Distributed resting-state functional connectivity patterns predict language performance in later adulthood

Language ability in later adulthood depends on functional brain organisation extending beyond classically defined language regions, yet whether resting-state functional connectivity (rs-FC) captures this distributed architecture, and whether predictive information is concentrated within a language-specific network or spread across domain-general systems, remains unclear. Using data from the Human Connectome Project Aging dataset (N = 445; age range: 40–86 years), we combined connectome-based predictive modelling (CPM) and graph-theory analyses to examine the intrinsic neural architecture supporting vocabulary comprehension (TPVT) and single-word reading recognition (READ). CPM yielded significant out-of-sample predictions for both measures (TPVT: r=.233, p=.001; READ: r=.196, p=.001), with predictive edges predominantly spanning inter-network connections across the default mode, fronto-parietal, and dorsal and ventral attention networks; language network edges constituted a non-majority of top-ranked features in both models. Graph-theory analyses of a functionally defined resting-state language network revealed that within-network global efficiency was positively associated with TPVT in lower-educated individuals but negatively so in higher-educated individuals, whose performance was instead linked to greater cross-network connectivity across left temporal, parietal, and prefrontal nodes. Node-wise PCs showed age-dependent reversals in left temporal and temporo-occipital regions for both tasks, with higher cross-network integration benefiting younger but not older adults, consistent with dedifferentiation accounts of cognitive ageing. Together, these findings position language performance in healthy ageing as a whole-brain phenomenon shaped by the interplay between domain-general and language-specific network organisation and its interaction with life-course factors, calling for a shift away from language-centric models toward frameworks that foreground distributed connectivity.