Feature-tuned synaptic inputs to somatostatin interneurons drive context-dependent processing

Mapping neural computation onto the functional microarchitecture of sensory circuits is essential for understanding how brain circuits transform input signals into coherent percepts. Many higher-order perceptual processes emerge in the cortex, yet relatively little is known about how specific connectivity motifs give rise to these computations. To address this challenge, we combined single cell and population-level physiological recordings and perturbation methods to map a context-dependent cortical computation onto the synaptic microarchitecture of the mouse primary visual cortex (V1). We demonstrate a precise, recurrent circuit between cortical pyramidal cells (PCs) and somatostatin (SST) inhibitory interneurons that mediates context-driven figure/ground modulation in V1. Through a like-to-like connectivity rule from PCs and SSTs, this circuit explains SSTs visual encoding properties and their resulting impact on contextual modulation in V1. These findings reveal key synaptic and circuit mechanisms that may underlie the earliest stages of scene segmentation in the visual cortex. Moreover, they raise the notion that feature-specific excitatory-inhibitory microcircuitry from PC to SSTs could be a general strategy that the cortex exploits to give rise to higher level computations.