Combination of HLA-A24, -DQA1*03, and -DR9 Contributes to Acute-Onset and Early Complete β-Cell Destruction in Type 1 Diabetes: Longitudinal Study of Residual β-Cell Function

The subjects consisted of 266 patients with type 1 diabetes, including 151 men and 115 women aged 34 ± 14 years (means ± SD) at the onset of diabetes, and 136 normal control subjects. The relationship between the mode of onset of diabetes and each HLA allele was analyzed in these subjects. Our longitudinal investigation of residual β-cell function was designed as a historical cohort study (20), and the onset of diabetes was defined as the starting point. To determine the temporal profile of β-cell destruction as precisely as possible, 70 patients in whom the first assessment of residual β-cell function revealed complete β-cell loss at >5 years after the onset of diabetes were excluded from the longitudinal study, because the time of complete β-cell loss was unknown over the 5-year period in these patients. By setting this 5-year limit, the error in determining the time of complete β-cell destruction was reduced to less than about one-tenth (<5 years) of the longest observation period (0.08–48.0 years; median 10 years). As a result, 196 patients with type 1 diabetes, including 115 men and 81 women aged 36 ± 14 years (means ± SD) at the onset of diabetes, were included in the longitudinal study. Some of these patients have been reported previously (2,13,21). All of the patients were Japanese residents of the Tokyo and Yokohama areas. A diagnosis of type 1 diabetes was made according to the ADA guidelines (1). In addition, daily urinary C-peptide excretion of <6.6 nmol/day or an integrated serum C-peptide value during the 100-g oral glucose tolerance test of <3.3 nmol/l was used to define type 1 diabetes, as described previously (2,6,13,21). Autoantibodies to GAD were positive at the onset of diabetes in 82.6% (123 of 149) of our patients. This study was approved by the Committee for Investigations Involving Human Subjects of Toranomon Hospital. All patients gave informed consent before enrollment in the study.

A sensitive C-peptide radioimmunoassay was used to assess residual β-cell function (8). We have previously defined a serum C-peptide response (CPR) of <0.033 nmol/l after a 100-g oral glucose load (ΔCPR) as indicating complete β-cell destruction (2,21). To allow more convenient assessment of residual β-cell function over time, a fasting serum C-peptide level below the detection limit (0.017 nmol/l) or a level <0.033 nmol/l at 2–3 h postprandially was also used as a criterion indicating complete β-cell destruction in this study. The time of 2–3 h postprandially was selected because the maximal serum C-peptide level was measured at this point in type 1 diabetic patients undergoing a 100-g oral glucose tolerance test (8). The relationships between these three serum C-peptide criteria for complete β-cell loss were analyzed in 57 of the 266 type 1 diabetic patients.

Assessment of residual β-cell function was done as follows. In 169 patients, the first measurement of serum C-peptide was performed at 5.9 ± 7.9 years (means ± SD) (median 2 years; range 0.08–37 years) after the onset of diabetes, and serum C-peptide was measured a total of 5.1 ± 2.9 times (median 10 times; range 2–14 times) during a disease duration of 13.3 ± 10.5 years (median 10 years; range 0.1–48 years). Complete β-cell loss that occurred among these 169 patients was confirmed by the next measurement of serum C-peptide in all (n = 54) but two patients. In 16 patients, complete β-cell loss was detected by the first assessment of β-cell function performed at 1.1 ± 0.33 years (median 0.45 years; range 0.08–4 years) after the onset of diabetes. In 13 of these 16 patients, complete β-cell loss was confirmed by the next measurement of serum C-peptide. In the four patients who underwent serum C-peptide measurement a total of three times, complete β-cell loss was observed >5 years (median 8 years; range 6–14 years) after the detection of residual β-cell function by the second assessment. Based on the above-mentioned 5-year limit, these patients were censored at the last time when residual β-cell function was detected because time of complete β-cell loss was unknown over the 5-year period. In seven patients, the presence of residual β-cell function was only investigated by one test at 4–24 years (median 14 years) after the onset of diabetes.

HLA-DR and -DQ alleles were typed by the previously described PCR–restriction fragment–length polymorphism (RFLP) methods (22–24). HLA-A, -C, -B, and -DR antigens were typed by the microcytotoxicity test (25), except in 38 patients for whom only HLA-A alleles were determined by PCR-RFLP method (26). In 91 patients, serological data of HLA-A alleles were confirmed by PCR-RFLP method. The DRB1-DQA1-DQB1 haplotype was established based on known linkage disequilibrium data (27).

