Concordance for Type 1 Diabetes in Identical Twins Is Affected by Insulin Genotype

We have studied and compared 40 concordant and 37 discordant monozygotic twins for type 1 diabetes at both HLA-DQBI and the insulin gene (INS) loci. Twins studied were selected blind to those included in our previous study 15 years ago of HLA-DR phenotype (14). A total of 54% of twin pairs were common to both studies.

Twins were ascertained from 1967 to 1997 by referral to the British Diabetic Twin Study; monozygosity was established in each twin pair as described previously (18,19). DNA was not available from all twins. All twins were invited to attend the Diabetes Department for blood sampling, but in the study period, only a fraction were able to do so. We therefore selected twin pairs who had attended the Diabetes Department consecutively from 1995 to 1997, who were either concordant or discordant for type 1 diabetes. The diabetic index twin from both concordant and discordant pairs had a comparable age at diagnosis, and nondiabetic twins from discordant pairs had a low risk of developing diabetes. Discordant pairs likely to remain discordant were identified by being >11 years from the diagnosis of the index twin and, at 11 years by having normal glucose tolerance and the absence of islet cell antibody (ICA), GAD antibody, or IA-2ic antibody.

Of 247 concordant twin pairs, we selected a consecutive series of 40 pairs attending the Diabetes Center (mean age of diagnosis of index twin 16 ± 9 years, range 1–39; 16 male pairs; median time to concordance 3.95 years, range 0.4–25). Of 109 discordant twin pairs, we identified a consecutive series of 37 pairs (mean age of diagnosis of index twin 19 ± 11 years, range 2–59; 20 male pairs) in which the index twin had a comparable age at diagnosis to the index twin of the concordant pairs. Nondiabetic twins underwent oral glucose tolerance testing (glucose was given as 75 g or 1.75 g/kg, whichever was less) to confirm that they had neither diabetes nor impaired glucose tolerance both initially and at intervals thereafter; diagnosis of type 1 diabetes was made according to standard guidelines (20). Sera were stored at –20°C and analyzed for each of three antibody assays. Analysis was performed on batched samples by observers blinded to the clinical status of the subjects.

IA-2ic and GAD₆₅ were measured by radioimmunoprecipitation assays as described previously (Promega, Madison, WI) (17). All samples were tested in duplicate, including positive and negative control standard sera. Each assay for IA-2ic and GAD₆₅ antibodies included serially diluted sera from a patient with stiff-man syndrome and a prediabetic individual (a twin who subsequently developed type 1 diabetes but was not part of this study) to further evaluate the cutoff level for positivity for GAD₆₅ and IA-2ic. Levels >3 SD higher than that of a control population were considered positive, as described previously (17). In the latest IA-2ic (unpublished) and GAD₆₅ antibody proficiency workshops, our assays had a sensitivity, specificity, validity, and consistency of 100% in each (21).

Undiluted sera were screened for ICA. The presence of ICA was established by testing serum with indirect immunofluorescence on a fresh group O cryofixed human pancreas (16). Positive tests were defined as ≥4 Juvenile Diabetes Foundation units. The ICA assay assessed in the ENDIT workshop (unpublished) had a sensitivity and specificity of 100%.

HLA-DQBI was studied by a polymerase chain reaction restriction fragment–length polymorphism method combined with group-specific primers adapted from Nomura et al. (22). DNA was extracted from blood (23), and a 241-bp region from the polymorphic second exon of the DQB1 gene was amplified in two separate PCRs, specific for DQwI and DQw2,3,4 alleles. Seven DQwI and six DQw2,3,4 alleles can be identified after restriction against panels of seven and five restriction enzymes, respectively (22,24). The digested and undigested products were visualized by ethidium bromide staining after electrophoresis in a 4.2% agarose gel.

INS was studied at the 1,127 Pst I polymorphic restriction site in the 3′ untranslated region of the insulin gene and at the −23 Hph I polymorphic restriction site (3). The digested and undigested products were visualized by ethidium bromide staining after electrophoresis in a 4.2% agarose gel. By convention, the disease associated allele at both sites is denoted as “+.”

All statistical analyses were performed using SPSS for Windows software (version 9; Chicago, IL). Differences in frequencies of genotypes and alleles were calculated by χ² analysis; all P values quoted are two-sided. Power analysis based on a 31% difference of DR3/DR4 heterozygotes between concordant and discordant twins, as observed in our previous study (14), indicated that 37 twin pairs in each group would be sufficient to show a difference at the 5% level with 80% power. The study was approved by the ethical committee of the Barts and the London National Health Service Trust. Written and informed consent was received from the twins or their parents, as appropriate.

Results are presented in Table 1. No differences were found between concordant and discordant twins for positivity of DQB1*0302, DQB1*0201, the combination of DQB1*0302/0201, and DQBI*0602/3. In contrast, 2 of 40 (5%) concordant twins compared with 8 of 37 (22%; P = 0.04) discordant possessed neither DQB1*0201 nor DQB1*0302.

