Organization of Neural Representations in Mouse Posterior Cortex for Dynamic Navigation Decisions

Abstract

During navigation in dynamic environments, animals adaptively incorporate sensory information into a plan to guide their movements. The neural underpinning of this behavior must integrate sensory processing, navigation planning, and motor execution, and furthermore adapt the rules governing their integration based on experience. In this work we investigated the organizing principles of neural representations in mouse posterior cortex during dynamic navigation decisions. We trained mice to perform a virtual navigation task based on rule switches and developed behavioral models to infer latent cognitive processes of distinct timescales. Mice exhibited diverse decision-making strategies across trials that influenced the dynamics of choice formation within a trial, which was embodied in running trajectories. Using two-photon calcium imaging, we densely sampled activity from large populations of neurons in posterior cortex and characterized the distribution of single-neuron encoding of various sensory, motor, and cognitive variables across areas. We found that, while neural encoding was highly distributed across posterior cortex, it was well-described with three spatially distinct gradients for visual cue, spatial position plus dynamics of choice formation, and locomotion, with peaks in visual, retrosplenial, and posterior parietal cortices, respectively. We then compared the conjunctive structures of single-neuron encoding and the population geometry of neural representations for multiple variables across areas to test whether these areas specialize in the ways they combine different variables to serve distinct computations. Surprisingly, all areas combined variables similarly instead of creating unique conjunctions of variables, resulting in a high-dimensional, complex representation of variable conjunctions shared across areas. These results lead us to infer that for navigation posterior cortical areas are functionally organized not in a hierarchy but in parallel, where areas are specialized to handle streams of information for distinct modalities but work coherently to synthesize a general-purpose state representation of the environment and behavior that can guide dynamic navigation decisions.