medRxiv · Apr 29, 2026
Longitudinal Brain Atrophy Patterns in Dementia and Cognitive Decline: the Framingham Heart Study
By Rehman H, Tao Q, Nolan JC et al.
Abstract Background Characterizing longitudinal patterns of brain atrophy that distinguish Alzheimer’s disease (AD) and related neurodegeneration along with normative aging remains a major challenge. We aimed to identify data-driven longitudinal brain atrophy components and evaluate their associations with plasma AD biomarkers and cognitive outcomes in a community-based cohort. Methods We analyzed 756 MRI scans from 300 participants in the Framingham Heart Study (mean 2.52 scans per participant; range 2–4). Linear mixed-effects models were used to identify MRI features associated with diagnostic group (cognitively normal [CN], mild cognitive impairment [MCI], and dementia). Significant features (n=211) were entered into a longitudinal multivariate decomposition framework (ANOVA Simultaneous Component Analysis with Assorted Linear functions; ALASCA) to derive principal components (PCs) capturing patterns of structural change over time. Associations between PCs and plasma AD biomarkers (p-Tau 181 , total Tau(t-Tau), glial fibrillary acidic protein [GFAP], neurofilament light chain [NfL], amyloid-β40 [Aβ 40 ], and amyloid-β42 [Aβ 42 ]) were evaluated using multivariable mixed-effects models adjusted for age, sex, education, and APOE ε4 status. Cognitive measures and neuroethological measures in a subset were used to assess the functional relevance and biological associations, respectively. Results The first three PCs explained ∼95% of the variance within the modeled MRI feature (n=211) set (PC1: 75.8%, PC2: 13.8%, PC3: 5.4%). PC1 captured medial temporal atrophy involving hippocampal subfields and basolateral amygdala and was associated with worse cognition and higher plasma AD biomarkers. Neuropathological analyses showed stronger associations of PC1-related atrophy with AD-related tau pathology in the absence of concomitant TDP-43 pathology. In contrast, PC2 reflected diffuse cortical gray–white matter contrast alterations across association cortices and showed distinct associations with biomarkers and cognition compared to PC1, consistent with overlapping aging- and neurodegeneration-related processes. PC3 showed limited variance and no consistent associations. Conclusion Longitudinal MRI-derived components capture distinct patterns of brain structural change associated with neurodegeneration. Medial temporal trajectories are closely associated with AD and related dementia, whereas cortical alterations likely reflect mixed aging- and disease-related processes. Integration of structural MRI with plasma biomarkers provides complementary information on disease expression and heterogeneity, supporting multimodal approaches for disease characterization and risk stratification.