Abstract

Two-photon calcium imaging has made it possible to record from tens or hundreds of thousands of neurons simultaneously, offering an opportunity to understand the large-scale structure of the neural code. I will discuss some applications of this approach asking how visual and non-visual information are encoded in large populations of mouse visual cortex, and how the nature of this code differs between cell types as identified by mouse lines or by post-hoc spatial transcriptomics.

Both visual and non-visual information are encoded with a geometry where the variance of the nth dimension scales as a power law n^(–alpha). The exponent is larger for spontaneous activity than for sensory responses, and is larger for inhibitory than excitatory populations, indicating a lower-dimensional, more redundant code. Around 30% of the reliable spontaneous variance of excitatory populations was predictable from behavior assessed by video, and around twice as much in inhibitory populations. The spaces of behaviorally- and visually-driven activity overlapped only in one dimension, which accounted for 25% of visually-evoked variance in inhibitory populations but only 5% in excitatory populations. Visual cortical responses to sound stimuli occupied the behavioral dimensions, and were coincident with startle responses, suggesting they primarily reflect sound-evoked changes in alertness.

Post-hoc transcriptomic analysis of imaged cells revealed that inhibitory cells coupled to spontaneous dimensions in a manner correlated on their fine transcriptomic type, but their sensory responses only correlated with coarse transcriptomic type. Analyses to date indicate no correlation between the spontaneous or visually-evoked activity of excitatory cells and their transcriptomes.