Flexible integration and segregation of large-scale networks during adaptive control

Abstract

Adaptive control characterizes the dynamic adjustment of cognitive control to changing environmental demand, and has obtained growing interests in its neural mechanism for the past two decades. Recent years, interpreting network reconfiguration in terms of integration and segregation has been proved to shed light on neural structure underlying various cognitive tasks. However, the relationship between network architecture and adaptive control remains unclear. Here, we quantified the network integration (global efficiency, participation coefficient, inter-subnetwork efficiency) and segregation (local efficiency, modularity) in the whole-brain and analyzed how these graph theory metrics were modulated by adaptive control. The results showed that the integration of the cognitive control network (the fronto-parietal network, FPN), the visual network (VIN) and the sensori-motor network (SMN) was significantly improved when conflict was rare, so as to cope with the incongruent trials of high cognitive control demands. Additionally, as the conflict proportion increased, the segregation of the cingulo-opercular network (CON) and the default mode network (DMN) significantly enhanced, which may contribute to specialized functioning or automatic processing, and help to solve conflict in a less resource-intensive mode. Finally, using graph metrics as features, the multivariate classifier reliably predicted the context condition. These results demonstrate how large-scale brain networks support adaptive control through flexible integration and segregation.