Computational modelling reveals neurobiological contributions to static and dynamic functional connectivity patterns

Functional connectivity (FC) is a widely used indicator of brain function in health and disease, yet its neurobiological underpinnings still need to be firmly established. Recent advances in computational modelling allow us to investigate the relationship between FC and neurobiology non-invasively. These techniques allow for targeted manipulations to study the effect of network disturbances on FC. Most modelling research has concentrated on replicating empirical static FC (sFC). However, FC changes over time, and its dynamic properties are closely linked to behaviour and symptomatology.

In this study, we adapted computational models to reflect both sFC and dynamic FC (dFC) of individuals, allowing for a more comprehensive characterisation of the neurobiological origins of FC. We modelled the brain activity of 200 healthy individuals based on empirical resting-state functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI) data. Simulations were conducted using a group-averaged structural connectome and four parameters guiding regional brain activity: i) G, a global coupling scaling parameter; ii) J_i, the local inhibitory current; iii) J_NMDA, the excitatory NMDA synaptic coupling parameter; and iv) w_p, the excitatory population recurrence weight. We evaluated the models based on four metrics: a) the sFC, b) the FC variance, c) the temporal correlation (TC), and d) the node cohesion (NC). The optimal model for each subject was identified by the fit to both sFC and TC. We analysed associations between brain-wide sFC and TC features with optimal model parameters and fits with a univariate correlation approach and multivariate prediction models. In addition, we used a group-average perturbation approach to investigate the effect of coupling in each region on overall network connectivity.

Our models could replicate empirical sFC and TC but not the FC variance and NC. Both fits and parameters exhibited strong associations with brain connectivity. G correlated positively and J_NMDA negatively with a range of static and dynamic FC features (|r| > 0.2, p(FDR) < 0.05). TC fit correlated negatively, and sFC fit positively with static and dynamic FC features. TC features were predictive of TC fit, sFC features of sFC fit (R ² > 0.5). Perturbation analysis revealed that the sFC fit was most impacted by coupling changes in the left paracentral gyrus (Δr = 0.07). In contrast, the left pars triangularis impacted the TC fit most strongly (Δr = 0.24).

Our findings indicate that neurobiological characteristics are associated with individual variability in sFC and dFC, and that sFC and dFC are shaped by small sets of distinct regions. In addition, we show that brain network modelling can replicate some, but not all, properties of dFC, and model fits are strongly influenced by specific FC patterns. By modelling both sFC and dFC, we could produce new insights into neurobiological mechanisms of brain network configurations.