PROJECT SUMMARY The cerebellum plays a key role in motor control, particularly in motor learning. However, there is increasing recognition that the cerebellum also participates in cognitive functions such as planning, set-shifting, abstract reasoning, attention and working memory. Indeed, cerebellar damage is associated with both motor and cognitive problems and has been linked to a wide range of neurological and neuropsychiatric disorders, from ataxia and dystonia to schizophrenia and autism. Despite this remarkable functional heterogeneity, the cerebellar microcircuit is remarkably homogeneous, both across its different regions and across different animal species. This has led to the theory that the cerebellum may accomplish its job by performing one universal computation, which can be applied to both motor and non-motor functional domains by connecting the cerebellum to different brain areas. Much of what we know about this universal computation comes from mechanistic studies that have revealed what the cerebellum does – and how it does it – by manipulating and recording neural activity in the cerebellum of mice and other animals performing simple sensorimotor learning tasks. Based on this previous work, a picture is emerging that the cerebellum is a ‘machine’ specialized for supervised learning, whose function can be summarized succinctly: to learn how to eliminate errors. While there is good support for this idea for learning tasks in which the error signals are low-level and conveyed directly from the sensory periphery to the cerebellum via subcortical pathways, there is almost nothing known about the function of high-level error signals originating in parts of the brain that are not directly connected to the sensory or motor periphery. The goal of this project is to start filling this gap by examining how the cerebellum operates when mice are trained in 3 new learning tasks recently developed, in which the error signals originate in the primary somatosensory cortex (aim 1), the cerebellum itself (aim 2) and the prefrontal cortex (aim 3). To assess function, the proposed experiments will record and manipulate the error signals present during the learning process with an unprecedented level of temporal and cellular specificity, using new high-density Neuropixels tools and cell-specific photostimulation with optogenetics. This research could help develop new therapeutic approaches to treat both motor as well as cognitive disorders associated with deficient connectivity between the cerebellum and the rest of the brain, not by targeting molecular mechanisms of neural plasticity, but the instructive error-related signals that drive them.