Spatiotemporally flexible subnetworks reveal the quasi-cyclic nature of integration and segregation in the human brain.

Abstract

Though the organization of functional brain networks is modular at its core, modularity does not capture the full range of dynamic interactions between individual brain areas nor at the level of subnetworks. In this paper we present a hierarchical model that represents both flexible and modular aspects of intrinsic brain organization across time by constructing spatiotemporally flexible subnetworks. We also demonstrate that segregation and integration are complementary and simultaneous events. The method is based on combining the instantaneous phase synchrony analysis (IPSA) framework with community detection to identify a small, yet representative set of subnetwork components at the finest level of spatial granularity. At the next level, subnetwork components are combined into spatiotemporally flexibly subnetworks where temporal lag in the recruitment of areas within subnetworks is captured. Since individual brain areas are permitted to be part of multiple interleaved subnetworks, both modularity as well as more flexible tendencies of connectivity are accommodated for in the model. Importantly, we show that assignment of subnetworks to the same community (integration) corresponds to positive phase coherence within and between subnetworks, while assignment to different communities (segregation) corresponds to negative phase coherence or orthogonality. Together with disintegration, i.e. the breakdown of internal coupling within subnetwork components, orthogonality facilitates reorganization between subnetworks. In addition, we show that the duration of periods of integration is a function of the coupling strength within subnetworks and subnetwork components which indicates an underlying metastable dynamical regime. Based on the main tendencies for either integration or segregation, subnetworks are further clustered into larger meta-networks that are shown to correspond to combinations of core resting-state networks. We also demonstrate that subnetworks and meta-networks are coarse graining strategies that captures the quasi-cyclic recurrence of global patterns of integration and segregation in the brain. Finally, the method allows us to estimate in broad terms the spectrum of flexible and/or modular tendencies for individual brain areas.