Population cerebellar neurons compute rewards with single-trial precision and engineerability

Precise behavioral control of the brain and related prosthetic design require neuronal algorithms that have single-event precision and temporal accuracy in real time, no ambiguity or trial average allowed. In reward-based learning, the brain must detect each reward timely to support adaptive behavior in dynamic environments, yet the underlying neuronal computations remain elusive. Here, we identify a cerebellar mechanism encoding rewards with single-trial precision. Utilizing cleared-tissue tractography, electrophysiology, and two-photon imaging in a mouse foraging task, we describe an Instantaneous Silencing of Population Neurons (ISPN) in the deep cerebellar nucleus (DCN), triggered by synchronized Purkinje cells (PCs) activity. This transient silencing emerges 100 ms after reward delivery and induces rebound phasic firings of ventral tegmental area (VTA) dopamine neurons for one reward event. Such single-trial precision and temporal accuracy enabled precise reward control, targeting either PC-to-DCN or DCN-to-VTA projections. A single 5-ms optogenetic pulse, leaving 99% of behavioral time untouched, is sufficient to elicit one ISPN for reward, shifting behavior with single-trial precision. Our findings underscore the cerebellum's critical role in real-time reward coding and illustrate how our brains utilize population coding to achieve both temporal and single-event precision, the necessary features in real-world behavioral encoding.