Cross-sectional and longitudinal trajectories of structural connectome aging diverge: a diffusion MRI study across the adult lifespan

Cross-sectional studies of structural-connectome organization in aging report inconsistent findings, and it is not always clear whether the inconsistency reflects pipeline differences or limits of cross-sectional design. We addressed this directly using diffusion-MRI data from 194 healthy adults aged 21–89 years (377 scans) drawn from the Dallas Lifespan Brain Study, of whom 115 had longitudinal follow-up. Whole-brain structural connectomes were reconstructed using constrained spherical deconvolution tractography over the 116-region Automated Anatomical Labeling atlas, and network robustness was quantified through threshold and attack simulations. We report cross-sectional age associations on baseline-only data with FDR correction, bootstrap mediation by fractional anisotropy, paired-difference and linear mixed-effects models for within-subject change, ICC(3,1) test–retest reliability, an explicit returner-versus-non-returner attrition test, and machine-learning age classification with permutation testing. Cross-sectionally, FA showed steep decline with age (r = -0.629, pFDR < 0.001), while topology metrics showed only modest, mostly positive associations (clustering: r = +0.246, pFDR = 0.003; threshold h50: r = +0.138, n.s.). Bootstrap mediation showed no reliable indirect effect through FA (a×b = +0.074, 95% CI [-0.003, +0.150]). Within the same individuals, longitudinal tracking revealed significant decline in robustness (t = -2.94, p = 0.004), threshold AUC, clustering, density, mean strength, and number of connections (all pFDR < 0.05); FA itself did not change reliably within subjects. ICCs of all network metrics were poor (< 0.30) and returners did not differ from non-returners on any baseline measure. Logistic regression separated younger (<50 yr) from older (≥50 yr) adults at 77.7% accuracy (AUC = 0.85; permutation p = 0.001). Cross-sectional and longitudinal designs yield qualitatively different conclusions about the aging connectome.