Patient stratification by genetic risk in Alzheimer’s disease is only effective in the presence of phenotypic heterogeneity

Abstract

Case-only designs in longitudinal cohorts are a valuable resource for identifying disease-relevant genes, pathways, and novel targets influencing disease progression. This is particularly relevant in Alzheimer’s disease (AD), where longitudinal cohorts measure disease “progression,” defined by rate of cognitive decline. Few of the identified drug targets for AD have been clinically tractable, and phenotypic heterogeneity is an obstacle to both clinical research and basic science. In four cohorts (n = 7241), we performed genome-wide association studies (GWAS) and Mendelian randomization (MR) to discover novel targets associated with progression and assess causal relationships. We tested opportunities for patient stratification by deriving polygenic risk scores (PRS) for AD risk and severity and tested the value of these scores in predicting progression. Genome-wide association studies identified no loci associated with progression at genome-wide significance (α = 5×10−8); MR analyses provided no significant evidence of an association between cognitive decline in AD patients and protein levels in brain, cerebrospinal fluid (CSF), and plasma. Polygenic risk scores for AD risk did not reliably stratify fast from slow progressors; however, a deeper investigation found that APOE ε4 status predicts amyloid-β and tau positive versus negative patients (odds ratio for an additional APOE ε4 allele = 5.78 [95% confidence interval: 3.76–8.89], P<0.001) when restricting to a subset of patients with available CSF biomarker data. These results provided no evidence for large-effect, common-variant loci involved in the rate of memory decline, suggesting that patient stratification based on common genetic risk factors for progression may have limited utility. Where clinically relevant biomarkers suggest diagnostic heterogeneity, there is evidence that a priori identified genetic risk factors may have value in patient stratification. Mendelian randomization was less tractable due to the lack of large-effect loci, and future analyses with increased samples sizes are needed to replicate and validate our results.