1Laboratoire De Neurosciences Cognitives et Adaptives , University of Strasbourg, France
*Email: a.pathak@unistra.fr
Introduction
Functional Connectivity (FC) quantifies statistical dependencies between brain regions but traditionally assumes stationarity. Dynamic Functional Connectivity (DFC) captures temporal fluctuations, offering insights into cognition and brain disorders [1]. However, DFC’s interpretation is debated, with concerns about neural vs. non-neural origins [2]. Arousal fluctuations, driven by neuromodulation, likely shape DFC. This study extends whole-brain models by incorporating time-varying neuromodulatory inputs, improving the replication of empirical DFC patterns. Findings suggest arousal plays a crucial role in DFC dynamics, refining our understanding of brain network organization. Methods The study analyzes resting-state fMRI data from 100 individuals in the Human Connectome Project [3]. Whole-brain models were built using structural connectivity data, employing two autonomous models: an oscillatory Stuart-Landau model [4] and a multistable Wong-Wang model [5]. A time-dependent modification, modeled as a Ornstein-Uhlenbeck process (tMFM) was introduced in the global excitability term. Dynamic Functional Connectivity (DFC) , measured using a sliding window approach and DFC speeds served as the model fitting targets. A Genetic Algorithm optimized model parameters by fitting simulated data to empirical observations, using statistical metrics (AIC/BIC) to compare model performance. Results Dynamic Functional Connectivity (DFC) was analyzed in resting-state fMRI using a sliding window approach, revealing two distinct phenotypes: drift and pulsatile. The drift phenotype showed a gradual slowing of dynamics, while the pulsatile phenotype exhibited brief, well-defined epochs of slow events. Two modeling approaches were explored: the eMFM (noise-driven bistability) and the MOM model (metastable oscillatory dynamics). Both generated transient DFC but failed to fully capture empirical patterns. Introducing arousal-linked modulations in excitability (tMFM) significantly improved model fit, with linear drift capturing drift phenotypes and mean-reverting dynamics modeling pulsatile phenotypes.
Discussion This study explores how incorporating time-varying parameters, specifically arousal-linked fluctuations, improves dynamic functional connectivity (DFC) modeling. Traditional models assume time-invariant dynamics, but evidence suggests cortical excitability varies with arousal. By integrating stochastic arousal terms into the eMFM framework (tMFM), we show that DFC is better captured as a time-dependent process. Compared to the oscillatory MOM model, tMFM more accurately reproduces empirical DFC patterns, though future work could extend MOM to include neuromodulatory influences. Additionally, linking DFC with pupillometry—an arousal proxy—could further refine models, offering deeper insights into neuromodulation, brain states, and cognition.
Acknowledgements The authors acknowledge support from PEPR BHT, Fondation Vaincre Alzheimers, CNRS and the University of Strasbourg References 1.https://doi.org/10.1016/j.neuroimage.2013.05.0792.https://doi.org/10.1162/imag_a_00366 3.10.1016/j.neuroimage.2016.05.062 4.https://doi.org/10.1038/s42005-022-00950-y 5.https://doi.org/10.1016/j.neuroimage.2014.11.001
