Cerebellar mechanisms underlying adaptive motor responses

Abstract

In order to understand the function of any brain structure, one must know what input/output transformation it performs. The term input/output transformation includes at least two stages. First, we must understand how inputs are processed. Second, we must know what the output activity encodes. Certain properties of the cerebellum make such an undertaking feasible. In this thesis I present the results of three main projects designed to study the input/output transformations of this major brain system from different angles. In the first project I investigated the relationship between spiking activity of cerebellar cortex principal neurons - Purkinje cells (PCs) - and eyelid conditioned response (CRs) profiles on a single trial basis. Systematically exploring a variety of encoding possibilities, I found that PCs do not directly encode a single kinematic variable of a CR. The best prediction was rather achieved via a dynamical model approach, where PCs provide a ‘drive’ to the eyelid plant, the dynamics of which are described by a differential equation. In the second project I addressed how the cerebellum deals with inherent uncertainty about the nature of sensory inputs. I found that under conditions of uncertainty, the cerebellum performed a probabilistic binary choice, scaling the probability of response with the similarity between current and trained stimuli. Importantly, if responses were made, their amplitude was close to the previously trained value, maintaining the adaptive nature of responses. Recordings from eyelid Purkinje cells localized this computation to cerebellar cortex. Results from large-scale computer simulation suggest that the efference copy signal is critical for the expression of target response amplitude. In the third project I studied cerebellar mechanisms of learning and expression of movement sequences. While the majority of movements we perform are composed of sequences, most of the knowledge about cerebellar learning and computation comes from tasks involving single, unitary movements. Hence, I designed a novel sequence training protocol to explicitly test the ability of the cerebellum to chain together a series of movements through associative learning processes. The results demonstrate a simple yet general framework for how the cerebellum can learn to produce a movement sequence.