Beta oscillations and pupil-linked arousal regulate the updating of cognitive models.

Abstract

The beta rhythm (neural oscillations in the range 13-30 Hz) is associated with the control of movement. Discrete movements (such as a button press) are bounded by pre-movement dip in beta power and post-movement beta rebound, which are thought to release and terminate the action, respectively. However, an additional or alternative role has been suggested for the beta rebound, in regulating the extent to which sensory feedback updates the task model. Using EEG and pupillometry in human volunteers of both sexes, we developed a task that dissociates updating of a cognitive model from motor control. We found that beta oscillations were modulated in two independent modes - phasic and tonic - with distinct computational correlates. When abstract task models are updated by integrating the outcomes of several actions, there is a tonic suppression of the beta rhythm, which originates at the time of the post movement rebound, spans multiple trials and is measurable at both the pre- and post- movement stages. Independently, phasic trial-to-trial fluctuations in beta power, primarily measurable in the pre-movement phase, were associated with motor readiness. The decrease in tonic beta power during model updating was mediated by an increase in tonic pupil dilation, a known correlate of model updating often considered a proxy for neuromodulatory tone.<b>Significance Statement</b> Beta oscillations have traditionally been linked to movement, but are increasingly implicated in learning and cognitive control. We show that these functions are supported by two computationally distinct modes of beta activity. A transient, phasic component reflects motor readiness immediately surrounding voluntary actions. In contrast, a sustained, tonic suppression of beta develops over multiple trials as evidence accumulates that an internal model of the environment should be updated. This tonic signal emerges from the post-movement beta rebound, tracks Bayesian estimates of state uncertainty, and parallels tonic pupil-linked arousal. These findings unify motor and cognitive accounts of beta oscillations by demonstrating that beta supports distinct computations on separate timescales.