PROJECT SUMMARY In the mammalian neocortex, cortical computations result from the activity of neural circuits composed mainly of glutamatergic excitatory and GABAergic inhibitory neurons (INs). INs profoundly influence cortical computations and dynamics, via a diversity of types, with distinct morphological, electrophysiological, and molecular properties. In mouse cortex, molecular markers identify four major IN classes: parvalbumin- (PV), somatostatin- (SST), vasoactive intestinal peptide- (VIP), and LAMP5- expressing INs. Mouse studies using Cre- recombinase lines have revealed class-specific IN connectivity and function, but whether these findings translate to non-human primates (NHPs) and humans remains unknown. Understanding IN function in NHPs is critical, as they are the closest model to humans, where IN dysfunction has been implicated in several neurological and psychiatric disorders, such as epilepsy, migraine, schizophrenia, and Alzheimer's disease. Studies of INs in NHPs have been hindered by the lack of cell-specific viral tools, but recent technological advances are beginning to enable cell-type specific targeting via enhancer-specific AAVs, albeit vector development is still based on a mouse-first approach. In the previous grant cycle, we validated in marmoset visual cortex several enhancer-AAVs for transducing specific IN types, and identified two vectors, one based on the h56D promoter, the other on the S5E2 enhancer, with high specificity and efficiency for GABAergic and PV- INs, respectively. We generated constructs of these vectors carrying various transgenes, and began investigating PV-IN connectivity and function. We found that the connectivity and function of PV-INs depends on cortical layer. In V1 input layer (L)4C, PV-inhibition is divisive and linearly controls gain but not orientation tuning (OT), but outside L4C it scales neuronal responses non-linearly and controls OT. Layer-differences in PV-IN function likely arise from both layer-spec