THE INFERIOR OLIVE RELAYS SENSORIMOTOR PREDICTION ERRORS TO SHAPE MOTOR LEARNING IN ZEBRAFISH

Inferior olive (IO) neurons project climbing fibres to the cerebellum and are proposed to signal prediction errors that drive adaptive motor learning. However, how these neurons compute, encode, or transform prediction errors remains contested. A prediction error can occur during mismatches between expected and actual sensory feedback for a given action or event in the world. To test this, we used transparent zebrafish larvae in which the conserved olivocerebellar circuit is fully optically accessible, enabling large-scale two-photon calcium and glutamate imaging combined with virtual-reality motor perturbations to examine the origin and transformation of prediction-error signals.
A model of the expected visual feedback produced by zebrafish swimming behaviour predicted how IO neurons should encode discrepancies between expectation and sensory input. IO activity matched these predictions, increasing proportionally when visual feedback deviated from expectation. Simultaneous imaging of excitatory input to the IO using iGluSnFR and climbing-fibre output to the cerebellum, however, revealed that prediction-error signatures were already present in IO inputs, suggesting that the IO does not compute errors de novo. Instead, input–output comparisons showed that the IO may act as a filter, transforming high frequency raw sensory–motor discrepancies into low frequency instructive teaching signals for cerebellum-dependent learning. Indeed, during cerebellar-dependent motor adaptation, ablation of the IO significantly impaired behavioural adaptation, cementing the role of the IO in relaying error-based teaching signals for effective cerebellar-dependent motor control.