Early Onset of Diabetes in the Proband Is the Major Determinant of Risk in HLA DR3-DQ2/DR4-DQ8 Siblings

Type 1 diabetes results from the autoimmune destruction of insulin-producing β-cells, a process that begins early in life. Islet autoantibodies are detectable by the age of 5 years—often by the age of 2 years—in children who go on to develop diabetes, and autoantibodies to insulin (generally the first to appear) have been detected as early as 6–12 months (1,2). Established islet autoimmunity is difficult to modulate, and therefore strategies aimed at primary prevention before the initiation of autoimmunity are needed. Trials of potential interventions at this stage will have to recruit children on the basis of genetic risk alone.

Recent whole-genome studies have resulted in an explosion of information regarding genetic susceptibility to type 1 diabetes, but the HLA region remains the most important genetic determinant. More than 90% of children with type 1 diabetes carry the HLA class II haplotypes DRB1*03-DQB1*02:01 (DR3-DQ2) and/or DRB1*04-DQB1*03:02 (DR4-DQ8), and the highest risk DR3-DQ2/DR4-DQ8 diplotype is present in 50% of cases of very early–onset diabetes (3). We and others have previously shown that the age of diagnosis of children with type 1 diabetes and the number of HLA haplotypes they share are determinants of risk of diabetes in their siblings (3,4). HLA class II determined risk can also be modulated by other factors close to or within the HLA region as well as by non-HLA genes (5).

Current trials to prevent the initiation of islet autoimmunity use HLA-based risk assessment to identify individuals eligible for inclusion, but they use different strategies to assign risk that depend on the potential toxicity of the planned intervention (6,7). The international Trial to Reduce IDDM in the Genetically at Risk (TRIGR) combined HLA-determined genetic susceptibility with a first-degree family history of type 1 diabetes to identify children with an estimated 15.4% risk of type 1 diabetes by age 10 who were suitable to include in a dietary modification trial (8). In the Finnish Diabetes Prediction and Prevention (DIPP study), high-risk HLA combinations identified children with no family history of type 1 diabetes who had a 36.4% (95% CI 20.0–52.8) risk of developing two or more antibodies and a 2.4% risk of diabetes within 5 years (9).

Other primary prevention trials are seeking to identify infants at much higher genetic risk. A study of 48 families with siblings matched for the high-risk DR3-DQ2/DR4-DQ8 genotype participating from birth in the prospective Diabetes Autoimmunity Study in the Young (DAISY) reported a dramatically increased risk of islet autoimmunity and diabetes in 29 siblings who shared both extended high-risk haplotypes identical by descent (IBD) with the diabetic proband. In this group, the estimated risk of islet autoimmunity was 63% (95% CI 44–85) by age 7, and risk of diabetes was 55% (95% CI 30–80) by age 12 (10). This compares with a 7% risk of diabetes by age 12 in HLA DR/DQ identical siblings who were not IBD. This strategy for identifying a subgroup of DR3-DQ2/DR4-DQ8 siblings at “extreme risk” based on IBD haplotypes has been used to recruit children for the Pre-POINT study, a clinical trial assessing the efficacy of oral insulin given very early in life (7).

However, a smaller, more recent study of 14 HLA DR3-DQ2/DR4-DQ8 IBD siblings in the Dutch Kolibrie cohort failed to replicate this extreme risk: only 2 of the 14 developed type 1 diabetes by the age of 15 (11). The aims of our study were therefore to evaluate the risks of autoimmunity and diabetes associated with the extended DR3-DQ2/DR4-DQ8 diplotype in a larger group of U.K. families with more than 25 years follow-up in the population-based Bart’s-Oxford (BOX) study of childhood diabetes and to identify other factors that might contribute to the discrepant results of previous studies.

The participants in this study were siblings of probands who had been notified to the BOX study, which, since 1985, has recruited more than 95% of the families of children who have developed type 1 diabetes before the age of 21 years in the former Oxford Health Authority Region of the U.K. (12). The background population is 95% Caucasian and the remainder originates mainly from the Indian subcontinent (data from the Office of Population Censuses and Surveys for 1991). All cases of type 1 diabetes were referred by diabetes specialists based on diagnosis according to World Health Organization criteria (13) and a clinical requirement for insulin treatment from diagnosis. Patients with secondary diabetes, known genetic subtypes including maturity onset diabetes of the young, or clinical type 2 diabetes were not included in the study.

Between 1985 and 2002 we recruited to the study 2,134 siblings from 1,745 families. HLA genotype data were available for 1,440 siblings (67%). Of these, 191 had the high-risk HLA DR3/DR4 genotype. Haplotype sharing was uncertain in 53 proband/sibling pairs. Therefore, 138 siblings were included in the analysis (Fig. 1).

