Gene‐neuroimaging brain model decodes neuropathological mechanisms in Alzheimer’s disease

Background Alzheimer’s disease (AD) is characterized by aberrations of several biological processes/factors (e.g., genes, misfolded proteins, atrophy). Investigating the interplay between these factors at multiple scales can advance our understanding of AD and facilitate the development of cost‐effective treatments while unifying biomarkers for early AD diagnosis and prevention. We propose a mathematical framework that models how gene expression (GE) modulates the interaction between six macroscopic imaging modalities in AD progression. The identified genes and pathways provide novel insight into AD pathology. Method 1) We pre‐processed 6 longitudinal neuroimaging data (beta‐amyloid and tau proteins, cerebral blood flow, glucose metabolism, R‐fMRI, and grey matter volume) of 157 healthy and 317 diseased subjects from ADNI. 2) Using GE data of 6 neurotypical brains from Allen Human Brain Atlas, we derived brain‐wide GE of 990 genes with leading roles in central biological functions. 3) We built a mathematical model that incorporates: (i) disease‐related longitudinal changes in neuroimaging data, (ii) GE‐modulated interactions between the different neuroimaging modalities, (iii) propagation of alterations resulting from (ii) across brain networks. 4) By evaluating the latent relationship between GE modulating effects and cognitive measures (MMSE, ADAS, executive function, memory score) of 127 subjects (converters) who advanced to subsequent stages in the AD continuum, causal genes and altered pathways in AD progression were identified. Result The model explains 64% of the common variance between GE‐dependent brain alterations and cognitive scores of the converters. We identified 89 disease‐driving genes, including some notable previously identified AD genetic determinants: APPB2, TOP2A, CLU, BACE1 and CD44. The biological factors causally modulated by each gene, and factor alterations resulting from the modulation were also reported; e.g., our model reveals that APPB2 gene modulates beta‐amyloid to cause longitudinal alteration of beta‐amyloid, while CLU could interact with vascular flow to alter tau protein (Fig. 1). We also identified 54 likely altered pathways in AD evolution. Conclusion We developed a mathematical model to investigate the influence of gene expression on multifactorial alterations of biological processes in AD progression. Our results are strongly consistent with previously reported studies and provided further insight into the molecular mechanism underlying AD progression.