No added value of Sex and APOE*4 stratification for Alzheimer’s Disease Genetic Risk Scores

Abstract

AbstractBackgroundGenetic risk scores (GRS), be it polygenic risk scores (PRS) or polygenic hazard scores (PHS), have shown promise to support genetic risk prediction and clinical trial recruitment for Alzheimer’s disease (AD). A recent study further suggested that sex‐matched AD PHS (e.g., male‐to‐male) outperform sex‐mismatched PHS (e.g., male‐to‐female), while no benefit was observed for PRS. Similar benefits may be expected for matching by APOE*4 positivity status, but this question has remained unexplored. The prior sex‐stratified work neither evaluated whether full sample GRS (males+females) outperform sex‐matched GRS, nor compared polygenic (including many variants regardless of significance) to oligogenic (including fewer more significant variants) effects. Here, we performed the largest‐to‐date sex‐stratified, and first APOE*4‐stratified, evaluation of AD GRS.MethodsThe study design is detailed in Figure‐1: genome‐wide association studies (GWAS) were performed in a Discovery cohort (LMM‐BOLT v.2.3.5) and GRS were constructed and evaluated in an independent Replication cohort (PRSice‐2 & R v3.6). All analyses performed multiple linear regression on an AD‐age score that models resilience to age‐related risk for AD (Figure‐2). Models adjusted for sex, APOE*4/APOE*2 dosage (rendering GRS estimates/effects independent of APOE), the first five genetic principal components (PC‐AiR; GENESIS; R v3.6), and array/sequencing center. GRS were evaluated at different P‐value thresholds to elucidate oligogenic versus polygenic effects, while all GRS with other covariates were combined to evaluate predictive performance in the Replication.ResultsStratified discovery GWAS are shown in Figure‐3. GRS tended to perform best at more stringent P‐value thresholds (i.e., oligogenic) and GRS derived from the full‐sample discovery virtually always outperformed stratified GRS (Figure‐4‐5). Predictive performance was similarly the highest for GRS derived from the full‐sample discovery (Table‐1).ConclusionsOur results were consistent with emerging literature supporting that AD is oligogenic and showed that the performance of sex‐ or APOE*4‐stratificatied GRS was outweighed by using full‐sample derived GRS. While not shown here, we made consistent observations when using a case‐control model (thus equivalent to PRS) without age adjustment. These findings have important implications to guide further efforts at translating AD GRS into clinical practice.aging after age 90.