Decision tree analysis of genetic risk for clinically heterogeneous Alzheimer’s disease

BACKGROUND Heritability of Alzheimer's disease (AD) is estimated at 74% and genetic contributors have been widely sought. The ε4 allele of apolipoprotein E (APOE) remains the strongest common risk factor for AD, with numerous other common variants contributing only modest risk for disease. Variability in clinical presentation of AD, which is typically amnestic (AmnAD) but can less commonly involve visuospatial, language and/or dysexecutive syndromes (atypical or AtAD), further complicates genetic analyses. Taking a multi-locus approach may increase the ability to identify individuals at highest risk for any AD syndrome. In this study, we sought to develop and investigate the utility of a multi-variant genetic risk assessment on a cohort of phenotypically heterogeneous patients with sporadic AD clinical diagnoses. METHODS We genotyped 75 variants in our cohort and, using a two-staged study design, we developed a 17-marker AD risk score in a Discovery cohort (n = 59 cases, n = 133 controls) then assessed its utility in a second Validation cohort (n = 126 cases, n = 150 controls). We also performed a data-driven decision tree analysis to identify genetic and/or demographic criteria that are most useful for accurately differentiating all AD cases from controls. RESULTS We confirmed APOE ε4 as a strong risk factor for AD. A 17-marker risk panel predicted AD significantly better than APOE genotype alone (P < 0.00001) in the Discovery cohort, but not in the Validation cohort. In decision tree analyses, we found that APOE best differentiated cases from controls only in AmnAD but not AtAD. In AtAD, HFE SNP rs1799945 was the strongest predictor of disease; variation in HFE has previously been implicated in AD risk in non-ε4 carriers. CONCLUSIONS Our study suggests that APOE ε4 remains the best predictor of broad AD risk when compared to multiple other genetic factors with modest effects, that phenotypic heterogeneity in broad AD can complicate simple polygenic risk modeling, and supports the association between HFE and AD risk in individuals without APOE ε4.