Corrective sub-movements link feedback to feedforward control in the cerebellum

Abstract

The ability to execute accurate movements is thought to rely on both anticipatory feedforward commands and rapid feedback corrections, yet how these control systems are integrated within cerebellar circuits remains unclear. Here, we show that corrective sub-movements (CSMs) - structured, feedback-driven adjustments occurring spontaneously during naturalistic mouse reaching - are not only encoded in anterior interposed (IntA) output neurons, but also act as instructive signals for feedforward learning. Closed-loop perturbations that trigger CSMs lead to learned shifts in the timing of future corrections, even in the absence of further perturbation. Strikingly, this learning depends on the timing of corrective responses, rather than the timing of the error itself, and is accompanied by physiological adaptation in cerebellar output neuronal firing rates. These findings reveal a cerebellar mechanism by which feedback responses train future anticipatory control signals, bridging the gap between reactive and anticipatory motor control in the cerebellum. Key Findings/ HighlightsO_LIMice generate rapid, precise corrective sub-movements (CSMs) during reach that counter both spontaneous variability and induced errors C_LIO_LICerebellar output neurons encode both predictive and corrective movements, mechanistically linking feedforward and feedback control C_LIO_LIClosed-loop optogenetic disruption of cerebellar output induced reach errors that were entirely compensated by CSMs C_LIO_LILearning was driven by the timing of corrective responses, not the timing of the initial error, suggesting an instructive role for CSMs C_LIO_LIBoth reach kinematics and cerebellar nuclear activity adapted over trials of repeated perturbation C_LIO_LIOptogenetic activation of cerebellar output drove errors and subsequent learning, but also blocked expression of that learning acutely, pinpointing cerebellar output as a key site of motor learning expression C_LIO_LIFindings suggest a neural mechanism by which feedback corrections influence learning of feedforward control policies in cerebellar circuits C_LI