Introduction

The appearance of circulating islet autoantibodies before the clinical onset of immune-mediated type 1 diabetes makes it possible to select subjects at increased risk of becoming hyperglycaemic among first-degree relatives of known patients [1–9]. However, prevention studies testing the potential of pharmacological interventions to arrest or slow down the subclinical disease process in antibody-positive relatives are still hampered by large differences in disease progression rate in marker-positive subjects [1–3, 10–12]. Therefore, the study design should benefit from further refining the inclusion criteria in order to select participants with a more homogeneous risk of the disease, thereby reducing the numbers needed to treat [3, 12].

A recent study in siblings of type 1 diabetic patients [13] confirmed the generally held view that the risk of diabetes increases with the number of islet autoantibody specificities [4–9], but in addition indicated that the presence of autoantibodies against the intracellular domain of insulinoma-associated protein-2 (IA-2 autoantibodies, IA-2A), which is often associated with multiple antibody positivity [13–16], confers a higher risk of rapid progression towards clinical onset than multiple antibody positivity per se [13]. In persistently IA-2A-negative siblings the risk of diabetes increased significantly with the number of other autoantibodies present (GADA, IAA and/or ICA), but the progression rate remained less than 10% within 5 years [13]. These findings warrant a search for additional biological markers that may complement autoantibody measurements for refining diabetes risk assessment. In this respect genetic markers should be considered, especially positivity for HLA DQA1*0301-DQB1*0302 (DQ8) and HLA DQA1*0501-DQB1*0201 (DQ2) risk haplotypes, alone or in heterozygous combination (DQ2/DQ8), the latter being preferentially associated with early-onset diabetes [17, 18].

Since type 1 diabetes prevention studies have included types of first-degree relatives apart from siblings, and sometimes even second-degree relatives [10, 11], it is also important to test the predictive value of marker combinations in different types of first-degree relatives, who have been reported to differ in terms of their disease risk [19, 20]. We therefore measured IAA, ICA, GADA and IA-2A yearly in a large group of siblings, offspring and parents of type 1 diabetic patients, and followed them in terms of diabetes development to investigate whether (1) the frequency and diabetes-predictive value of the various islet autoantibodies tested, alone or in combination, varied according to the type of kinship to the proband; (2) the presence of HLA DQ risk haplotypes or genotypes could influence the progression rate to diabetes in antibody-positive relatives and explain (part of) the differences among relatives; and (3) specific combinations of HLA DQ haplotypes or genotypes and antibody profiles could further refine risk assessment.

Subjects and methods

Subjects and data collection

Siblings (n=1,727), offspring (n=2,379) and parents (n=483) of type 1 diabetic patients were consecutively recruited by the Belgian Diabetes Registry (BDR) between 1 January 1989 and 31 December 2001 according to previously defined criteria [13, 21]. At entry, blood was sampled and a short questionnaire with demographic, familial and personal information was completed. The relatives were not prescreened for ICA or any other autoantibody before entering the study. The group of first-degree relatives (n=4,589), aged 0–39 years, was followed for a median (interquartile range) period of 52 (43–59) months. The predictive value of biological markers was based on the results of baseline samples. The non-diabetic control subjects (n=790), aged 0–39 years, were recruited among blood donors, laboratory personnel and children attending wards for minor surgery, including correction of phimosis. None of the controls’ relatives had type 1 diabetes. Overall, there was no difference in age between relatives and the control group, but on average siblings were younger than parents (p<0.001) and older than offspring (p<0.001) (Table 1). BMI was recorded and analysed after transformation into a standard deviation score (SDS) in comparison with data from a reference cohort comprising 15,636 male and 14,899 female subjects recruited between 1978 and 1990 [22, 23]. Relatives who developed diabetes during follow-up were identified through repeated contacts with Belgian endocrinologists and paediatricians, self-reporting through yearly questionnaires and a link with the BDR patient database, where newly diagnosed diabetic patients under age 40 are registered. All participating first-degree relatives gave informed consent and the study was approved by the ethics committees of BDR and participating university hospitals and was carried out in accordance with the Declaration of Helsinki as revised in 2000 (http://www.wma.net./e/policy/b3.htm).

