Generative replay across hippocampal-neocortical circuits

Abstract

Flexible behaviour requires predictions that extend beyond direct experience, yet the neural mechanisms that support this process remain unclear. A candidate mechanism occurs during hippocampal sharp-wave ripples (SWRs), in which generative replay combines representations for discrete past events into novel activity patterns. However, it remains unknown whether comparable computations occur in neocortex. Using simultaneous electrophysiology and calcium imaging across hippocampus and neocortex, we record neural activity in freely moving mice during an inference task and in subsequent sleep. We show that neuronal sequences in primary visual cortex (V1) encode relationships that were never directly experienced, providing evidence for generative replay outside the hippocampus. Generative replay in V1 is predicted by preceding hippocampal activity, suggesting that hippocampal activity provides a teaching signal to reorganise neocortical representations during sleep. These findings identify a hippocampal-neocortical mechanism for generative replay which may support the construction of a hierarchical generative model to enable flexible behaviour.