A tree based approach to modelling disease associations in the MHC, reveals evidence of class II-independent type 1 diabetes susceptibility loci. J.M.M. Howson, S. Nejentsev, N.M. Walker, S. Field, J.S. Szeszko, H.E. Stevens, D.G. Clayton, J.A. Todd. JDRF/WT Diabetes and Inflammation laboratory, Cambridge Institute for Medical Research, Cambridge University, Cambridge, United Kingdom.
   The MHC class II loci, HLA-DRB1 and HLA-DQB1, are well known to be involved in the etiology of type 1 diabetes (T1D). Not all the linkage in the region however, is thought to be accounted for by these loci alone. Over the past 10 years, the hunt has intensified to find other MHC loci that also contribute to T1D susceptibility. A number of positive and negative associations have been found with HLA-A, -B, -C from the class I region, MICA, BTNL2 and HLA-DPB1. However many of these studies suffered from small sample sizes and resort to subgroup analysis due to statistical difficulties in the modelling of the class II effects (100s of genotypes). This is why no non-class II loci from the region are generally accepted to be associated with T1D in a specific and convincing way. Here we describe the use of recursive partitioning to the modelling of the confounding effects of HLA-DRB1 and HLA-DQB1 caused by linkage disequilibrium between their alleles and non-class II test loci. The grouping method does not rely on risk estimates. However the splitting criteria are based on categorizing individuals as cases and controls according to genotype, so can be considered a risk-based method. Both phased and unphased genotypes were considered. We applied recursive partitioning to 458 UK and 365 USA families typed at 82 microsatellites and SNPs from across the MHC. We obtained evidence for two non-class II loci associated with T1D. The results were replicated in an independent sample set of 1500 British cases and 1500 controls, which further strengthens the candidacy of these immune response genes in T1D susceptibility.