Sensory prediction errors predict motor prediction errors

Abstract

When sensory inputs can be predicted by an organisms own actions or external environmental cues, neural activity is often attenuated compared to sensory inputs that are unpredictable. We have recently demonstrated that attenuation to predictable inputs is also observed when stimulating the motor system with transcranial magnetic stimulation (TMS). Akin to sensory attenuation, predictable TMS excites the motor system less effectively than unpredicted TMS. However, it remains unclear whether these motor prediction signals are related to, or even dependent on, sensory prediction. Using dual-site TMS to target two brain regions, we arranged different warning cues to predict different regions of stimulation and measured motor attenuation using motor-evoked potentials. We found that expecting TMS over the motor cortex produced stronger attenuation than expecting TMS over a non-motor region, confirming that the attenuation observed is directly linked to activation of the motor system and not due to the sensory by-products of TMS. Using combined TMS-EEG, we measured motor attenuation with motor-evoked potentials, and simultaneously measured sensory attenuation to the sound of TMS (a coil "click") with auditory-evoked potentials. We found that both motor and auditory potentials were attenuated to predictable TMS compared to unpredictable TMS. Critically, the magnitude of auditory attenuation predicted the magnitude of motor attenuation. Our results reveal a close correspondence between error processing in the sensory and motor systems. The findings provide compelling evidence that predictive coding is governed by domain-general properties across distinct neural systems, which share common mechanisms responsible for all forms of predictive learning.