Table 1 Characteristics of first-degree relatives of type 1 diabetic patients and non-diabetic healthy control subjects under age 40 years
Full size table

Assays

ICA were determined by indirect immunofluorescence and endpoint titres expressed as Juvenile Diabetes Foundation (JDF) units [24]. IA-2A, GADA and IAA were measured by liquid-phase radiobinding assays and expressed as percentage of tracer bound in haemolysis-free sera [24]. Cutoff values for antibody positivity were determined as the 99th percentile of antibody levels obtained in 790 non-diabetic control subjects after omission of outlying values, and were ≥12 JDF units for ICA, ≥0.6% for IAA, ≥2.6% for GADA and ≥0.4% for IA-2A [24]. Between-day coefficients of variation determined on human control sera were 11% (n=474) for IA-2A, 10% (n=427) for GADA and 9% (n=413) for IAA at the level of 1.8, 7.8 and 7.2% tracer binding, respectively. In the IDW Combinatorial Workshop [25] diagnostic sensitivity adjusted for 99% specificity was 73% for ICA, 85% for GADA and 36% for IAA. In the DASP 2002 [26], sensitivity and specificity were 36 and 98% for IAA, 88 and 96% for GADA, and 62 and 97% for IA-2A, respectively. cDNAs for the preparation of radiolabelled GAD and the intracellular domain of IA-2 (amino acids 603–980) were kindly donated by Professor Å. Lernmark (University of Washington, Seattle, WA, USA) and Dr. M. Christie (King’s College School of Medicine and Dentistry, London, UK), respectively. All initially antibody-positive relatives (n=373) as well as the vast majority of the entire group (3,780/4,589) were genotyped for HLA DQA1 and DQB1 as previously described [18]. Plasma glucose was measured as before [13].

Statistical analysis

Statistical differences between groups were assessed with the Chi-square test with Yates’ correction or Fisher’s exact test for categorical variables, and with the Mann–Whitney U-test for continuous variables. Diabetes-free survival was estimated by Kaplan–Meier analysis [27] and differences in progression to diabetes with the log–rank test [28]. The Cox proportional hazards model, performed by the forward stepwise method, was used to investigate the independent contributions of risk factors identified by univariate analysis, with calculation of 95% confidence intervals as hazard ratios [29]. All statistical tests were performed two-tailed using SPSS for Windows 11.0 (SPSS, Chicago, IL, USA) and considered significant whenever p<0.05 or, in case of k comparisons, whenever p<0.05/k (Bonferroni adjustment) [30].

Results

Autoantibody positivity

At first sampling, 4% of siblings tested positive for ICA, which tended to be higher than in offspring of a diabetic father, and was significantly higher than in offspring of a diabetic mother, in parents and in control subjects (overall p<0.001; Table 2). The hierarchy of antibody positivity according to kinship was similar for the other individual antibodies as well as for the number of antibodies (Table 2). Within each category of relatives the highest frequencies were noted for GADA and IAA, and the lowest for IA-2A (Table 2). Overall, 11.1% of siblings (p<0.001 vs all other subject groups) and 7.1% of offspring of a diabetic father presented at least one type of antibody at first sampling. In siblings and offspring there were no differences in antibody frequency according to age (0–9 vs 10–39 years), except for IAA, for which the frequency was significantly higher at younger ages (not shown).

Table 2 Prevalence of autoantibodies in non-diabetic control subjects and in the initial sample of first-degree relatives of type 1 diabetic patients according to relationship to the proband
Full size table

Progression to type 1 diabetes

During follow-up, 31 siblings, ten offspring of a diabetic father, two offspring of a diabetic mother and none of the parents developed type 1 diabetes with a progression frequency of, respectively, 18, 8, 2 and 0‰ (p<0.001) and a median (interquartile range) time to diabetes of 28 (9–44) months. Their median (interquartile range) age at diagnosis was 13 (10–20) years, with a male/female ratio of 23:20 (1.15). The various types of prediabetic relatives did not differ significantly overall in demographic and genetic parameters or in median time to diabetes (not shown).

Progression to diabetes was significantly associated with antibody positivity, especially IA-2A, and increased with the number of autoantibodies present at first sampling with (not shown) or without (Table 3) inclusion of ICA (p<0.001). However, when considering possible combinations of two molecular antibodies (n=70), progression was significantly higher in the presence than in the absence of IA-2A (p<0.001) (Table 3).

Table 3 Five-year diabetes-free survival rate according to type and number of antibodies at first sampling, IA-2A status and HLA DQ genotype in first-degree relatives under 40 years
Full size table

Initially IA-2A-positive relatives (n=58)

Although siblings and offspring of a diabetic father tended to develop more diabetes than offspring of a diabetic mother or parents, there was no significant difference in risk according to kinship (overall p>0.05; not shown). However, when all IA-2A-positive relatives were analysed according to HLA DQ risk status, significantly more subjects progressed to diabetes in the presence than in the absence of the DQ8 susceptibility haplotype (p=0.038). Especially the DQ2/DQ8 genotype (n=17) was associated with the highest 5-year progression rate [95% confidence interval] towards diabetes (84% [65–100%] vs 32% [16–48%] for all other genotypes (n=41), p=0.003; Fig. 1). This result was confirmed in the group positive for IA-2A and IAA or IA-2A and GADA (p<0.001; Table 3).

