Abstract

How different complex features of the dynamics emerge from a given structural connectivity (SC) is of great interest for our understanding of brain function, in health and disease. Among the complex dynamical features that are commonly described, the continuously evolving correlation patterns of the brain (Functional Connectivity Dynamics, FCD) has been extensively characterized and described as a marker for aging and psychiatric disorders. More recently, the existence of high-order interdependencies of either synergistic or redundant types has also been shown to change with aging, with brain activity from older healthy subjects showing more redundancy. To understand how different topological features of the structural connectome can be related to these features of the functional connectivity, we built a set of 240-node structural connectomes with varying degrees of integration and segregation, assessed with metrics from network analysis. These connectomes included artificial networks such as Watts-Strogatz or modular networks, and empirical human connectomes that were perturbed to manipulate their integration and segregation measures. The set of over 40 SCs spanned a broad range of the small-world coefficient omega, sweeping from lattice-like to small-world to random-like networks. Then the SCs were used to simulate neural activity using a Wilson-Cowan model modified with homeostatic inhibitory plasticity, to ensure an oscillatory regime over a wide range of the global connectivity parameter G. Then, the simulated activity was assessed for the FCD, using the variance of the FCD matrix as an index for dynamical richness, and for the emergence of synergistic interdependencies, by randomly sampling groups of 3, 4, or 5 nodes. The nature of the high-order dependencies was assessed using the O-information metric, that is negative for systems dominated by synergy and positive for those with more redundancy. In all the networks, the higher variance of FCD and the most synergistic n-plets were found at an intermediate value of G, were the metastability is also high and the network synchrony is moderate. When looking at the structural features, the networks that maximized both complex dynamical features were those having a positive omega index but close to zero. This indicates networks of the small-world type, which have a balance between their integration and segregation metrics. Our results are in agreement with what has been observed in the brain SC and its dynamics as the brain ages.