Figure 6. Proposed model for decision-making network in cerebellum, and the interaction with brain-wide neurodynamics.

(A) Trial-averaged ipsilateral cerebellar activity for different cell types (granule, Purkinje and eurydendroid cells). Dashed line: red for heat onset; black for turn initiation. Only cells showing a significant difference in activity between correct and incorrect trials (p<0.001, t-test) are displayed, which accounted for >70% of the total recorded cells within each cell type. Left panels: correct trials. Middle panels: incorrect trials. Right panels: difference in average activity between correct and incorrect trials.

(i). The granule cells display the strongest observed pre-motor activity, as well as post-turn activity both in correct and incorrect trials.

(ii). The Purkinje cells display weaker pre-motor activity, especially in correct trials, compared to the granule cells. However, post-turn activity averaged across neurons is significantly higher in incorrect trials (p < 0.05, one-tailed Wilcoxon rank sum test).

(iii). The eurydendroid cells display some pre- and post-motor activity, driven by input from granule cells.

(iv). Diagram of the cerebellar microcircuitry. The main cerebellar inputs are the mossy fibers from various brain regions and the climbing fibers from the inferior olive. The activity of the granule cells (Figure 6A (i)) provides excitatory input via the parallel fibers to the Purkinje and eurydendroid cells, as well as inhibitory interneurons, which can suppress Purkinje cells and thus enable excitation of eurydendroid cells via the parallel fibers.

(B) Schematic overview of the proposed neurocircuitry underlying sensory information integration, learning, motor planning and action selection in our task: The heat stimulus activates the telencephalon, which inhibits thalamic activity and activates the habenula, which can regulate learning, reward prediction and the brain state. The cerebellum integrates this multimodal information, as well as that of various precerebellar nuclei, via the mossy fibers to prepare and generate movements. Cerebellar dynamics are relayed to the ARTR via the eurydendroid cells, initiating a motor command. Following movement initiation, the cerebellum receives feedback from the inferior olive and sends motor information back to the thalamus, thereby forming a closed loop for learning.