Scientific lectures: The inferior olive transmits salient sensorimotor prediction errors across timescales in zebrafish

Learning occurs as animals detect and correct errors by comparing the predicted and actual sensory consequences of their actions. The inferior olive, one of the inputs to the cerebellum, is key in models of such predictive cancellation, yet direct evidence that the olive encodes sensorimotor prediction errors is limited. Using two-photon calcium imaging and glutamate sensor recordings in larvalzebrafish in virtual reality, we show that inferior-olive neurons indeed report the magnitude of the discrepancy between expected and actual visual reafference. Forward- and backward-tuned cells exhibit distinct, nonlinear relationships to swim strength and manipulated reafference that reflect motor-dependent prediction errors.

To determine how these prediction errors are computed, we imaged excitatory inputs to show how forward- and backward-tuned populations each transforms their excitatory drive. Surprisingly, we find that both receive excitation that mirrors their motor-modulated output, suggesting that prediction errors are computed in presynaptic nuclei; however, each population filters their input, preferentially relaying mismatches to the cerebellum when swim strength and visual speeds fall into distinct regimes. This enables the IO to balance actual reafference against the expectation set by swim strength. IO population activity filters salient, ethologically relevant mismatches at different timescales, and selective IO ablation abolishes behavioural responses to those stimuli.

Together, these results identify the inferior olive as a population-level saliency encoder of sensorimotor prediction errors that supports learning across multiple timescales.

Hosted by Dafni HADJIECONOMOU and Claire WYART

If you would like to meet the speaker, please contact us.

Conference location

Please join the conference in Paris Brain Institute auditorium.