Clinical data and family history of diabetes, including type of diabetes and age at diagnosis, were collected by interview upon entry to the study. The median time between diagnosis of the proband and the enrollment of siblings was 0.2 years (range 0.1–2.5 years). Of 138 siblings, 7 had developed diabetes before entry to the study. Families were followed by annual visit, telephone call, or postal questionnaire, and the diabetes status of every family member was ascertained on each occasion. A further 12 siblings developed diabetes during follow-up. Serum samples used to determine islet autoantibodies were collected routinely from consenting unaffected siblings at entry to the study, with repeat sampling if any family member was positive for one or more autoantibody. At least one sample was tested for autoantibodies in 94 of 131 siblings (72%) who were nondiabetic at entry to the study. The first available sample was measured in each sibling, and all those in this study who were found to be positive for islet autoantibodies were identified using the first sample. The median age at collection of the first sample was 12.9 years (interquartile range [IQR] 8.1–16.4), 13.2 years at collection of the last sample (IQR 8.4–16.8), and 25.4 years (IQR 20.6–29.9) at last contact.

HLA class II DRB1 and DQB1 and HLA class I A and B genotyping was carried out by PCR using a DYNAL Reli SSO system (Invitrogen, Paisley, U.K.). The type 1 diabetes–associated haplotype HLA-DRB1*04-DQB1*03:02 was abbreviated to DR4-DQ8 and HLA-DRB1*03-DQB1*02:01 was abbreviated to DR3-DQ2. IBD siblings were recognized by analyzing the transmission of extended haplotypes from HLA class II to HLA class IA from both parents to affected and unaffected children (10). This will be described as extended HLA identity.

Samples from all participants were tested for islet cell antibodies and autoantibodies to glutamate decarboxylase, islet antigen-2, and insulin, as previously described (14). Data on antibodies to zinc transporter 8 were not available from this cohort. Samples were judged to be antibody positive if levels were at or above the 97.5th percentile of levels in a control population of 2,860 schoolchildren from the same region (15). The islet cell antibody assay was shown to have 81% sensitivity and 86% specificity in the First Immunology of Diabetes Society Combined Antibody Workshop (16). Radiobinding assays for autoantibodies to glutamate decarboxylase, islet antigen-2, and insulin had sensitivities of 91, 74, and 58%, respectively, with 99% sensitivity in the first Diabetes Antibody Standardization Proficiency Evaluation (17).

Risks of diabetes and autoimmunity in siblings were estimated by survival analysis using the log-rank test to compare groups based on HLA sharing and age at diagnosis of diabetes in the proband. Islet autoimmunity was defined as more than two autoantibodies above the 97.5th percentile. Cumulative risks were calculated for clinical diagnosis of diabetes and a composite end point of diabetes and/or islet autoimmunity. For the analyses of diabetes risk, time to event was defined from birth to the development of diabetes or date of last contact; for risk of diabetes and/or autoimmunity time to event was defined as birth to the development of diabetes or first detection of two or more antibodies or date of last sample. Time-to-event outcomes for combined predictive markers (HLA haplotype sharing and age at diagnosis of the diabetic proband) were analyzed using Cox proportional hazards models. As samples for autoantibody testing were not available from birth in this cohort, the cumulative risks of positivity for two or more islet autoantibodies were calculated using the following assumptions: 1) all siblings who were antibody negative had not previously been positive for two or more antibodies, 2) all siblings with at least two autoantibodies remained in this category until diagnosis of diabetes or the date of last contact (end of follow-up), and 3) the date of collection of the first antibody-positive sample was used as the initiation of islet autoimmunity. The risk of the composite end point of diabetes and autoimmunity could not be calculated in 30 individuals for whom no serum sample was available and who remained nondiabetic at last contact. A total of 108 individuals were therefore included in that analysis.

Of 138 informative HLA DR3-DQ2/DR4-DQ8 siblings, 87 (63%) shared both haplotypes with their diabetic proband, 40 (29%) shared one haplotype, and 11 (8%) shared neither. Of the siblings analyzed, 69 (50%) were from families in which the diabetic proband was younger than age 10 years at the time of diagnosis (median 5.4 years [IQR 2.5–7.7 years]), and the remaining 69 were from families in which the proband was diagnosed after this age (median 13.4 years [IQR 11.4–15.2 years]). The analysis included 34 siblings from multiplex families in which more than one first-degree relative had type 1 diabetes. The characteristics of the cohort are shown in Table 1.

