Conditional meta-analysis stratifying on detailed HLA genotypes identifies a novel type 1 diabetes locus around TCF19 in the MHC

Abstract

The human leukocyte antigen (HLA) class II genes HLA-DRB1, -DQA1 and -DQB1 are the strongest genetic factors for type 1 diabetes (T1D). Additional loci in the major histocompatibility complex (MHC) are difficult to identify due to the region's high gene density and complex linkage disequilibrium (LD). To facilitate the association analysis, two novel algorithms were implemented in this study: one for phasing the multi-allelic HLA genotypes in trio families, and one for partitioning the HLA strata in conditional testing. Screening and replication were performed on two large and independent datasets: the Wellcome Trust Case-Control Consortium (WTCCC) dataset of 2,000 cases and 1,504 controls, and the T1D Genetics Consortium (T1DGC) dataset of 2,300 nuclear families. After imputation, the two datasets have 1,941 common SNPs in the MHC, of which 22 were successfully tested and replicated based on the statistical testing stratifying on the detailed DRB1 and DQB1 genotypes. Further conditional tests using the combined dataset confirmed eight novel SNP associations around 31.3 Mb on chromosome 6 (rs3094663, p = 1.66 × 10(-11) and rs2523619, p = 2.77 × 10(-10) conditional on the DR/DQ genotypes). A subsequent LD analysis established TCF19, POU5F1, CCHCR1 and PSORS1C1 as potential causal genes for the observed association.

Fig. 1

Summary of analytical plan. a Processing of WTCCC dataset. b Processing of T1DGC dataset. c Procedures of statistical and LD analyses

Fig. 2

p values of the WTCCC dataset conditional on the HLA-DRB1-DQB1 genotypes based on the stratified Cochran-Armitage trend test. The classical MHC subregions and some relevant genes are included for reference. Position is along chromosome 6, NCBI genome build 36.3

Fig. 3

p values of the T1DGC dataset conditional on the HLA-DRB1-DQA1-DQB1 haplotypes based on Mantel–Haenszel test. The classical MHC subregions and some relevant genes are included for reference. Position is along chromosome 6, NCBI genome build 36.3

Fig. 4

LD between the 22 replicated SNPs. The figure in each box gives the first two decimal places of the r ² value. LD blocks identified by the confidence interval method were outlined. IDs in blue indicate SNPs that pass screening with WTCCC dataset, whereas IDs in green indicate SNPs that pass screening with T1DGC dataset. All SNPs showing novel associations are highlighted in bold (color figure online)

Fig. 5

p values conditional on HLA-DRB1-DQB1 together with HLA- a A, b B, c C and d DPB1, respectively. p values of a–c are generated by applying stratified Cochran–Armitage trend test on the combined dataset, whereas p values in d are generated by applying Mantel–Haenszel test on the phased T1DGC haplotypes. Significance levels were indicated in the appropriate regions under test. The classical MHC subregions and some relevant genes are included for reference. Position is along chromosome 6, NCBI genome build 36.3

Fig. 6

LD between the eight replicated SNPs with novel T1D associations and the missense SNPs within a TCF19, b CCHCR1 and c PSORS1C1. Only SNPs available in the CEU HapMap dataset were studied and any redundant SNPs were removed. All missense SNPs are highlighted in bold and red. Boxes of D′ = 1 are outlined in yellow (color figure online)