Abstract
Neurons in primary visual cortex (V1) show contrast-dependent spatial frequency tuning: at low contrasts, they tend to prefer lower spatial frequencies. A linear spatial filter followed by a nonlinearity cannot explain this behavior, unless that nonlinearity is spatial frequency dependent. One well known nonlinearity in cortical cells is contrast gain control. Here, we show that spatial frequency tuned contrast gain control can explain the contrast-dependent tuning of V1 neurons. We measured responses to optimally-oriented sinusoidal gratings and grating mixtures presented at a range of contrasts to well-isolated single neurons recorded from opiate-anesthetized paralyzed macaque monkeys. The responses to mixtures helped us to dissociate the linear and non-linear components of the response. We fit individual cell responses with a model consisting of a linear spatial filter whose output is divided by a gain control signal computed from the weighted sum of the squared responses of a pool of similar filters tuned to a range of spatial frequencies, plus a small adjustable parameter. We compared the performance of the model with and without spatial frequency dependent pooling. For most cells, spatial frequency dependent pooling improves the fits; for 25% of cells, the improvement is significant. The tuned gain control often captures the contrast dependence of spatial frequency tuning, including shifts towards lower preferred frequency at lower contrasts. This is achieved by preferentially weighting the responses of filters with frequencies below the preferred spatial frequency of the cell. This demonstrates that a tuned normalization can account for contrast-dependent changes in spatial frequency tuning.
Acknowledgement: Simons Foundation, NIH, HHMI