Abstract
Purpose :
A central goal in visual neuroscience is to predict neural responses to complex visual inputs, especially those encountered during natural stimulation. Achieving this goal will require extending our descriptions of sensory computation, typically probed using controlled yet artificial visual stimuli, to the complex and dynamic conditions that characterize natural vision. For example, nonlinear integration across visual space has been demonstrated in many classes of retinal ganglion cells (RGCs), but it is unknown what impact this has on natural stimulus encoding. Here we test whether spatially nonlinear responses are elicited by natural visual stimuli and explore the consequences of nonlinear spatial integration for constructing models that predict RGC responses to natural inputs.
Methods :
We used extracellular and whole-cell patch recordings to measure spike and synaptic current responses of parasol (magnocellular-projecting) RGCs in an in vitro macaque retinal preparation. We used artificial visual stimuli, like spots and gratings, to characterize nonlinear RF structure as well as naturalistic visual stimuli, which included measured eye movements, to explore whether this nonlinear RF structure is important under physiological stimulation conditions. Finally, we constructed RF models to guide experiments and make predictions about how nonlinear RF structure impacts natural stimulus responses.
Results :
Experiments using artificial stimuli showed that both On and Off parasol RGCs show nonlinear integration at a small (sub-RF) spatial scale. However Off, but not On, parasols showed nonlinear responses to natural visual stimuli. Using measured excitatory subunit nonlinearities and RF modeling, we show that the nonlinear responses of Off parasols are due to the more sharply rectified excitatory subunits in the Off parasol RF, relative to those in the On parasol RF.
Conclusions :
We have shown that 1) nonlinear parasol responses to natural images are due to the presence of excitatory subunits within the RF; 2) the degree of rectification of subunit output determines whether a nonlinear RF can be approximated as spatially linear in the context of natural stimuli; and 3) accounting for excitatory nonlinear subunits can substantially improve models that predict neural responses to natural images.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.