PUPIL-LINKED AROUSAL REFLECTS CHANGE POINT PROBABILITY IN A PASSIVE CONTEXT

Predictions drive our interpretation of the world and need to be updated after unexpected changes in the environment. The precise neural circuits that implement this fundamental building block of sensory inference are still unknown. Here, we aimed to isolate the neural correlates of prediction updating by bypassing manual reports and their associated activity.
Participants performed a passive and active version of a novel protocol. During simultaneous EEG and pupillometry, participants listened to a sequence of pure tones whose frequency was sampled from one of two distributions (tone duration, 0.5s; inter-tone-interval, 1s; frequency range, 0.5-2kHz; overlap distributions, 20%). After each tone there was a 5% chance of a “change point”, meaning that the current generative distribution switched. The active context entailed counting state changes.
We found robust tone-evoked pupil responses in the passive context, albeit smaller than during active. Tone-evoked pupil responses in both contexts were linearly related to “change point probability” (CPP), derived from a Bayesian belief updating model. The strength of passive CPP encoding was about 85% (±16% S.E.M.) of that in the active condition and occurred slightly later during the tone-evoked response. In line with the pupillometry findings, we observed enhanced N200, P300 and alpha band suppression to stimuli signaling change points in both passive and active contexts.
These findings indicate that physiological signatures of prediction updating occur reliably during a passive protocol. This approach will enable the study of aberrant processing in clinical populations and underlying mechanisms in animal models, thereby strengthening their translational link.