Cortico-striatal dynamics across working memory stages

ABSTRACT Working memory depends on the temporary retention and manipulation of information, bridging the gap between short-term memory and information processing functions. However, when the same working memory task is repeated over several days, it raises the question of whether the rule or task set becomes automated (or proceduralized). The medial prefrontal cortex (mPFC) is crucial for working memory. Yet, the role of the dorsolateral striatum (DLS) in the automation (proceduralization) of rules or task sets remains to be clarified. Using a longitudinal approach of the “delay non-match to place” (DNMP) task in a T-maze combined to chemogenetic inhibition of the mPFC or DLS in mice, we show that the mPFC becomes less critical in the maintenance phase of the task as behaviour progressively shifts toward automation. During this phase, the DLS facilitates automated processing. Accordingly, silencing through chemogenetic inhibition of the DLS during maintenance triggers an adaptation in learning strategies, reactivating a goal-directed behaviour. Our findings strengthen memory traces as a dynamic reorganization of neural networks, challenging the classical view of information migration between brain structures. We here propose that the memory trace remains in a dormant state—less energy-consuming for the system—while still allowing for rapid flexibility in case of task modification.