiles were constrained to lie within the subspace spanned by spatial and temporal STC components, and the parameters of the nonlinear combination rule and feedback signal were fit using maximum likelihood. This model accurately reproduced the observed STA, contrast-response function, and detailed spatial and temporal structure of STC components. It also provided a substantially more accurate prediction of novel spike trains than a single-filter model characterized using only the STA.

  • Listing of all publications