Shared and distinct cortical mechanisms for working memory and decision-making

The dorsolateral prefrontal cortex (DLPFC) and lateral intraparietal cortex (LIP) in the primate brain are critically involved in working memory during tasks that require the retention of information over a delay. These same regions have also been implicated in reinforcement learning (RL), where information about an animal’s choice and its outcome is retained to update future reward expectations based on past experiences. We investigated whether spatial memory, required across different behavioral contexts, relies on a shared neural mechanism. To explore this, we analyzed neural activity recorded from rhesus monkeys engaged in three distinct tasks—the oculomotor delayed response task (ODR), a visual search task, and the matching pennies game—each requiring the retention and use of similar spatial information under different cognitive demands. The ODR task demands only prospective memory, as the selection of action is dictated by the location of visual cue, and the subject must retain this intended action for execution after a temporal delay. In contrast, the matching pennies task engages both retrospective and prospective memory: retrospective memory of previous choice and its outcome to inform decision-making, while prospective memory is needed to carry out that decision. Visual search task, by comparison, does not explicitly require either retrospective or prospective memory.

Our analysis revealed that neural signals encoding retrospective memory of the animal’s choice in the visual search and matching pennies tasks were not correlated with the prospective working memory signals of visually cued locations in the ODR task, in either the DLPFC or LIP. Moreover, retrospective choice signals in the visual search and matching pennies tasks were not correlated with each other. In contrast, neural activity related to upcoming choices (prospective memory) in the LIP showed significant correlations across all three tasks. In the DLPFC, prospective choice signals were correlated between the visual search and ODR tasks, but not between those tasks and matching pennies. Additionally, in the DLPFC, neural signals representing previously rewarded choices were significantly correlated with working memory signals during the ODR task.

These results suggest that the LIP supports a consistent, shared mechanism for prospective memory linking a committed action to its eventual execution. In contrast, the DLPFC might mediate the transformation of retrospective memory-integrating past choices and outcomes – into a decision and its associated prospective memory.