Basic Science and Pathogenesis.

Background: Age is the largest risk factor for late-onset Alzheimer's Disease (LOAD). Although >80 genetic loci have been associated with LOAD, little is known about the age dependencies of these associations except the APOE region.

Method: We performed cross-ancestry and ancestry-specific genome-wide gene-age interaction and age-stratified association study using TOPMed-imputed genome-wide association study (GWAS) data from Alzheimer's Disease Genetics Consortium (ADGC) including 34,833 non-Hispanic Whites (NHW), 7,264 African Americans (AA), 3,232 East Asians (EA), and 2,024 Caribbean Hispanics (CH) aged 60 years and older. We chose an approximate median age-at-onset threshold of 75 to dichotomize participants into younger (N:N = 10, 905:11,624) and older (N:N = 9,982:14,842) groups. We first evaluated the association of LOAD with the interaction of SNP × age within each cohort using a linear mixed-effect model including a binary term for the age group, numeric age, sex, and the first 5 principal components (PCs) of ancestry. Results from each cohort were combined using a fixed-effect model to jointly estimate the regression coefficients of SNP and SNP × age terms, both within-ancestry and cross-ancestry analyses. We also performed an age-stratified analysis that included all controls regardless of age to increase power and using the same linear model without the interaction term. Results across cohorts were combined by meta-analysis using a fixed-effect model within-ancestry and a modified random-effect model across ancestry.

Result: In cross-ancestry joint meta-analysis, we identified 6 genome-wide significant (GWS) loci, 4 of which showed nominal evidence of age-dependent association with LOAD, including CD2AP (rs5876027, P = 2.33 × 10, P = 0.043), PICALM (rs583162, P = 2.34 × 10, P = 7.60 × 10), APOE (rs429358, P = 0, P = 1.58 × 10) and LILRA5 (rs1761461, P = 4.86 × 10, P = 0.032), for the first time. Within-ancestry joint meta-analysis identified novel associations with VXN (rs146432976, P = 2.29 × 10, P = 3.08 × 10) in AAs and PALM2-AKAP2 (rs67191673, P = 3.46 × 10, P = 0.013) in EAs. CR1 (rs6661489, P = 3.54 × 10) and TREM2 (rs75932628, P = 8.61 × 10) were associated with LOAD in NHWs but showed no evidence of age-dependent effects. In cross-ancestry age-stratified GWAS, BIN1, PICALM, and APOE region genes reached GWS in both strata, while MS4A6A was only significant in older group (rs7232, OR = 0.88, P = 4.23 × 10). A GWS association with PALM2-AKAP2 (rs67191673, OR = 1.94, P = 3.64 × 10) was identified in the EA younger group.

Conclusion: We identified several ancestry-specific age-dependent association with previously established and novel loci.