Abstract
Anxiety results in sub-optimal motor performance and learning; yet, the precise mechanisms through which these modifications occur remain unknown. Using a reward-based motor sequence learning paradigm, we show that concurrent and prior anxiety states impair learning by biasing estimates about the hidden performance goal and the stability of such estimates over time (volatility). In an electroencephalography study, three groups of participants completed our motor task, which had separate phases for motor exploration (baseline) and reward-based learning. Anxiety was manipulated either during the initial baseline exploration phase or while learning. We show that anxiety induced at baseline reduced motor variability, undermining subsequent reward-based learning. Mechanistically, however, the most direct consequence of state anxiety was an underestimation of the hidden performance goal and a higher tendency to believe that the goal was unstable over time. Further, anxiety decreased uncertainty about volatility, which attenuated the update of beliefs about this quantity. Changes in the amplitude and burst distribution of sensorimotor and prefrontal beta oscillations were observed at baseline, which were primarily explained by the anxiety induction. These changes extended to the subsequent learning phase, where phasic increases in beta power and in the rate of long (> 500 ms) oscillation bursts following reward feedback were linked to smaller updates in predictions about volatility, with a higher anxiety-related increase explaining the biased volatility estimates. These data suggest that state anxiety alters the dynamics of beta oscillations during general performance, yet more prominently during reward processing, thereby impairing proper updating of motor predictions when learning in unstable environments.
Footnotes
Computational Model of reward-based learning has been implemented