By the time of analysis, 19 of the 138 DR3-DQ2/DR4-DQ8 siblings had developed type 1 diabetes. Of these, 15 (11%) were HLA identical, 3 (2%) were haplo-identical, and 1 (1%) did not share either high-risk haplotype with the proband. For 16 diabetic siblings, diabetes had been diagnosed in the proband before age 10 years, and for 3 siblings the proband had been diagnosed after age 10.

The cumulative risk of diabetes by age 15 was 16.6% (95% CI 8.6–24.6) in the HLA-identical siblings, compared with 6% (95% CI 0–12.4) in those sharing one or no haplotype (P = 0.095); cumulative risk was 22.5% (95% CI 12.5–32.5) in the siblings of probands diagnosed when younger than age 10, compared with 3% (95% CI 0–6.8) in those whose proband was diagnosed after age 10 (P = 0.001). Among the 25 siblings from multiplex families, the risk of diabetes was 13.4% (95% CI 6.3–18.9).

As shown in Fig. 2A, the effect of extended HLA haplotypes on cumulative risk of diabetes varied with the age of the proband at onset of type 1 diabetes. Among the 50 HLA-identical DR3-DQ2/DR4-DQ8 siblings from families in which the proband was diagnosed before age 10, the cumulative risk of diabetes by age 15 was 25% (95% CI 13–38), compared with 5.4% (95% CI 0–13) in the 37 siblings from families in which the proband was diagnosed after age 10 (P = 0.009).

Of the 108 siblings included in this analysis, 8 had progressed to the combined end point of islet autoimmunity (more than two antibodies) and/or diabetes by age 5, 17 progressed by age 10, and 23 progressed by age 15. The overall cumulative risk of islet autoimmunity and/or diabetes by age 15 was 26% (95% CI 16–35%); it was 31% (18–44) in HLA-identical siblings, compared with 18% (4.5–31) in those sharing one or no haplotype (P = 0.113). In 53 siblings from families in which the proband was diagnosed before age 10, the risk of islet autoimmunity and/or diabetes was 52% (33–70%) by age 15, compared with 8.6% (0–17) risk by age 15 in 55 siblings from families in which the proband was diagnosed after age 10 (P < 0.001).

As shown in Fig. 2B, the age of the proband at the onset of type 1 diabetes modulated the combined risk of autoimmunity and/or diabetes. Among the 36 HLA-identical DR3-DQ2/DR4-DQ8 siblings from families in which the proband was diagnosed before age 10, the cumulative risk of autoimmunity and/or diabetes by age 15 was 61% (95% CI 13–38) compared with 4.7% (95% CI 0–16) in the 29 siblings from families in which the proband was diagnosed after age 10 (P < 0.0001).

Cox regression analysis showed that the age of the proband when diabetes was diagnosed was an independent determinant of risk of diabetes and of the combined end point of autoimmunity and/or diabetes, but HLA haplotype sharing was not. Coming from a family in which the proband was diagnosed before age 10 was associated with 5.9-fold increased risk of diabetes (95% CI 1.7–20.5; P = 0.006) and 7.4-fold increased risk of the combined end point (95% CI 2.7–20.3; P < 0.001), whereas being HLA identical with the proband was associated with a hazard ratio of 1.9 (95% CI 0.6–5.7; P = 0.265) for the development of diabetes and a hazard ratio of 1.6 (95% CI 0.6–3.8; P = 0.322) for the combined end point.

In this study we examined the risk of diabetes and islet autoimmunity in HLA DR3-DQ2/DR4-DQ8 siblings of children with type 1 diabetes in relation to the number of extended HLA haplotypes shared with the proband and the age at which the proband’s diabetes was diagnosed. We observed that increased risk was concentrated in the subgroup of siblings who were HLA identical to the proband (sharing both haplotypes) compared with siblings who shared one or no haplotype. We have shown for the first time that risk is higher in the DR3-DQ2/DR4-DQ8 siblings of children whose diabetes was diagnosed at a young age and that this factor is the dominant determinant.

The existing literature in this area has yielded conflicting results. Using an approach similar to that adopted by us, a study of 48 families from the DAISY cohort reported that HLA DR3-DQ2/DR4-DQ8 siblings IBD with the proband for both extended HLA haplotypes had a 55% risk (95% CI 30–80) of developing diabetes by age 12 (10). The same study also found a 63% risk (95% CI 41–84) of autoimmunity by age 7 and an 85% risk (95% CI 61–100) by age 15 in this group, compared with 20% risk (95% CI 0–42) by age 15 for those who shared only one or no haplotype. However, a smaller, more recent study of 14 DR3-DQ2/DR4-DQ8 siblings sharing the IBD haplotypes failed to replicate these findings after up to 12 years’ follow-up (11).

