Background: Alzheimer’s disease (AD) pathogenesis arises from multifactorial mechanisms that include dysregulation of microRNAs (miRNAs). As small and highly conserved non-coding RNAs that post-transcriptionally regulate gene expression and remain s...
Background: Alzheimer’s disease (AD) pathogenesis arises from multifactorial mechanisms that include dysregulation of microRNAs (miRNAs). As small and highly conserved non-coding RNAs that post-transcriptionally regulate gene expression and remain stable in peripheral blood, miRNAs have emerged as candidates for prognostic and diagnostic biomarker development. In parallel, cerebral amyloid angiopathy (CAA) and AD share amyloid-β pathology but in distinct anatomical compartments—vascular walls in CAA versus brain parenchyma in AD. Elucidating human CAA mechanisms is therefore essential for anticipating and mitigating complications of Aβ-targeting immunotherapies; yet, despite its clinical relevance, these mechanisms remain incompletely defined.
Methods: In study 1, we constructed miRNA co-expression networks from brain tissue profiles in the Religious Orders Study and Rush Memory and Aging Project (ROS/MAP; N = 702) to identify modules associated with AD-related neuropathological traits, diagnosis of AD dementia, and global cognition. We then prioritized hub miRNAs within AD-relevant modules and evaluated their associations with AD-related neuropathological traits, diagnosis of AD dementia, and global cognition. After selecting target genes of the hub miRNAs, we performed pathway-based enrichment analysis. For replication, we conducted a consensus miRNA co-expression network analysis integrating ROS/MAP with an independent Gene Expression Omnibus (GEO) dataset (N = 16). Finally, we assessed the classification performance of hub miRNAs for AD dementia using machine-learning models. In study 2, we analyzed 64 plasma miRNAs measured in 145 participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), including 74 patients with probable AD and 71 cognitively normal (CN) older adults. We applied principal component analysis (PCA) with factor rotation to reduce dimensionality, identified AD-associated principal components (PCs) and their key miRNAs with factor loadings higher than 0.80. Associations of these PCs and key miRNAs with amyloid/tau/neurodegeneration (A/T/N) biomarkers and cognition were then evaluated. After identifying the candidate target genes of key miRNAs, we performed pathway enrichment analysis. We conducted mediation analyses to assess the indirect effect of the miRNA–A/T/N associations on AD and cognition. Finally, we assessed the classification performance of key miRNAs for AD using machine-learning models. In study 3, we integrated proteomic, neuropathologic, and clinical data from two independent cohorts using postmortem brain proteomics from ROS/MAP (N = 400) and CSF proteomics from ADNI (N = 59). In ROS/MAP, we conducted proteome-wide association analyses with CAA severity, AD-related neuropathological traits, diagnosis of AD dementia, and global cognition. We then constructed protein co-abundance network to define CAA-associated modules and hub proteins, followed by pathway enrichment to interpret biological significance. We performed cell-type deconvolution and diffusion map–based pseudotime modeling to infer cellular origins and temporal dynamics of CAA-related proteins. We further assessed the classification performance of CAA-associated proteins for CAA using machine-learning models. Finally, we performed a replication analysis in an independent ADNI cohort.
Results: In study 1, network analysis identified a module that was significantly associated with diagnosis of AD dementia and global cognition. Within this module, five hub miRNAs (miR-129-5p, miR-433, miR-1260, miR-200a, and miR-221) demonstrated significant associations with diagnosis of AD dementia and/or AD-related neuropathological traits, with miR-129-5p showing the strongest and most consistent effects across all phenotypes. Pathway-based enrichment analysis of target genes regulated by these hub miRNAs revealed significant biological processes including ErbB, AMPK, MAPK, and mTOR signaling. Consensus network analysis further identified two AD dementia-associated modules and two hub miRNAs (miR-129-5p and miR-221). Finally, machine-learning analysis demonstrated that the inclusion of the five hub miRNAs improved AD classification performance by 6.3% compared with a baseline model of age, sex, and apolipoprotein E (APOE) ε4 carrier status, increasing the area under the curve (AUC) to 0.807. In study 2, PCA identified one PC that was significantly associated with AD. The PC was also associated with CSF p-tau levels, hippocampal volume, and cognition. Two key miRNAs (miR-423-5p and miR-92a-3p) within the PC were significantly associated with AD. Lower levels of both miR-423-5p and miR-92a-3p were associated with reduced hippocampal volume and poorer cognition, while lower miR-423-5p levels were additionally associated with greater brain amyloid deposition. Pathway-based enrichment analysis highlighted several significant biological processes, including memory, protein phosphorylation, and the phosphatidylinositol-3-phosphate biosynthetic process. Mediation analyses indicated that miR-423-5p, but not miR-92a-3p, had indirect effects on AD and memory performance through brain amyloid deposition and hippocampal atrophy. Finally, machine-learning models demonstrated that inclusion of the two key miRNAs improved the classification performance of demographic information-based models for AD. In study 3, integrative analyses identified 30 proteins associated with CAA severity, enriched for biological processes including canonical Wnt signaling, lipoprotein metabolism, and protein tetramerization that were distinct from those observed in Consortium to Establish a Registry for Alzheimer’s Disease (CERAD)- and Braak-related protein profiles. Protein co-abundance network analysis further identified two CAA-associated modules that showed no significant associations with CERAD scores or Braak stages. Cell-type deconvolution and pathway enrichment indicated a microglia-enriched module linked to vesicle-mediated transport and an astrocyte-enriched module linked to potassium ion transmembrane transport. Diffusion map–based pseudotime modeling of the 30 CAA-associated proteins inferred trajectories that increased with CAA severity and captured broader AD-related neuropathological traits, including CERAD and Braak pathology and global cognition. A three-protein panel (RCSD1, FRZB, and AKAP6) selected by permutation feature importance, when combined with age, sex, and APOE ε4 carrier status, improved AUC for CAA classification from 0.585 to 0.721 using machine-learning models. Finally, in ADNI, 11 of 23 discovery proteins with available measurements replicated for CAA severity with concordant effect directions.
Conclusion: Integrative network and machine-learning analyses of brain tissue–based miRNA profiles identified signatures, most notably miR-129-5p, as associated with AD dementia, its related neuropathological traits, and cognition, thereby refining our understanding of AD pathogenesis and improving diagnostic classification performance. Complementing these tissue findings, plasma miR-423-5p and miR-92a-3p were linked to brain amyloid burden and cognitive decline, underscoring their mechanistic relevance and supporting their potential as minimally invasive, blood-based biomarkers for prognosis and diagnosis. In parallel, our proteomic profiling and integrative analyses delineated a molecular architecture of CAA that is distinct from classical AD pathology, nominating specific candidate biomarkers and pathways that implicate vascular and glial mechanisms. Together, these results establish complementary biomarker frameworks that can inform biologically grounded diagnostics and guide targeted therapeutic strategies, a priority made increasingly urgent by the clinical adoption of Aβ-directed treatments.