Switching state-space models enable decoding of replay across multiple spatial environments

Replay, a phenomenon whereby cells reactivate representations of spatial trajectories, is thought to be crucial for the storage and consolidation of spatial memories. This phenomenon is commonly studied with respect to a single spatial environment. However, sleeping rats can replay trajectories of the spatial environment they previously experienced and awake rats can replay spatial trajectories of the spatial environment they are in and previously experienced environments. This means that care is needed when interpreting replay, as it may represent a replay of the current environment or another environment the animal has experienced. It is possible that replays that experimenters previously thought were not interpretable are actually coherent replays of another experience. More broadly, experiences occur across environments, and how the hippocampus and other brain areas might multiplex the replay of different experiences remains unknown. These issues motivated us to develop decoding methods that account for multiple spatial environments and provide a measure of statistical confidence with regards to which environment and trajectory are being represented. Here we take advantage of a switching point process state space model framework to develop such a method, leveraging spatial rate maps in each environment and prior knowledge about the spatial structure of environments. We show on simulated data how this model can turn "noisy" spatial trajectories decodes based on considering only one environment into interpretable trajectories in another environment. We then show on real data that we can decode replay of multiple environments. We believe this state space model will be important for understanding how the brain’s storage and consolidation mechanisms work in the context of more complex, real-world experiences. Moreover, as these methods can be applied to data across brain areas, they provide a robust framework for understanding how replay is represented across regions.