According to the time from the diagnosis of diabetes to the start of insulin therapy, we divided the patients into those with acute-onset (≤12 months) type 1 diabetes (n = 138) and slow-onset (>12 months) type 1 diabetes (n = 99), as described previously (28,29). By reviewing the medical records, we detected patients who developed ketosis or ketoacidosis within 1 week of the onset of hyperglycemic symptoms and had a concomitant rise of pancreatic exocrine enzymes at the onset, and we classified such patients as having fulminant type 1 diabetes (n = 11) (7). Eighteen patients could not be allocated to a subgroup because their medical records were incomplete.

The Kaplan-Meier method was used to estimate the cumulative incidence of complete β-cell destruction, and differences between the cumulative incidence curves were assessed by the log-rank test (30). Person-time incidence rates were expressed as the number of events per 100 patient-years based on the ratio of the observed number of events to the total number of patient-years of exposure. Cox’s proportional hazards model (30) was used to examine the combined influence of the variables on the risk of complete β-cell loss in type 1 diabetes. The Mann-Whitney U test was used for unpaired data. Differences of frequency in two or three groups were assessed by Fisher’s exact probability test or the contingency χ² test, respectively. To calculate corrected P values, the raw values were multiplied by the number of alleles examined at each locus. Results are expressed as means ± SD. Statistical significance was defined as P < 0.05. All analyses were performed with the Stat View 4.02J statistical package and Survival Tools for StatView (Abacus Concepts, Berkeley, CA).

The correspondence between judgment of complete β-cell destruction based on ΔCPR versus the fasting serum C-peptide level or based on ΔCPR versus serum C-peptide level at 2–3 h postprandially was 84.2% (48 of 57) and 93.0% (54 of 57), respectively.

Because the combination of HLA-A24 and HLA-DQA1*03 was associated with accelerated β-cell destruction in a previous cross-sectional study (13), we first examined the effect of this allele combination on the temporal profile of complete β-cell loss. Among 177 patients who could be typed for the HLA-A and -DQA1 loci, all but 1 (99.4%) had HLA-A24 and/or HLA-DQA1*03, which showed a higher prevalence than in the normal control subjects (86.8%, 118 of 136, P < 0.0001). The cumulative incidence of complete β-cell loss differed among the patients with both HLA-A24 and HLA-DQA1*03 (n = 98), those with HLA-DQA1*03 only (n = 59), and those with HLA-A24 only (n = 19) (4.22, 2.37, and 0.90 per 100 patient-years, respectively, P = 0.012) (Fig. 1A). Patients who possessed both HLA-A24 and HLA-DQA1*03 (n = 98) showed earlier complete loss of β-cell function than the remaining patients (n = 79) (4.22 vs. 1.89 per 100 patient-years, P = 0.0057). Among 146 type 1 diabetic patients with HLA-DQA1*03, there were 225 DQ haplotypes involving DQA1*03, which had the following breakdown: 82 (36.4%) were DQA1*03-DQB1*0303, 88 (39.1%) were DQA1*03-DQB1*0401, 44 (20.0%) were DQA1*03-DQB1*0302, 6 (2.7%) were DQA1*03-DQB1*0402, and 5 (2.2%) were DQA1*03-DQB1*0301. All of the HLA-A24 alleles typed at the DNA level (n = 75) were A*2402.

Patients with HLA-DR9 (DRB1*0901) (n = 69) showed earlier complete loss of β-cell function than those without HLA-DR9 (n = 116) (4.53 vs. 2.27 per 100 patient-years, P = 0.0093) (Fig. 1B). Homozygotes for HLA-DR9 (n = 26) showed earlier complete β-cell loss than the remaining patients (n = 170) (P = 0.0004) and earlier loss than in heterozygotes for HLA-DR9 (n = 42) (P = 0.023) (7.46, 2.51, and 3.28 per 100 patient-years, respectively). A higher cumulative incidence of complete β-cell loss was also observed in patients with the DQA1*03-DQB1*0303 haplotype (n = 60), which is in linkage disequilibrium with HLA-DR9 (27), compared with those without this haplotype (n = 103) (4.33 vs. 2.55 per 100 patient-years, P = 0.0497). The frequency of HLA-A24 plus HLA-DQA1*03 did not differ between patients with HLA-DR9 (57.4%, 39 of 68) and patients without HLA-DR9 (55.1%, 59 of 107) (P = 0.88).

