Sensory origin of oculomotor variability revealed by MT population activity

Motor outputs vary even in response to identical sensory inputs, yet the origin of this variability within the sensorimotor pathway remains unresolved. Here, we evaluate how variability in sensory representations can explain behavioural fluctuations in a reflexive oculomotor task. We used Neuropixels probes to record neuronal activity in area MT of marmosets during ocular following responses and applied partial least squares regression to extract the shared variance between population activity and eye movements. Stimulus-evoked activity reliably predicted trial-by-trial variability in open-loop eye velocity. Additionally, closed-loop analyses revealed trial-by-trial correspondence between eye movements and subsequent neural responses. These results demonstrate that variability in sensory neural populations contributes to motor variability, supporting the claim that sensory noise is propagated through the sensorimotor pathway. Surprisingly, however, traditional models trained to only capture neural variability across trials poorly predicted behavioural variations, suggesting that only a subset of sensory representations is accessible to the motor system.