Conserved neural population geometry supports behavioral generalization

Abstract

Learned behaviors often generalize across contexts even though neural representations can vary substantially between environments. How the brain preserves task-relevant information across these changing neural representations remains unclear. This problem is particularly evident in the hippocampus because hippocampal spatial representations reorganize across environments even when learned behaviors generalize successfully. Here, we recorded hippocampal neuronal activity using calcium imaging while rats performed a conditioning task in two distinct environments. Despite remapping of spatial representations, the low-dimensional organization of task related population activity was conserved across contexts, and the temporal relationships among neural population states during task execution were preserved. Strikingly, this relational organization was conserved not only across contexts for individual animals, but also across animals, despite differences in the underlying neural populations and individual experiences. Moreover, this conserved organization was sufficient to support transfer of task decoding across animals, demonstrating that task information could generalize between independently learned neural representations without shared neuron identities. These findings identify a conserved population level neural organization through which the hippocampus can support behavioral generalization despite contextual remapping and suggest that shared organizational neural mechanisms and structures may underlie behavioral generalization across individuals.