Multi‐omics characterization of brain‐based pseudotime estimates for Alzheimer’s disease progression

Abstract Background The cascade of molecular changes that leads to Alzheimer’s disease (AD) onset and progression remains incompletely understood. Recently, researchers have utilized manifold learning techniques to construct pseudo‐temporal models of the disease using brain transcriptomics data and quantify the progression of individuals along this trajectory using pseudotime (Mukherjee et al. Nat Commun 2020;11:5781). Downstream analyses provided insights into potentially disease‐driving pathways, including links to mitochondrial dysfunction. Here, we embedded these models into a multi‐omics context to enable a more comprehensive molecular characterization of disease progression. Method The AD Atlas ( https://adatlas.org/ ) integrates multi‐scale molecular data from different studies and cohorts, including omics QTLs, correlation networks, differential expression data, as well as omics associations with AD and endophenotypes. Here, we used the AD Atlas to annotate metabolites that were significantly associated with brain‐based pseudotime estimates. Metabolite (n = 667) levels were measured in brain tissue samples (dorsolateral prefrontal cortex) from 154 female ROS/MAP participants using untargeted metabolomics and tested for association using linear regression. We used the resulting set of significant metabolites as input for the AD Atlas to extract a multi‐omics context network augmented with associations to AD. Subsequently, we applied pathway enrichment analysis to derive overrepresented biological processes potentially involved in disease progression. Result In total, 89 of the 667 metabolites showed a significant association with pseudotime after Bonferroni adjustment, 34 of which could be mapped to the AD Atlas database. The resulting multi‐omics network contained a total of 619 genes, 197 metabolites and links to 12 AD‐related phenotypes, including CSF amyloid pathology, brain glucose uptake measured by FDG‐PET and cognitive measures. Five of the 34 metabolites and nearly one‐third of the genes contained in the network showed significant associations with AD, with transcriptional changes most pronounced in the temporal cortex (n = 193). Enrichment analysis revealed functional links to neurotransmission and bioenergetics, pathways previously implicated in the pathogenesis of AD. Conclusion By using metabolites as proxies for transcriptome‐derived pseudotime, we were able to investigate the molecular underpinnings of AD progression in a multi‐omics context. Our analysis provides further molecular evidence for pathways implicated in AD and emphasizes the potential of such an approach for future studies.