In resting state of animals, firing sequences of hippocampal place cells corresponding to the previously experienced path are frequently replayed. Interestingly, firing sequences are replayed not only in the forward direction (same ordering as the real experience), but also in the reverse direction (Foster et al., 2006). Reverse replays are mainly observed when animals are consuming rewards after the exploration, therefore it has been speculated that they may be related to encoding paths to rewards. However, the conventional recurrent network models for place cell replays learn only forward replays, and reverse replays in such models potentiate paths that diverge from the reward. To investigate the mechanism and roles of reverse replays, we constructed the biologically plausible recurrent network model that has short-term and long-term plasticity. We found that firing sequences in this network model potentiate the recurrent connections in the reverse direction more strongly, thus frequent reverse replays in the reward position result in increasing selectivity of paths that converge towards the reward. This model provides a possible explanation for how reverse replays in the goal location are involved in learning and consolidation of the optimal forward path.

研究助成：Research funds : KAKENHI(15H04265)