Individual brain similarity networks across aging and the Alzheimer's disease continuum

Abstract

Aging is accompanied by gradual changes in brain structure and cognition, but these changes do not unfold in the same way for every individual. Some people maintain cognitive function into late life, whereas others experience decline in memory, executive function, processing speed, or other cognitive domains. This variability makes it difficult to distinguish typical aging from early signs of neurodegenerative disease. This distinction is important because age-related brain changes may interact with pathological processes that eventually contribute to neurodegenerative diseases such as Alzheimer's disease (AD). AD is one of the most common and severe causes of dementia. Its pathological processes, including amyloid-beta (AB) deposition and tau accumulation, begin many years before clinical symptoms appear. During this long preclinical phase, individuals may remain cognitively normal while subtle biological changes are already developing in the brain. This creates a major challenge for early detection, disease monitoring, and prevention trials, where biomarkers must capture early, distributed, and person-specific alterations before overt atrophy or clinical impairment becomes evident. Current biomarkers provide important information about aging and AD, but each captures only part of the disease process. Regional magnetic resonance imaging (MRI) measures, such as gray matter volume and cortical thickness, are sensitive to tissue loss but may miss coordinated structural changes across the brain. Anatomical and functional connectivity measures derived from diffusion- weighted imaging (DWI) and resting-state functional MRI (rs-fMRI) provide network-level information, but can be affected by acquisition and preprocessing challenges. Fluid biomarkers reflect molecular pathology but do not describe the spatial organization of brain changes. Conventional positron emission tomography (PET) measures often summarize amyloid and tau burden within predefined regions, potentially overlooking inter-individual variability in pathological distribution. These limitations suggest that early aging- and AD-related changes may be better captured by approaches that measure individualized deviations in inter-regional brain organization. The overarching aim of this thesis was therefore to validate and extend a perturbation-based framework for studying brain aging and AD. In this framework, each individual is added to a reference population, and the resulting change in the normative inter-regional covariance pattern is quantified. This change defines the individual's brain similarity network, with each connection representing a person- specific deviation in the similarity between brain regions. By capturing these deviations, the framework was designed to detect subtle and early changes in brain organization with greater sensitivity than the conventional imaging and fluid biomarkers described above. To address this aim, the thesis was organized into four studies that move from normative brain aging, to methodological comparison, to preclinical AD, and finally to molecular pathology across the AD continuum. Study I examined whether the perturbation-based brain similarity framework captures age-related variation in brain organization beyond established connectivity measures. Individual brain similarity networks were derived from T1-weighted MRI in cognitively normal adults and compared with anatomical connectivity from DWI and functional connectivity from rs-fMRI. Using graph neural network models and graph- theoretical analyses, brain similarity networks predicted age and cognitive performance more accurately than anatomical and functional connectivity, detected age-related network alterations earlier in adulthood, and showed associations with cortical cytoarchitecture. Given that individual gray matter networks can be constructed using several other MRI-based approaches, study II compared perturbation-based similarity networks with correlation-based networks, Jensen-Shannon divergence networks, Mahalanobis distance-based networks, and conventional regional gray matter volume. These measures were evaluated in relation to age-related trajectories, cognition, cardiovascular risk burden and inter-individual heterogeneity. Perturbation-based networks showed the strongest sensitivity to aging, captured widespread nonlinear structural variation, were the only network measure consistently associated with cognitive performance and cardiovascular risk, and revealed increasing heterogeneity in later life. Building on these aging studies, Study III applied the perturbation-based brain similarity framework to preclinical AD in cognitively normal individuals stratified by amyloid status. Brain similarity was compared with cortical thickness, volumetric MRI measures, and cerebrospinal fluid (CSF) and plasma biomarkers for predicting longitudinal cognitive decline and diagnostic conversion. Across cohorts, brain similarity predicted decline in multiple cognitive domains and outperformed conventional structural imaging measures, while providing complementary information to fluid biomarkers. It also achieved higher accuracy in predicting conversion to mild cognitive impairment (MCI) or AD, and its spatial distribution was associated with cytoarchitectural organization. Finally, Study IV extended the perturbation-based framework beyond structural MRI to molecular pathology. Individual tau and amyloid connectomes were constructed from molecular PET imaging across the AD continuum and evaluated for subject identification, disease tracking, diagnostic discrimination, and cognitive prediction. These molecular connectomes acted as subject-specific fingerprints, increased across disease stage and longitudinal follow-up, and outperformed conventional PET composite standardized uptake value ratio (SUVR) and staging measures. Their susceptibility to AD was further linked to transcriptomic pathways including apoptosis, pyrimidine metabolism, and histone acetylation. Together, these findings support a shift from region-centric biomarkers toward individualized network-based representations of brain organization. Brain similarity and molecular connectome mapping capture distributed, biologically meaningful deviations that emerge during normal aging, become clinically relevant in preclinical AD, and progress across the disease continuum. This framework may improve early detection and disease monitoring and support the development of personalized approaches for AD research and clinical trials. List of scientific papers I. Zufiria-Gerbolés B, Sun J, Pineda J, Volpe G, Mijalkov M, Pereira JB. Similar minds age alike: an MRI similarity approach for predicting age-related cognitive decline. npj Aging. 2026;12:39. https://doi.org/10.1038/s41514-026-00345-1 II. Sun J, Mijalkov M, Volpe G, Passaretti M, Zufiria-Gerbolés B, Zhao H, Tiunn I, Pereira JB. Individual gray matter networks predict age- related changes across the lifespan. [Manuscript] III. Sun J, Zufiria-Gerboles B, Passaretti M, Volpe G, Mijalkov M, Pereira JB, for the Alzheimer's Disease Neuroimaging Initiative. Tracking early cognitive decline in preclinical AD with brain MRI similarity. Alzheimer's & Dementia. 2026;22:e71170. https://doi.org/10.1002/alz.71170 IV. Xu Z, Mijalkov M, Sun J, Chang Y-W, Sala A, Volpe G, Severino M, Veronese M, Garcia-Ptacek S, Pereira JB, for the Alzheimer's Disease Neuroimaging Initiative. Mapping individual molecular connectomes in Alzheimer's disease. Alzheimer's & Dementia. 2026;22:e71310. https://doi.org/10.1002/alz.71310