Neural mechanisms of planning
OA Version
Citation
Abstract
Planning is foundational to human cognition, enabling flexible, goal-directed behavior. This cognitive process is often altered in disorders such as schizophrenia and obsessive-compulsive disorder. Despite its relevance, the precise brain mechanisms underlying planning are poorly understood. A challenge in studying planning comes from the various levels of abstraction at which planning can occur. Here, I consider planning as the evaluation, selection, and preparation of future actions in anticipation of future needs. The aim of this dissertation is to resolve some of the gaps in our understanding of planning and its neurobiology. In motor planning, the execution of a precise action requires that the appropriate movement is performed at the correct time. Previous work has established the role of preparatory ramping activity in motor planning. Inspired by past work on memory, I asked whether variability in ramping dynamics among cortical neurons can provide information about the ‘what’ and ‘when’ of motor plans. By evaluating a publicly available dataset, I show that the variability of ramping speeds across neurons in the mouse frontal motor cortex reflects a pattern of heterogeneity that contributes to the population encoding of the ‘what’ and ‘when’ of motor plans. These findings draw parallels between the neural correlates of prospective actions and retrospective events. At a higher level of abstraction, planning can be performed using an internal model of how the world may change or react to our actions. Using this world model, actions can be evaluated based on their potential outcomes and the value of those outcomes. To study the mechanisms underlying model-based planning, I trained rats to perform a two-step decision-making task. Improvements in training standardization and automation allowed behavior to be monitored throughout the acquisition of this task, revealing two learning processes: an early phase where rats learn task statistics independently of rewards, and a later phase of rapid, goal-directed learning associated with model-based planning. This reveals how animals adapt their strategies over time to optimize performance. A core computation in model-based planning is the evaluation of actions using a learned world model. The neural circuit supporting this function is unknown. The retrosplenial cortex is a strong candidate region given its involvement in value-based decision making and spatial reasoning. Using a potent and selective chemogenetic strategy for inactivation, I tested whether the retrosplenial cortex is required for model-based planning in rats during the two-step task. Surprisingly, the inactivation of the retrosplenial cortex increased the degree to which rats' behavior shows evidence of model-based planning, while decreasing patterns of model-free, reward-seeking behavior. This result suggests that this brain region may mediate interactions between model-based and model-free control of behavior, providing a potential neural mechanism for this trade-off. In summary, the findings presented in this dissertation offer new insights into the neural mechanisms of planning. In particular, they highlight cortical contributions to planning across two spatial and temporal scales: the preparation of an imminent action seconds prior to its execution and the evaluation of alternative actions based on the outcomes they may produce multiple steps into the future.
Description
2024