Our study of 138 informative families undergoing regular follow-up represents the largest study of IBD extended HLA haplotypes and sibling risk to date. As reported in the DAISY cohort, we found that HLA-identical DR3-DQ2/DR4-DQ8 siblings were at increased risk but the magnitude of risk was less. The estimates of diabetes/autoimmunity by age 15 were similarly high in both studies despite different definitions of islet autoimmunity: the DAISY study used single islet autoantibody positivity while the BOX study used more than two autoantibodies, a more accurate predictor of diabetes.

We have shown that this risk was concentrated in siblings of probands diagnosed when younger than 10 years old, while there was little association with the degree of haplotype sharing in siblings of children diagnosed after age 10. The influence of the age of the proband at the onset of diabetes on the risk among first-degree relatives has been described previously; the risk of progression to diabetes in monozygotic twins of probands diagnosed before 15 years of age was almost twice that of twins of probands diagnosed after age 25, and twins of patients diagnosed when younger than 10 years old were estimated to have a 50% risk of developing diabetes within 6 years (18). Although the investigators noted that this increased risk did not seem to be related to HLA, the frequency of the high-risk HLA DR2-DQ2/DR4-DQ8 diplotype was relatively low: only 15 of 79 twin pairs (19%) diagnosed before the age of 25 years were positive for HLA DR2-DQ2/DR4-DQ8, compared with 5 of 21 twin pairs (24%) diagnosed later. We also previously showed that the overall risk of diabetes was higher in siblings of probands diagnosed with type 1 diabetes very early in life (3). Given the high frequency of the HLA genotype DR3-DQ2/DR4-DQ8 we observed in early-onset type 1 diabetes, as have others (19), our interpretation was that increased HLA susceptibility in siblings of probands with early-onset diabetes was the dominant mediator of increased risk of diabetes. Our current study disproves this.

Rapid progression resulting in early onset of type 1 diabetes is associated with less preserved β-cell function (20), suggesting a more aggressive autoimmune response. To date, most genetic studies of early-onset diabetes have focused on HLA class I and II, particularly DR3/4-DQ8, A*24:02, B*18:01, and B*39:06 (19,21,22), as well as class I C/natural killer cell receptor interactions (23), yet our study strongly indicates that important factors outside the HLA modulate risk. These are likely to include other genes and potentially other epigenetic and environmental modulators. Our data support a recent report from the Nationwide Italian Twin Study highlighting the importance of exposure during fetal or neonatal life over genetics in their effect on concordance rates (24). There is increasing focus on maternal and intrauterine conditions that modulate genetic risk of type 1 diabetes (25). Potential factors include viral infections (26), maternally derived microbiome (27), weight gain in early life (28), and maternal/fetal cell transfer in pregnancy (29). Further studies focused on the determinants of early-onset autoimmune diabetes are required.

A limitation of our study is that islet autoantibody testing did not start in infancy and was not carried out at defined intervals in all families. We therefore had to make a number of assumptions, including use of the date of the first islet autoantibody–positive sample as the date of antibody appearance, although antibodies may have been present for many years. This assumption is likely to have resulted in an underestimation of the risk of islet autoantibody positivity in younger siblings. In addition, we may have missed transient antibody positivity that would have been detected in a prospective study from birth. A further consideration is that data on antibodies to zinc transporter 8 were not available in this cohort. Therefore, our estimates probably represent the lower limits of the true risk of islet autoimmunity.

This study represents an important clarification of increased risk of diabetes in IBD siblings of children with type 1 diabetes with long-term follow-up. It highlights the importance of integrating clinical data such as age of the proband at onset of diabetes and family history in estimating individual risk. It also offers a simple way of stratifying risk in DR3-DQ2/DR4-DQ2 siblings to identify those at extremely high genetic risk for future therapeutic trials.

Acknowledgments. The authors are grateful to the Bart’s-Oxford study families for participating in this research project and to Kyla Chandler and Saba Rokni, University of Bristol, for excellent technical assistance.

Funding. This study was funded by grants from Diabetes UK, the Wellcome Trust, and the North Bristol NHS Trust.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. K.M.G. wrote the manuscript; researched data; coordinated sample collection and analysis; reviewed, edited, and discussed the manuscript; and contributed to discussion. R.J.A., I.W., and A.J.K.W. researched data; coordinated sample collection and analysis; reviewed, edited, and discussed the manuscript; and contributed to discussion. P.J.B. wrote the manuscript; carried out the statistical analysis; reviewed, edited, and discussed the manuscript; and contributed to discussion. P.J.B. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.