Using polygenic risk scores to assess the importance of microglia in Alzheimer’s disease

Abstract

AbstractBackgroundAlzheimer’s disease (AD) is a highly polygenic disorder; this has been demonstrated by the identification of a large number of risk loci from genome‐wide association studies (GWAS), and the ability of polygenic risk scores (PRS) to differentiate between AD cases and controls with high prediction accuracy (AUC = 78.2%) [Escott‐Price et al. 2015]. Many AD risk alleles have been shown to be located at open chromatin sites in immune cells, specifically microglia [Tansey et al. 2018].MethodA whole genome PRS was produced for individuals in the Genetic and Environmental Risk for AD (GERAD) data (N = 5,113) [Harold et al. 2009]. The score was weighted using effect sizes from the Kunkle GWAS [Kunkle et al. 2019], excluding GERAD individuals, and only single nucleotide polymorphisms (SNPs) with a p‐value less than 0.5 are included in the score. Linkage disequilibrium between SNPs was removed using standard pruning methods. In addition to a whole genome PRS, which uses all available SNPs, a microglia‐specific PRS was also produced, using SNPs which reside within the microglia regions defined by Gosselin et al. 2017. We also produced microglia‐specific PRSs with microglia genes defined from alternative sources for comparison.ResultWe compare the association between the whole genome and microglia‐specific PRSs, which show a moderate correlation (r = 0.24, p = 7 × 10−10). The prediction accuracy of both PRSs was determined; preliminary results show that prediction accuracy does not decrease dramatically when comparing whole genome and microglia PRS (AUC decreases by 3% and R2 drops to 0.04 from 0.08). We also determine the proportion of prediction accuracy and SNP‐based heritability explained by the microglial component of the polygenic risk score.ConclusionAD research has been directed towards microglia as it has been shown to be an important cell in neurodegenerative disorders [Young et al. 2019]. The decrease in prediction accuracy observed is due to the reduced number of SNPs; the goal to improving prediction accuracy is reducing SNP numbers whilst retaining heritability [Wray et al. 2019]. Microglial PRS show improved signal‐to‐noise ratio by reducing SNP numbers and are therefore a valuable approach to experimental design and cell line selection.