Exploring the shared genetic architecture of modifiable risk factors and related endophenotypes of Alzheimer’s disease: A genomic SEM study

AbstractBackgroundThe mechanisms by which modifiable risk factors influence Alzheimer’s disease (AD) remain incompletely understood. Previous work using linkage disequilibrium score regression (LDSC) indicates that many of the modifiable risk factors have high genetic correlation with one another. Modelling this shared genetic architecture between risk factors, associated endophenotypes and AD could yield important insights into shared aetiological pathways.MethodWe used genomic structural equation modelling (SEM) to model the shared genetic architecture between potentially modifiable risk factors (including anxiety, depression, diabetes, education, hearing loss, hypertension, obesity, sleep, smoking) and related endophenotypes (including inflammatory markers and metabolites). Genomic SEM uses summary statistics from genome‐wide association studies (downloaded from leading consortia, the Neale lab group, CTG lab group and UK Biobank) to calculate the genetic covariance between traits using multivariable LDSC. We then used exploratory factor analysis (EFA) to model the patterns of shared genetic architecture into latent constructs.ResultMultivariable LDSC showed multiple correlations between traits. EFA showed that AD shares genetic covariance with circulating metabolites and cytokines, but not modifiable risk factors. A second factor showed loadings of age‐related hearing loss with cytokines and metabolites distinct from their variance with Alzheimer’s disease. A third factor comprised the majority of other risk factors including depression, insomnia and smoking, but not diabetes or obesity, which did not share significant covariance with any other AD‐related traits.ConclusionWe found high levels of shared genetic architecture between AD risk factors that were not shared with AD itself. This could suggest that rather than a direct causal effect on AD, risk factors exert their influence via a common intermediary (e.g. pro‐inflammatory states and/or cognitive reserve). Further work is required to establish model fit statistics, identify common SNPs related to these factors and conduct multivariable Mendelian randomisation to measure causal pathways that could lead to AD. This approach could provide important biological insights into the mechanisms underpinning the shared genetic architecture of AD risk factors, and explore how variation at these loci functionally impacts on AD pathophysiology, elucidating aetiological pathways, and ultimately prioritising targets for drug discovery and efforts to prevent AD.