Project Summary/Abstract Prospectively selecting an appropriate sequence of actions from a large number of alternatives requires extensive computation. Although behavioral and modeling studies have advanced our understanding of how the internal model of the animal’s environment might be interrogated to evaluate the outcomes of future actions, the biological mechanisms of such computation remain poorly understood. In addition, new techniques to monitor the activity of many neurons in multiple brain areas are enabling researchers to systematically test theories of how temporally extended computations for high-order cognition like planning might be subserved by broadly distributed neural activity. To capture these exciting opportunities, we have chosen to study the neural basis of planning during the four-in-a-row task, which is both rich enough to keep the essential elements of complex planning and tractable for rigorous computational and neurophysiological investigations. The objective of the four-in-a row task is to place four consecutive stones in a grid before the opponent does so. This task allows us to study the neural processes for iteratively evaluating possible outcomes or values of alternative action sequences across many well-defined states. Recent work has characterized the computational strategies of humans in this task and how they are refined during practice. In addition, our preliminary studies show that non-human primates can be trained to perform the same task. In Aim 1, we will apply rigorous statistical and machine learning techniques to understand how the animals use feature-based evaluation and potentially tree search to make a move. In Aims 2 and 3, we will examine how computational aspects of planning rely on the coordination between neural populations in fronto-striatal and fronto-hippocampal networks. In Aim 2, we will simultaneously record the activity of many neurons in the dorsolateral and dorsomedial prefrontal cortex and in the caudate nucleus to test how these regions are involved in integrating various task features during action selection. In Aim 3, we will simultaneously record from the prefrontal cortex and hippocampus to test whether dynamic changes in the state representation in the HPC are reflected by successive transformation of neural signals related to values and actions in the PFC. We will also test whether sequential moves decoded from neural activity in the HPFC and PFC during replays correspond to past and future choices made by the animal. The results from these experiments will improve our understanding of how the brain develops good plans adaptively.