Longitudinal genome wide association studies in Alzheimer’s disease

Abstract

AbstractBackgroundA large number of subjects are generally required for the genome wide association studies (GWAS) in complex traits such as Alzheimer’s disease (AD)1,2. Here we incorporated longitudinal data in AD‐GWAS to improve the statistical power for the identification of AD‐associated genetic variants with a limited sample‐size. Time‐varying genetic contributions towards AD could also be modeled and captured in this longitudinal GWAS.Method1,877 subjects from the ADNI database3 with genotyping data available were included in this analysis. Subjects’ genotype data were imputed to the Haplotype Reference Consortium using the Michigan Imputation Server4, with 9,573,130 single nucleotide polymorphisms (SNPs) remaining after the quality control step. Subjects’ longitudinal diagnosis at each visit were obtained from ADNI, providing 10,832 phenotypes for 1,877 subjects. Subjects’ diagnoses were further binarized into the normal and diseased groups. We applied the retrospective varying coefficient mixed model association test (RVMMAT) to detect time‐varying genetic effect on this longitudinal binary phenotype5. Briefly, dynamic genetic effect was modeled using smoothing splines and estimated by maximizing a double penalized quasi‐likelihood function via a retrospective approach. Subjects’ sex, age at each diagnosis, and the first 5 principal components of whole genome were included as covariates. A categorical variable for genotyping platforms was included as an additional covariate.ResultRVMMAT showed no evidence of inflation in the quantile‐quantile (Q‐Q) plot, with an inflation factor (lambda) of 0.99 (Fig. 1A). 45 SNPs reached genome‐wide significance after false discovery rate correction for multiple comparisons (Fig. 1(B)). Among these genetic variants, 35 SNPs were clustered at the APOE, APOC1, TOMM40, and NECTIN2 genes on chromosome 19, and 6 SNPs were associated with SLAIN2 gene on chromosome 4.ConclusionWe demonstrate that RVMMAT could boost the statistical power in AD‐GWAS with a limited sample‐size. We expect this method to benefit the identification of genetic variants associated with pathological or clinical biomarker‐based longitudinal phenotypes in AD.