Results are presented in Table 2. The type 1 diabetes–associated +/+ genotype was increased for the 5′ Hph I (P = 0.005; Table 2) polymorphism. Similarly, the 3′ Pst I INS polymorphism was increased in frequency in concordant twins (frequency of +/+ genotype was 87.5% in concordant twins and 65% in discordant twins; P = 0.035).

Results are presented in Table 3. A total of 2 of 40 (5%) concordant twins compared with 12 of 37 (32.4%) discordant twins possessed neither DQB1*0201/DQB1*0302 nor the high-risk Hph I INS +/+ genotype (P = 0.002).

Our hypothesis was that discordance, compared with concordance, for type 1 diabetes in monozygotic twins is caused by a decreased load of both MHC and non-MHC susceptibility genes. From our observations, we conclude that the “load” of both MHC and non-MHC susceptibility genes does have an impact on disease penetrance of type 1 diabetes. Thus, the 5′ Hph I locus of the insulin gene was more prevalent in concordant than discordant twin pairs. Furthermore, there was an increased frequency of no high-risk HLA-DQB1 alleles in discordant compared with concordant twins. HLA-DQ protective alleles were as prevalent in concordant pairs as in discordant pairs. Lastly, only 5% concordant vs. 32.4% discordant twin pairs with type 1 diabetes possessed neither the HLA-DQB1*0302/0201 combination nor the INS Hph I high-risk genotype.

Differences between monozygous twins cannot be determined by germ-line genes. However, there are a number of ways genes could contribute to such disease discordance, including an increase in either protection or susceptibility to the disease. In the absence of a comprehensive understanding of the genes that contribute to genetic susceptibility to diabetes, the analysis of monozygotic twins discordant and concordant for diabetes provides a powerful method to assess the impact of a particular gene on disease expression.

Using DQB1 typing, we have confirmed the importance of the total load of MHC alleles as a determinant of concordance of disease, although we have been unable to confirm our previous results by studying individual DQB1 alleles compared with the previous DR serological typing (14). In that previous study, we found an increased frequency of the heterozygous phenotype HLA-DR3 and HLA-DR4 (linked to HLA-DQB1*0201 and DQB*0302, respectively) in concordant compared with discordant monozygotic twins. Several explanations are possible for the discrepant findings. First, there is a need to study a larger sample population to confirm or refute the results. The numbers recruited for this study represent the maximum number available at the time of the study; however, prospective sampling has been initiated. The discordant twin pairs were selected in the previous study because the index twin had been diagnosed >5 years prior, so the risk of diabetes developing in the nondiabetic twin was less than 10% (19). In the present study, we selected discordant pairs in which the index twin had been diagnosed for more >11 years, and the nondiabetic twin had normal glucose tolerance and no diabetes-associated autoantibodies. In the present study, the risk of the nondiabetic twins developing diabetes is calculated as <2% (16,19). The more precise identification of twin pairs likely to remain discordant for diabetes is an important feature of this present study. Comparison of our discordant twins with a previous study of an ICA-negative control population (n = 63) using the same methodology (24) confirms the importance of HLA-DQ in the susceptibility to type 1 diabetes in these twins. There was an increase in DQB1*0302 in discordant twins (65% positive) compared with controls (15.9% positive; P = 0.0004) and a decrease in the protective DQB1*0602/3 allele (5.4% positive compared with 30.2%; P = 0.003). This argues against a purely environmental cause for diabetes, even in twins discordant for disease. Last, the HLA-DR study of 1983 and the current HLA-DQB1 study are not directly comparable with regard to HLA class II typing. Although both HLA-DQB1*0201 and DQB1*0302 are in linkage disequilibrium with HLA-DR3 and DR4, respectively, they also form haplotypes with other DR antigens.

No twin studies have assessed the impact of non-HLA genes on susceptibility to type 1 diabetes. We found an increased frequency of diabetes-associated insulin genotypes in the concordant twins. The insulin gene is a plausible candidate gene for type 1 diabetes because of its β-cell–specific expression. It has been suggested that differences in the insulin gene transcription could depend on sequence variation in the disease-associated hypervariable region, although different studies have revealed discrepant results, depending on the reporter construct and the in vivo/vitro models used (25,26,27,28,29,30). Interestingly, whereas twins concordant for diabetes had a higher frequency of disease-associated insulin genotypes than discordant pairs, twins discordant for diabetes did not have a greater frequency than did the islet cell–negative control population (31) (Table 2). This supports the hypothesis that there is a decreased load of diabetes susceptibility genes in discordant compared with concordant twins. Last, this study would also support the interaction of HLA-DQ and insulin gene alleles in determining susceptibility and protection to type 1 diabetes.

This study was supported by the Diabetic Twin Research Trust, the Joint Research Board of St. Bartholomew’s Hospital, and the British Diabetic Association.

K.A.M. was a British Diabetic Association R.D. Lawrence Research fellow during this study.

We thank the twins and physicians who made these studies possible, as well as Dr. David Pyke, who initiated the twin study.