A Population Coupling Model Identifies Reduced Propagation from V1 to Higher Visual Areas During Locomotion

Abstract Point process generalized linear models (GLMs) have been a major tool for studying coordinated activity across populations of neurons. These models typically quantify how the spiking of a single neuron depends on the past activity of other neurons at multiple time lags, and the resulting neuron-to-neuron interactions are then aggregated to obtain population coupling effects. However, when neurons within the same population exhibit similar spiking patterns, explicitly modeling individual interactions can become redundant and unnecessarily increase model complexity. In such cases, population-level formulations may offer a more efficient alternative. For example, biophysical population models often characterize circuit dynamics using the average firing rate across neurons within a population, and recent data-driven approaches have similarly demonstrated the utility of population-level statistics for capturing crosspopulation interactions. Motivated by this consideration, we reformulate the GLM framework to operate directly at the population level. The resulting model, which we call pop-GLM, provides a computationally efficient method for estimating coupling between populations. In a simulated dataset, we show that pop-GLM achieves greater sensitivity in detecting coupling effects and can account for trial-to-trial variation in stimulus drive, which would otherwise introduce bias. We introduce a modification to the conventional GLM structure that is required to achieve the population-level formulation. We then apply pop-GLM to real data and identify decreased functional connectivity from primary visual area (V1) to a higher visual area during locomotion, a change not detected by single-neuron GLMs. Author summary A central goal of systems neuroscience is to understand how multiple populations of neurons, across different brain areas, interact as a coordinated circuit to produce perception and behavior. A common way to measure such interactions is to model how each neuron’s activity depends on the recent activity of others, and then combine these neuron-by-neuron estimates. When neurons in an area respond similarly, however, this becomes inefficient and hard to interpret. We instead formulated and investigated a new method that estimates functional interactions directly between two whole populations of spiking neurons, and showed it can be more sensitive and robust than previous approaches. To illustrate, we discovered decreased interaction between two mouse visual areas during locomotion, a change that previous techniques could not detect. The method should aid investigators in searching for important functional relationships across populations of neurons, with precise time-scale resolution.