Mice identify subgoal locations through an action-driven mapping process

Mammals instinctively explore and form mental maps of their spatial environments. Models of cognitive mapping in neuroscience tend to depict map-learning as a process of random or biased diffusion; however, in practice, animals explore spaces using structured, purposeful, sensory-guided actions. A promising model system for probing the relationship between spontaneous exploration and spatial cognition is threat-evoked escape behavior in mice. Notably, Shamash et al. 2021 examined how mice learn routes to a shelter when the direct path is blocked by an obstacle. They found that during a 20-min free exploration period, mice memorized allocentric subgoal locations at the obstacle edges, helping them to perform efficient two-step escape routes past the obstacle. Here we demonstrate that a particular class of movements - runs targeting an obstacle edge during exploration - plays a causal role in triggering subgoal memorization. We used closed-loop neural manipulations to interrupt running movements during exploration. Blocking edge-directed runs abolished subgoal learning. In contrast, three similar stimulation protocols that spared edge-directed runs had no such effect. We next examined the distribution of locations from which mice executed subgoal escapes. This revealed that the decision of whether to pursue a subgoal incorporates information about the mouse's position relative to the environment's layout. Thus, mice use an action-driven learning process to identify subgoals, and these subgoals are then integrated into a map-based planning process. From a reinforcement-learning perspective, this process fits with a model-free/model-based hybrid called the successor representation (Dayan 1993), albeit with several modifications. From a cognitive-science perspective, it matches the sensorimotor enactivism framework Clark 1999; Ballard et al. 1997). Overall, our results indicate that action-driven mapping may be an important component of gaining useful information about the environment and suggest the possibility of a tight link between the hippocampal mapping network and the cortico-striatal action-learning circuit.