Factorization of Alzheimer's disease genetic risk influences allow patient stratification, predicting disease onset, cognitive decline, and cell-type specific responses

Abstract

Late-onset Alzheimers disease (AD) is a complex and heterogeneous neurodegenerative disease with significant genetic components implicated in at least 97 loci from AD genome-wide association studies. While various distinct AD subtypes have been identified based on brain or CSF molecular profiling, contribution of the genetic signatures in distinguishing the AD subtypes is lacking. Here, we leveraged large snRNA-seq postmortem brain data with an empirical Bayes matrix factorization (EBMF) approach to study common effects of 197 AD risk variants on neuronal and glial cell transcriptome, enabling factor-based polygenic score (fPGS) and patient clustering based on their functional genetic profiles. We confirmed that each factor captures specific AD risk variant influences on cell types and known AD-associated biological processes, such as mitochondrial activity, endo-lysosomal activity, mRNA processing, neuroinflammation, or calcium signaling. Further, we found that most fPGS were predicting a certain neuropathological or AD-associated molecular condition. Notably, fPGS3 predicts somatic mutation burden in excitatory neurons, and fPGS7 predicts epigenome erosion in excitatory neurons associated with lipid transport disorders, increased mitochondrial activity, and increased Tau pathology. Finally, unsupervised clustering analysis of individuals with mild cognitive impairment and AD based on their fPGS profiles enable us to identify seven clusters, which are differentiated by the APOE genotype. Among them, five groups were subdivided within APOE3 homozygotes, which remarkably predicted either the disease severity or the disease onset with up to 6 years differences among groups. Resilient groups were associated with reduced matrisome and reactivity in astrocytes and increased cholesterol metabolic pathways in oligodendrocytes and OPCs. Overall, the analyses confirmed the ability of EBMF to stratify AD risk variant influences into molecularly and clinically coherent factors that allow genetic predictions of particular cell types and alterations of biological processes impacting the disease.