Within-dataset relationships between SEV and cognitive control.

We found that lower SEV, indicating reduced flexibility in brain state engagement, was correlated with worse out-of-scanner task-based and questionnaire measures of cognitive control across all three transdiagnostic datasets.

SEV predicted cognitive control in external samples.

A) Linear models were trained to predict cognitive control using SEV in the main dataset before the models were applied to the validation datasets. We also performed the analysis again after swapping the datasets (i.e., training on one of the validation datasets before testing on the main dataset). We found that SEV can successfully predict both inhibition B) and shift C) in previously unseen individuals, regardless of which sample was used for model training.

Brain networks underpinning flexible engagement of brain states.

We identified brain networks supporting SEV in main A), inhibition validation B), and shift validation C) datasets using 10-fold connectome-based predictive modeling. The lower triangular portion (red) indicated the positive SEV network, whereas the upper triangular part (blue) indicated the negative SEV network. We normalized the number of significant edges by the total network size (i.e., the number of edges between or within networks). Both the SEV positive D) and negative E) networks predicted resting-state SEV in the other two datasets. Correlation values between SEV network functional connectivity and SEV are shown in matrices. MF, medial frontal network; FP, frontoparietal network; DMN, default mode network; Mot, motor network; VI, visual I network; VII, visual II network; VAs: visual association network; SN, salience network; SC, subcortical network; CBL, cerebellum; FC: functional connectivity; SEV: state engagement variability.

Moment-to-moment changes in SEV and cognitive control networks aligned over time.

In our main dataset, we successfully identified brain networks predicting inhibition and shift A). Using these brain networks, we investigated if moment-to-moment engagement of the SEV network aligned with moment-to-moment engagement of the cognitive control network B). Moments of higher SEV corresponded to moments of functional connectivity related to better cognitive control. We further found that the temporal alignment between SEV and shift network dynamics was stronger in HCs than in patients in the main and inhibition validation datasets C).