Brain age gradients as intermediate phenotypes linking plasma p-tau217 to cognition in community-dwelling older adults

Deep learning-based brain age models quantify regional deviations from normative aging and may capture structural changes relevant to dementia risk. Plasma phosphorylated tau-217 (p-tau217) is a scalable Alzheimer’s disease biomarker, but its relationship to brain aging and cognition in cognitively unimpaired adults is unclear. In this cross-sectional study, we tested whether brain age patterns serve as indirect pathways linking plasma p-tau217 to cognition in the Aging Brain Cohort (ABC). Neuroimaging data from 518 adults (mean age = 43.7 years, 70.8% female) were analyzed using a validated deep learning brain age model, and decomposed via exploratory factor analysis into six gradients: frontal, dorsal, ventral, left frontotemporal, right frontotemporoparietal, and bilateral parietal. In a parallel mediation model including all six gradients as simultaneous mediators in adults aged ≥60 years ( N = 71), a significant specific indirect effect of plasma p-tau217 on Montreal Cognitive Assessment (MoCA) scores was observed through accelerated right frontotemporoparietal aging ( β = −0.111, 95% CI [−0.313, −0.010], p = 0.031). No other indirect pathways were significant, and neither the total nor direct effect was significant. These findings suggest a specific brain aging phenotype as a potential intermediate pathway linking tau-related pathology to cognition prior to clinical impairment.