Recurrent Excitatory Networks of the Cerebral Cortex: Patterned Inputs, Transformations, and Responses During Natural Vision

Abstract:

The brain is a highly interconnected system, with extensive connectivity both between brain areas and also within areas. Inside cortical areas and layers, neurons receive a majority of their inputs from other local neurons within a few hundred microns' distance. These recurrent connections, mostly excitatory-excitatory connections, can dramatically shape neurons' responses: their ongoing activity, their individual responses, and how the network responds to input. Thus, recurrent connectivity can have dramatic effects on neural computation. The Histed group uses cellular-resolution brain stimulation in vivo to understand how neurons are coupled to one another within cortical networks and to understand how this coupling affects activity, computation, and behavioral control. We have made progress understanding how single cells transform their inputs (single cell response functions, activation functions, or f-I curves) in vivo, and how the cortical recurrent network transforms dynamic inputs which change over time. One important finding is that the upper cortical layers of V1 in the mouse have dynamic responses, created by recurrent interactions, which filter static visual inputs and produce sparse and reliable responses during natural visual experience. This work shows how natural visual responses, often stronger, sparser, and more reliable than responses to edges or gratings, can be created by an interaction of inputs with the cortical local circuitry. Some of the functions we find for recurrent circuits in the brain parallel functions for recurrent circuits in machine learning systems (e.g. ChatGPT). We have also developed viral tools for stable and consistent expression of a calcium indicator and opsin (GCaMP8s, soma-targeted ChrimsonR) to enable cell-specific optogenetic stimulation experiments and shared these with the community.