Biophysical Modeling and Optical Imaging Tools for Studies of Cerebellar Motor Learning.

Motor learning is a precious capacity that humans, from toddlers to athletes and concert pianists, rely on to move with coordination and grace. The neural circuit in the cerebellum that helps us learn new movements and adjust our performance of familiar reflexes is remarkable for its quasi-crystalline architecture, which has been conserved at the cellular level across vertebrate species. This thesis develops and applies tools based on biophysical and computational modeling of neural circuits, as well as analysis of data sets from optical microscopy experiments, to study mechanisms of motor learning in the cerebellum. In the first part of the thesis, we use theoretical methods, including electrical compartmental modeling of single cells and a statistical theory of learning in networks, to explore hypotheses concerning temporal aspects of learned movements. We propose that the geometric structure of the cerebellar cortex may be useful for improving the neural representation of time. We go on to use a "lock and key" theory to explain a longstanding observation regarding the required causal temporal relationship among sensory stimuli for effective motor training. Our theory relates this timing aspect of motor learning to a biophysical mechanism, rebound conductances, present in the neurons of the deep cerebellar nuclei. In the second half of the thesis we develop and apply computational and statistical analysis procedures for analyzing experimental data sets from optical imaging experiments to test hypotheses regarding motor learning in the cerebellum. We present tools that can extract intracellular physiological signals, including neuronal spike trains and glial calcium transients, from noisy and high-dimensional imaging data sets recorded in living subjects. Using a novel miniaturized microscope, we measured intracellular signals as well as blood flow in freely moving mice during rest and activity. To examine intracellular signals in finer detail, we examined data recorded by two-photon imaging in head-fixed mice free to run on an exercise wheel. Using this technique, we discovered three forms of dynamic activation in networks of cerebellar Bergmann glia, one of which is associated with the onset of locomotor activity. High-dimensional optical imaging experiments require new computational approaches to data analysis. We developed automated procedures, including spatio-temporal independent component analysis, to identify the locations and extract the temporal dynamics of individual cells in our movie data sets. Our procedures solve the cell sorting problem that is analogous to the challenge of spike sorting in extracellular electrophysiology. Cell sorting of calcium imaging data revealed patterns of correlated complex spiking in groups of Purkinje cell dendrites. Using in vivo optical microscopy we showed that microzones of correlated complex spiking are stable over time and between resting and actively moving behavioral states. Optical imaging in combination with statistical and computational approaches to data analysis may help in coming years to answer longstanding questions concerning the cellular and network mechanisms of motor learning in the cerebellum.