Abstract
We examine, across nine human brain regions, the spectrum of genome-wide gene expression associations with Alzheimer’s disease (AD) neuropathology using 1,746 human individuals from three AD studies. We introduce a new computational approach, DECODER, that leverages discrepancies across different brain regions or different studies in order to identify robust expression markers for complex neuropathological phenotypes. Our computational evaluation experiments demonstrate: (1) the possibility of performing meta-analysis in the highly challenging AD setting where datasets involve study-specific confounders or brain region-specific biological processes, (2) DECODER’S potential as a general meta-analysis framework widely applicable to various diseases (e.g., AD, cancer) or phenotypes (e.g., neuropathology, survival), and (3) provide new insights into the similarity of brain regions in terms of expression associations with AD hallmarks. We further extend these computational advances through in vivo validation of novel genes using a transgenic Caenorhabditis elegans model expressing AD-associated Aβ. Our approach yields several novel genetic modifiers of Aβ toxicity and pinpoints Complex I of the mitochondrial electron transport chain (mETC) as a critical mediator of proteostasis and a promising potential pharmacological avenue toward treating AD.