Mechanistic basis of dynamic and heterogeneous divisive normalization in visual cortex

Abstract

Neocortical computation emerges from the dynamic interplay of excitation and inhibition, operating in a loose balance regime where recurrent and external inputs contribute comparably to neuronal activity. Neurons display broad heterogeneity in synaptic inputs and firing rates, making it essential to explain the full distribution of responses, not just the mean, when elucidating mechanisms of dynamics and computation. We examined divisive normalization in mouse visual cortex using population calcium imaging of excitatory and parvalbumin (PV) inhibitory neurons, combined with computational models of varying complexity. We found that suppression in PV neurons was transiently reduced, driven by the dynamics of subcortical input, and that heterogeneity in suppression strength was linked to population correlations, variability in excitatory-inhibitory balance, and suppression of both subcortical and local cortical inputs. Our results link local recurrent connectivity to the diversity of normalizing responses in cortex, providing a mechanistic basis for functional heterogeneity in this computation.