The role of low-frequency variants in type 1 diabetes (T1D) susceptibility still remains to be clarified. In the present study, we analyzed low-frequency variants of the T1D candidate genes in Japanese. We first screened for protein-changing variants of 24 T1D candidate genes in 96 T1D patients and 96 control subjects, and then the association with T1D was tested in 706 T1D patients and 863 control subjects recruited from the collaborating institutions in Japan. In total, 56 protein-changing variants were discovered; among them, 34 were low-frequency variants (allele frequency < 5%). The association analysis of the low-frequency variants revealed that only the A908V variant of GLIS3 was strongly associated with resistance to T1D (Haldane's odds ratio = 0.046, p = 8.21 × 10(-4), and pc=2.22 × 10(-2)). GLIS3 is a zinc finger transcription factor that is highly expressed in pancreatic beta cells, and regulates beta cell development and insulin gene expression. GLIS3 mRNA is also moderately expressed in the human thymus. The precise mechanism responsible for the association is unclear at present, but the A908V variant may affect autoimmunity to the GLIS3 protein itself; the 908V containing epitope may induce central or peripheral tolerance more efficiently than that of 908A.

Since vascular endothelial growth factor (VEGF) has a strong effect on induction of vascular permeability, VEGF is an attractive candidate gene for development of diabetic macular edema (ME). Among the 378 patients with type 2 diabetes studied, 203 patients had no retinopathy, 93 had non-proliferative diabetic retinopathy (NPDR), and 82 had proliferative diabetic retinopathy (PDR). ME was present in 16 patients with NPDR and 47 patients with PDR. We genotyped three VEGF polymorphisms: C-2,578A, G-1,154A, and C-634G. Genotype and allele distribution of C-634G, but not C-2,578A or G-1,154A, were significantly different between patients with and without diabetic retinopathy. Logistic regression analysis revealed that the C-634G genotype was a risk factor for DR (p = 0.002), and furthermore for ME (p = 0.047), independently from severity of DR, with the -634C allele increasing the risk. Macular thickness measured by optical coherence tomography was correlated with the C-634G genotype, with the trend increasing with the presence of more -634C alleles (p = 0.006). Stepwise regression analysis showed that duration of diabetes and presence of the C-634G genotype were independent predictors of macular thickness. In addition, basic transcriptional activity levels associated with the -634C allele were greater compared to those seen with the -634G allele in human glioma and lymphoblastic T-lymphocyte cells. These results demonstrate that the VEGF C-634G polymorphism is a genetic risk factor for ME as well as DR.