Structural Connectivity Selectively Constrains Intrinsic BOLD Timescales through Graph-Smooth Neural Activity

Structural connectivity provides the anatomical architecture supporting large-scale brain dynamics, yet how it constrains intrinsic temporal organization remains unclear. Previous studies have reported associations between resting-state fMRI timescales and structural connectivity strength, but the signal components underlying this relationship are unknown. Here, we use a graph signal processing framework to investigate intrinsic temporal properties of brain activity. Regional Blood Oxygenation Level Dependent (BOLD) signals were modeled as graph signals on the structural connectome and decomposed into low-frequency (structure-coupled) and high-frequency (structure-decoupled) components using graph spectral filtering. Using diffusion MRI-derived structural connectivity and resting-state fMRI from 100 unrelated Human Connectome Project participants, intrinsic timescales were quantified using relatively low-frequency power and related to regional structural connectivity strength while controlling for regional volume. Intrinsic timescales derived from structure-coupled signals showed robust positive associations with structural connectivity strength at both group and individual levels, whereas structure-decoupled signals exhibited substantially weaker relationships. Slow structure-decoupled dynamics were preferentially expressed in higher-order association cortex. Graph-spectral null models further demonstrated that these effects depended on the empirical organization of the structural connectome. Together, these findings provide a graph-spectral interpretation of structure-timescale coupling, demonstrating that structural connectivity selectively constrains the temporal statistics of graph-smooth neural activity.