73 of antibody-bound peptide repertoire variability in LLD was used as a number of independent tests (708), obtaining a study-wide threshold of 5.67x10-11. For each peptide’s summary statistics we extracted genome-wide significant associations (p<5x10-8) for clumping. We clumped variants in windows of 1,000 Kb if they had a minimal R2 (computed from LLD genotypes) of at least 0.1 using PLINK. Leading variants of each clump were then annotated using the Ensembl Variant Effect Predictor and the grCh37 human build.45 LD between our identified leading variants and other publicly reported variants was estimated in the CEU population from the 1,000 genomes using the LDlink webtool.46,47 HLA imputation and association The chromosome 6 regionwith 25–34Mb that contains theMHC genes was extracted. Imputation of the HLA region, including HLA alleles, polymorphic amino acids, SNP variants and indels, was then performed using SNP2HLA (v2) with the Type 1 Diabetes Genetics Consortium (T1DGC) reference panel (2,767 unrelated European descent individuals) and HLA Reference Panel.48,49 Next, we combined both imputed and genotyped SNPs, HLA alleles and amino acid variants, resulting in a total of 8,926 variants. Variants with MAF < 0.05 and imputation quality score (INFO) < 0.5 were removed before association. HLA to peptide association was performed using linear models in 1,175 participants, while controlling for age, sex, PhIP-Seq plate and disease subtypes (Crohn’s disease/ulcerative colitis, only specific to IBD cohort). Summary statistics from both datasets were further meta-analyzed using a fixed-effects model in PLINK v1.9. The statistical significance threshold was determined by dividing the usual P-value 0.05 threshold level by the number of independent features tested (66 PCs were needed to reach 90% of HLA feature variability in LLD, while 708 PCs were needed to capture 90% of the peptide variability, resulting in 46,728 independent tests), resulting in a threshold of 1x10-6. FDR was estimated using the Benjamini-Hochberg method.35 Modeling of peptide presentation in HLA complexes To explore whether HLA–peptide associations potentially point to HLA-II ability to display a specific peptide, we performed computational modeling of the complex–peptide interaction. The protein sequences of DR3, DR4, DR14, DR15 andDQ2were obtained from the IPDIMGT/ HLA database50 and aligned against the entire Protein Data Bank database using pBLAST. Protein structures displaying a 100% of amino acid identity with the HLA-II database sequences were chosen to build the peptide binding modes. Those structures correspond to the HLA complexes DR3:7N19, DR4:1D5M, DR14:6ATF, DR15:1YMM, DQ2:6PX6 and DQ8:2NNA. Proteins other than HLA-II, water molecules and heteroatoms were removed from the structures prior to modeling. The NetMHCIIpan-4.051 server was then used to predict peptide binding to the corresponding associated HLA alleles: DRB1*1501 for Lactobacillus phage LfeInf; DRB1*0301, DQA1*0501DQB1*0201 and DRB1*1401 for Streptococcus agalactiae C5a peptidase; and DRB1*0401 and DQA1*03-DQB1*0302 for Human mastadenovirus minor core protein. The DRB1*1401 for Streptococcus agalactiae C5a peptidase was selected as a no binding negative control for these experiments. Following the identification of the peptide core by NetMHCIIpan-4.0, the protein structures and identified peptide core were submitted to HPEPDOCK Server for peptide– Determinants of the human antibody epitope repertoire
RkJQdWJsaXNoZXIy MjY0ODMw