Abstract
Understanding the role of different inhibitory interneuron subtypes in cortical computations is essential for explaining sensory processing in the neocortex. Orientation tuning in primate primary visual cortex (V1) provides a canonical model for studying how cortical sensory circuits and inhibitory interneurons compute relevant stimulus features. The selective feedforward convergence of non-orientation-selective thalamic afferents establishes initial orientation tuning in the granular V1 input layer. As signals propagate through the cortical microcircuit, orientation tuning sharpens in extra-granular layers, yet the underlying mechanisms and the contribution of specific inhibitory neuron subtypes remain unresolved. To study the role of the largest cortical inhibitory neuron subclass, parvalbumin-expressing (PV⁺) interneurons, in this V1 computation, we combined laminar extracellular recordings with bidirectional optogenetic manipulations of PV⁺ cells in marmoset V1. We find striking laminar specificity: in the granular layer, PV⁺ cells implement divisive/ multiplicative linear gain control, whereas in extra-granular layers they exert tuned nonlinear suppression that enhances orientation tuning. Computational modeling suggests that PV+ neurons control gain by modulating a neuron's spiking threshold, and orientation tuning by modulating a neuron's input noise, which regulates the neuron's input-output function. Our findings reconcile discrepancies in previous rodent studies, reveal important species differences, and establish a framework for understanding layer-dependent inhibitory computations in the primate cortex.