Association study
If the sample is affected by diseases that may influence IgE levels, we will include it in the category of ”any disease.” The same applies to medication usage, which will be consolidated under ”any medication.” The covariates included age, sex, PC1 to PC10, any disease status, and any medication. Bolt-LMM was used to perform the association analysis of IgE16. We regressed IgE by covariates and extracted the rank-based inverse normalize residuals as the input phenotype. No covariates were included during the association test. Multi-SNP-based conditional & joint association analysis using GWAS summary data (GCTA-COJO) was used to conduct conditional analysis and extract the independent signals17. We applied ANNOVAR to conduct cytoband and functional annotation. In order to replicate our results, we conducted meta-analysis of lead SNP ± 250kb region with IgE summary statistics from Tohoku Medical Megabank Organization (ToMMo)9. Meta-analysis was conducted via METAL.
HLA imputation & association analysis
We applied HLA genotype imputation with attribute bagging (HIBAG)18 to do the imputation of HLA with 4-digit resolution, including the HLA class I genes (HLA -A ,HLA-B , and HLA-C ) and class II genes (HLA-DPB1 ,HLA-DQA1 , HLA-DQB1 and HLA-DRB1 ). The detailed methods have been described previously19. We removed the 2-degree relative (n=4,071) before the association analysis. The R package - MiDasHLA20 was used to conduct the association analysis, and the covariates were the same as previously described. The HLA allele with the possibility < 0.8, frequency less than 0.01 and the p-value of HWE < 1x10-5 were removed from the analysis. We further tested the haplotype association of HLA-DQA1 and HLA-DQB1and serum IgE level.