Fig. 1
figure 1

Diabetes-free survival in IA-2A-positive relatives (n=58) after stratification for all possible combinations of the presence or absence of both risk haplotypes, HLA DQ2 and HLA DQ8 (DQ2/non-DQ8 [— - — -] vs nonDQ2/DQ8 [- - - -] vs nonDQ2/nonDQ8 [····] vs DQ2/DQ8 [—]). In each panel the 5-year diabetes-free survival (95% confidence interval) is shown for each arm.

Full size image

Initially IA-2A-negative relatives (n=4,531)

Progression to diabetes within 5 years was significantly associated with the number of other molecular antibodies at initial sampling (overall p<0.001), but did not exceed 10% even in the presence of IAA and GADA (Table 3). Similar results were observed when also considering ICA (not shown). Within the IA-2A-negative relatives positive for ICA, GADA and/or IAA (n=315), there was no significant difference in progression rate according to the type of kinship (not shown). In contrast, with the IA-2A-positive subjects progression to diabetes did not differ according to DQ8 status (not shown), but there was significantly more progression [95% confidence interval] in carriers of DQ2 (8% [4–13%]) than in relatives without DQ2 (1% [0–3%]; p<0.001).

Characteristics of initially antibody-positive relatives

After stratification for IA-2A status, prediabetic and non-diabetic relatives did not differ significantly in terms of age, male/female ratio, SDS-BMI, type of relationship, random glycaemia, or antibody frequency and levels (in case of antibody positivity) at first sampling.

HLA DQ risk haplotypes and genotype according to type of relationship

All relatives under age 20 years (n=2,812) were HLA-DQ-genotyped and stratified according to the type of relationship. Siblings carried DQ2 (46 vs 39%; p=0.002), especially the high-risk DQ2/DQ8 genotype (12 vs 6%; p<0.001), more frequently than offspring, regardless of the gender of the diabetic patient. Siblings and offspring did not differ significantly in DQ8 prevalence (35 vs 33%).

Cox regression analysis

Univariate analyses (enter method) were first performed in which each of the potential risk factors for diabetes was confirmed to be significantly associated with progression to diabetes (Table 4). To assess the independent contributions of these identified biological predictors, multivariate forward stepwise Cox regression analysis was carried out. In a first multivariate model including all genotyped relatives (n=3,780), the number of detected antibodies, IA-2A, HLA DQ2/DQ8 and being a sibling contributed independently to the risk of diabetes (Table 4). Next, we investigated the hazard ratio for diabetes of the parameters selected in model 1 in the initially antibody-positive subjects at increased risk (model 2). Only IA-2A and HLA DQ2/DQ8 were selected as significant independent predictors (p<0.001) (Table 4). Similar results were obtained when ICA was omitted from the analysis (not shown).

Table 4 Cox regression analysis in all genotyped relatives (n=3,780; model 1) and in initially antibody-positive relatives (n=373; model 2)
Full size table

Discussion

In the present study in 4,589 first-degree relatives of type 1 diabetic patients analysed at study entry for diabetes autoantibodies (IAA, ICA, GADA, IA-2A) and HLA DQ genotype and followed for several years, the frequency of individual or multiple autoantibody positivity and progression to diabetes tended to be graded according to the known hierarchy of risk among first-degree relatives (siblings > offspring diabetic father ≫ offspring diabetic mother ≅ parents) [19, 20]. In relatives with circulating autoantibodies at study entry, the progression rate to diabetes depended much more on HLA DQ genotype than on the type of relationship to the proband. Differences in frequency of HLA DQ risk haplotypes or genotypes could partially explain differences in progression rates between siblings and offspring, but not differences between offspring of a diabetic father and offspring of a diabetic mother.

