While animal and human decision strategies are typically explained by model-free and model-based reinforcement learning (RL), their choice sequences often follow simple procedures based on working memory (WM) of past actions and rewards. Here, we address how working memory-based choice strategies, such as win-stay-lose-switch (WSLS), are represented in the prefrontal and motor cortico-basal ganglia loops by simultaneous recording of neuronal activities in the dorsomedial striatum (DMS), the dorsolateral striatum (DLS), the medial prefrontal cortex (mPFC), and the primary motor cortex (M1). In order to compare neuronal representations when rats employ working memory-based strategies, we developed a new task paradigm, a continuous/intermittent choice task, consisting of choice and no-choice trials. While the continuous condition (CC) consisted of only choice trials, in the intermittent condition (IC), a no-choice trial was inserted after each choice trial to disrupt working memory of the previous choice and reward. Behaviors in CC showed high proportions of win-stay and lose-switch choices, which could be regarded as "a noisy WSLS strategy." Poisson regression of neural spikes revealed encoding specifically in CC of the previous action and reward before action choice and prospective coding of WSLS action during action execution. A striking finding was that the DLS and M1 in the motor cortico-basal ganglia loop carry substantial WM information about previous choices, rewards, and their interactions, in addition to current action coding.
Keywords: basal ganglia; motor cortex; prefrontal cortex; reinforcement learning; striatum; working memory.
Copyright © 2023 Yoshizawa et al.