Accounting for visual cortex variability with distributed neural activity states

Sensory neuron responses vary across repeated presentations of the same stimuli, but whether this trial-to-trial variability represents noise versus unidentified signals remains unresolved (1–3). Some of the variability can be attributed to correlations of neural activity with locomotion, arousal, and other overt movements (4–7). We hypothesized that distributed brain states, i.e., patterns of neural activity observable in other brain regions, may account for more of the unexplained trial-to-trial variability in responses of sensory cortical neurons. To test this, we used Neuropixels 2.0 probes to record neural activity in the mouse primary visual cortex (VISp) while subjects passively viewed images. We recorded videos of behavior alongside neural activity from other brain regions: in some experiments, we recorded spiking activity of anterior cingulate area (ACAd) and in others we recorded widefield calcium signals from the dorsal cortex. We then used reduced rank regression to determine what fraction of the variability in visual responses was attributable to observed behavior and to distant neural activity. Our results indicate that distant neural activity explained 57.6±5% (mean ± SEM) of the maximum explainable VISp variability, significantly (p < 0.001) more than the 24.7±4% explained by the previously demonstrated behavioral factors. The ability to account for so much variability prompted us to consider whether the remaining unexplained variability was consistent with the common assumption that sensory cortical neurons emit spikes according to a Poisson process with time-varying rate (8). We found that the residual unexplained variance of VISp neurons was below that expected from a Poisson process, with a Fano factor of 0.42±0.02, substantially less than 1. Our results argue that the Poisson-like variability arises not from noise but rather is explained by activity patterns shared with other connected brain regions.