The present study also confirmed that IA-2A signals a higher risk of impending clinical onset of diabetes than multiple antibody positivity per se [13]. IA-2A positivity at study entry conferred greater than 50% risk of diabetes within 5 years regardless of the number of other antibodies present or the type of relationship to the proband. In the absence of IA-2A, other antibodies are also predictive of diabetes but to a much lower degree. Although the number of relatives included and their follow-up time may be smaller than in some other reports [6, 9], the present study has analysed a large unselected group of first-degree relatives with a substantial follow-up, unbiased towards certain antibody positivities in the absence of prescreening for ICA or enrichment in prediabetic relatives, as may be the case in some studies [4, 7], and with a wider age range than in other studies [6]. Our confirmation [13] of the primacy of IA-2A positivity over the number of autoantibodies—at variance with other claims [9]—should not be ascribed to differences in study size but rather to differences in study protocol and/or the way the data were analysed; indeed, when expressed in the same way as in other studies [4] our results also indicate an increasing risk of type 1 diabetes with the number of (molecular) antibodies present.

Both in IA-2A-positive and in IA-2A-negative patients with circulating diabetes autoantibodies, HLA DQ genotyping was found to provide additional information in terms of diabetes prediction. Cox regression analysis confirmed that the presence of DQ2/DQ8 complemented IA-2A positivity in the prediction model and overruled differences associated with the type of relationship and the number of antibodies. In IA-2A-positive relatives carrying DQ2/DQ8, progression to diabetes averaged 84% within 5 years, in line with the preferential association of DQ2/DQ8 with early-onset type 1 diabetes [3, 17, 18]; hence, simultaneous positivity for both markers can almost be considered a diagnostic criterion for diabetes long before glycaemia starts to rise, and defines a group of relatives at homogeneously high risk of type 1 diabetes. Carriers of DQ8 in combination with a haplotype other than DQ2 also tended to progress more rapidly to diabetes than DQ8-negative relatives, thereby confirming the preferential association of IA-2A and DQ8/DR4 in new-onset patients [31–33]. Recently, the strong diabetes-predictive value of IA-2A, especially in cases where there are high levels and specific subclasses, was independently confirmed in persistently antibody-positive relatives [34]. The use of combined positivity for IA-2A (above percentile 99) and HLA DQ2/DQ8 in one blood sample may, however, be more practical for the enrolment of high-risk subjects in prevention trials.

Future work should aim to increase the specificity of screening in relatives positive for other antibodies. In the absence of IA-2A, the presence of at least one other antibody type (IAA, ICA or GADA) did confer a significant—albeit low risk of diabetes. This risk increased with the number of antibodies present, as indicated by both survival analysis and Cox regression, and in carriers of HLA DQ2, regardless of DQ8 status, it is consistent with the reported preferential association of GADA with DQ2 in new-onset patients [31–33]. The differential HLA DQ enhancement of progression to diabetes according to IA-2A status supports previous suggestions that different diabetogenic pathways may converge to the same clinical endpoint of immune-mediated type 1 diabetes [35, 36]. Nevertheless, progression to diabetes in antibody-positive carriers of DQ2 remains low in the absence of IA-2A. Age, gender, BMI and 5′ INS genotype (K. Decochez, unpublished)—all known to influence the risk of diabetes [3, 18, 20, 22]—could not improve the prediction of diabetes so far in this group, but demonstration of their possible added value would require follow-up of larger study groups for longer periods. Disease prediction may further benefit from the assessment of the persistence of antibodies, changes in antibody levels and SDS-BMI over time and, most likely, from the repeated assessment of beta cell function and insulin resistance [22, 24, 37–42]. The most standardised functional tests (intravenous glucose tolerance test, clamps) are, however, more difficult to implement on a larger scale in non-diabetic risk groups. Since IA-2A can at present only be detected in about 70% of prediabetic subjects [13], prediction strategies would also benefit from an increase in sensitivity of IA-2A screening. In the present study the highly predictive combined presence of IA-2A and DQ2/DQ8 identifies only about 30% (13 out of 43) of the future diabetic relatives; therefore additional, e.g. functional, parameters would also be most welcome here to improve screening sensitivity.

In conclusion, (multiple) antibody positivity and progression to diabetes was graded according to the known hierarchy of risk among relatives, and confirmed our previous observation in siblings of the primacy of IA-2A positivity over multiple antibody positivity for the prediction of diabetes. Progression to diabetes was much more dependent on HLA DQ status than on the nature of kinship. Differences in the frequencies of HLA DQ risk genotypes could partly explain differences in the risk of diabetes between siblings and offspring, but not between offspring of a diabetic mother or father. In IA-2A-positive relatives there was significantly more progression to diabetes in the presence of DQ8, particularly when in heterozygous combination with DQ2; in the presence of islet antibodies different from IA-2A there was more progression in carriers of DQ2 regardless of DQ8 status. This differential HLA DQ enhancement according to IA-2A status may suggest the existence of at least two distinct pathogenic pathways for immune-mediated type 1 diabetes.