Open Collections

Abstract

Coordinated behavior arises from distributed brain networks, yet cellular electrophysiology and brain-wide imaging each capture only part of the relevant scale. This dissertation develops multiscale recording and analysis approaches to understand cerebellar computation at two complementary levels: how individual cerebellar neurons are embedded in cortex-wide networks and reorganize with brain state, and how population activity within the cerebellar cortex is structured and reweighted during learning. Chapter 2 (first data chapter) advances a chronic multimodal approach (“Mesotrode”) that combines mesoscale widefield calcium imaging with compact, multi-site tetrode electrophysiology while maintaining optical access and mechanical stability for longitudinal experiments. Chapter 3 combines widefield cortical imaging with Neuropixels recordings in cerebellum to map the cortical embedding of single cerebellar neurons. Cell-to-cortex correlation maps, a proxy for functional connectivity patterns, show diverse cortical affiliation motifs that remap between anesthetized and awake states. Sensory-responsive neurons exhibit stereotyped connectivity patterns, and state-dependent changes in sensory tuning are mirrored by coordinated changes in intrinsic coupling. Chapter 4 uses wide-field two-photon imaging to track climbing-fiber-evoked complex spikes from large populations of Purkinje dendrites across lobules V, VI, Simplex, and Crus I as mice learn a cross-modal perceptual decision task. To dissociate mixed selectivity from correlated behavior, I fit a ridge-regression encoding model incorporating task and movement variables and quantify each predictor group’s unique contribution. Movement explains substantial trial-to-trial variance, yet dendrites also form spatially clustered domains of mixed tuning. With learning, contextual rule cue responses decline, contralateral sensory encoding sharpens, and pre-movement choice signals become more prominent without major topographic remapping. Apparent reward-locked responses are largely attributable to sensory artifacts; after correction, reward-related signals are weak, heterogeneous, but retain information about trial context, possibly due to stereotyped movements developed through training. Together, these studies establish tools for combining multiscale neural recordings with analysis frameworks that relate cerebellar activity to both cortex-wide dynamics and local population structure. The results indicate that cerebellar signals are embedded in distributed cortical networks that remap across states, while learning selectively reweights complex spikes representations within largely stable microzonal functional units in the cerebellum.