Tom P. Franken¹, Fatemeh Didehvar¹; ¹Washington University in St. Louis

The lightness of a surface is strongly influenced by scene context. Prior studies found that individual neurons in early visual cortex can exhibit context effects similar to perception, but such effects are often heterogenous. It remains therefore poorly understood how context affects neural representation at the population level. To tackle this question, we used Neuropixels to record from columnar populations of neurons in visual cortical Area V4 in a macaque. During these recordings the animal fixated on scenes tiled into adjacent surface elements. The classical receptive fields (cRFs) of the neurons were centered within one element of fixed luminance. Random luminance noise was independently added to the other elements, and we presented >10000 such unique scenes per session (FixedLocalLum). In the same sessions we also recorded neural activity to scenes in which the luminance in the cRF was varied (VarLocalLum). To reveal context effects on the representation of luminance at the population level, we developed a between-condition decoding approach. We trained random forest decoders to decode the luminance in the cRFs, from spike counts of all neurons in a session to the VarLocalLum scenes. We then tested these decoders on the FixedLocalLum data. For the 1000 FixedLocalLum scenes with respectively the highest and the lowest decoded luminances, we averaged the pixel value element-wise and subtracted these averages. This revealed the contextual elements that significantly influenced the decoded luminance from a column (permutation test p<0.05). We find that these elements often form non-uniform patterns, similar to reverse correlation patterns that we identified in individual units. Our data suggest net, non-uniform context effects on the representation of local luminance in columnar populations of V4 neurons. More broadly, our method represents a novel approach based on reverse correlation that can be deployed to study context effects in large simultaneously recorded populations.

Acknowledgements: This research was supported by the National Eye Institute of the National Institutes of Health under Award Numbers R00EY031795 and R21EY036566, and the Small Grants Program from the McDonnell Center for Systems Neuroscience at Washington University in St. Louis..