Eight amino acid positions in five HLA class I and II genes explain the MHC association to type 1 diabetes risk. X. Hu1,2,3,4,5,6, B. Han1,2,3,4,5, S. Onengut-Gumuscu7, W. Chen7, A. J. Deutsche1,2,3,4,5,6, T. L. Lenz2,5, P. I. W. de Bakker8,9, S. S. Rich7, S. Raychaudhuri1,2,3,4,5,10 1) Division of Rheumatology, Immunology and Allergy, Department of Medicine, Brigham and Women's Hospital, Boston, MA, USA; 2) Division of Genetics, Department of Medicine, Brigham and Women's Hospital, Boston, MA, USA; 3) Partners Center for Personalized Genetic Medicine, Boston, MA, USA; 4) 3.Partners Center for Personalized Genetic Medicine, Boston, MA, USA Program in Medical and Population Genetics, Broad Institute of MIT and Harvard, Cambridge, MA, USA; 5) Harvard Medical School, Boston, MA USA; 6) Harvard-MIT Division of Health Sciences and Technology, Boston, MA USA; 7) Center for Public Health Genomics, University of Virginia, Charlottesville, VA, USA; 8) Department of Medical Genetics, University Medical Center Utrecht, Utrecht, Netherlands; 9) Department of Epidemiology, University Medical Center Utrecht, Utrecht, Netherlands; 10) Faculty of Medical and Human Sciences, University of Manchester, Manchester, UK.

   Type 1 diabetes (T1D) is a highly heritable metabolic disorder caused by autoimmune destruction of pancreatic cells. Variation in the human leukocyte antigen (HLA) genes, encoding the major histocompatibility complex (MHC) molecules, explains approximately 50% of phenotypic variance for T1D. It is postulated that certain amino acids in the MHC peptide-binding grooves may alter antigen binding or presentation. However high levels of polymorphism and extensive linkage disequilibrium prohibited fine-mapping in the region; for decades, risk was not robustly attributed to specific amino acids other than DQ1-57. Recent developments in statistical imputation enabled HLA fine-mapping through accurate in silico typing of classical allelic and amino acid polymorphisms. Here we imputed and tested HLA variants in 16,085 Caucasian T1D patients and controls. Method: We genotyped 6,670 T1D patients and 9,416 healthy controls collected through the Type 1 Diabetes Genetics Consortium on the ImmunoChip platform. Using SNP2HLA and a reference panel of individuals with typed classical alleles, we imputed 424 2- and 4-digit alleles and 399 amino acid residues in eight HLA class I and II genes. We performed logistic regression and conditional analyses to evaluate the effect of each variant. Results: We confirmed that the most statistically significant association was at DQ1-57 (p=10-917), which explained most of the DQB1 association. Conditioning on this position, we identified a comparably strong independent association signal at DR1-13 (p=10-570). At this position, histidine conferred risk (OR=4.01) while arginine conferred protection (OR=0.08) against T1D. Conditioning on these two positions, we identified six additional amino acid positions that independently conferred risk: DR1-71 (p=10-54), B-158 (p=10-54), A-105 (p =10-39), DR1-86 (p=10-26), C-24 (p=10-18), and C-173 (p=10-11). Discussion: These eight positions almost completely explained the T1D association across the MHC. The two most significant positions, DQ1-57 and DR1-13, showed stronger association than any of the respective classical alleles, underscoring the importance of comprehensive analysis including amino acid residues. DR1-13 was previously unreported in T1D, but known to confer risk to rheumatoid arthritis and follicular lymphoma, suggesting its potentially common role in autoimmune diseases. Our results may aid in discovering autoantigens and physiochemical basis of peptide-MHC-T cell binding.