There were no other HLA alleles or haplotypes that showed an association with earlier complete β-cell loss, including HLA-A24 (P = 0.17), HLA-DQA1*03 (P = 0.11), HLA-A2 (P = 0.33), HLA-B54 (P = 0.49), HLA-DR4 (P = 0.80), DQA1*03-DQB1*0302 (P = 0.31), DQA1*03-DQB1*0401 (P = 0.78), and other alleles or haplotypes (HLA-A11, -A26, -A31, -A33, -B35, -B39, -B44, -B52, -B60, -B61, -B62, -Cw1, -Cw3, -Cw4, -Cw7, -DR1, -DR6, -DR8, and DQA1*0102-DQB1*0604 [data not shown]). Accordingly, the combined effect of HLA-A24, HLA-DQA1*03, and HLA-DR9 was examined. The patients who possessed all of HLA-A24, HLA-DQA1*03, and HLA-DR9 (n = 37) showed complete loss of β-cell function much earlier than those who did not have this three-allele combination (n = 136) (6.19 vs. 2.45 per 100 patient-years, P = 0.0011) (Fig. 1C).

In contrast, the incidence of complete β-cell destruction tended to be lower in patients with HLA-DR2 (n = 26) than in those without HLA-DR2 (n = 159), but the difference was not significant (1.75 vs. 3.31 per 100 patient-years, P = 0.077). When we excluded patients with fulminant type 1 diabetes (n = 9) from the analysis, possession of HLA-DR2 (n = 21) was associated with a lower rate of progression to complete β-cell loss than in patients without HLA-DR2 (n = 155) (0.95 vs. 3.08 per 100 patient-years, P = 0.011) (Fig. 1D). Among the 21 patients with HLA-DR2, 10 patients had the DRB1*1501-DQA1*0102-DQB1*0602 haplotype, 10 patients had the DRB1*1502-DQA1*0103-DQB1*0601 haplotype, and 1 patient had the DRB1*1501-QA1*05-DQB1*0301 haplotype. The frequency of HLA-A24 plus HLA-DQA1*03, HLA-DR9 alone, and the combination of HLA-A24, HLA-DQA1*03, and HLA-DR9 did not differ between the patients with or without HLA-DR2 (data not shown).

Clinical characteristics, including sex distribution, age at onset of diabetes, BMI at onset, and time until insulin therapy, were compared between patients with or without the above-mentioned HLA alleles or allele combinations. Patients with both HLA-A24 and HLA-DQA1*03 had an older age at onset (38.5 ± 14.2 years) than the remaining patients (32.7 ± 12.1 year, P = 0.0084). The time until insulin therapy was shorter in patients with the combination of HLA-A24, HLA-DQA1*03, and HLA-DR9 (10.5 ± 16.2 months) than in the remaining patients (27.5 ± 47.2 months, P = 0.014). There were no differences in the clinical characteristics in other comparisons.

Complete β-cell destruction occurred in the order of fulminant (n = 10), acute-onset (n = 97), and slow-onset type 1 diabetes (n = 85) (15.88, 3.93, and 1.71 per 100 patient-years, respectively, P < 0.0001). Therefore, the frequencies of HLA alleles, allele combinations, and haplotypes that influenced complete β-cell destruction were investigated to search for genetic factors influencing the mode of onset by including the patients without longitudinal data on residual β-cell function (Table 1).

Although differences in the frequency of HLA-A24 did not reach significance among the three subgroups of patients (P = 0.051), its frequency was higher in acute-onset type 1 diabetes than in slow-onset type 1 diabetes (P = 0.0023). For the combination of HLA-A24 plus HLA-DQA1*03, the frequency differed among the three groups (P = 0.016) and was higher in acute-onset than slow-onset type 1 diabetes (P = 0.0068). Furthermore, when the combination of HLA-A24, HLA-DQA1*03, and HLA-DR9 was analyzed, the difference in the frequency of this three-allele combination among the three groups became striking (P = 0.0005). Frequency of this three-allele combination was also strikingly higher in acute-onset than slow-onset type 1 diabetes (P = 0.0002). Neither the frequency of patients with HLA-DR9 nor that of homozygote for HLA-DR9 showed a significant difference among or between the subgroups of patients. In comparison with normal control subjects, the alleles or allele combinations that showed an increase in patients with acute-onset type 1 diabetes were HLA-A24 plus HLA-DQA1*03 (68.5% [85 of 124] vs. 41.2% [56 of 136], P < 0.0001), HLA-DR9 alone (44.8% [60 of 134] vs. 26.5% [36 of 136], P = 0.0022, Pc = 0.020), and the combination of HLA-A24, HLA-DQA1*03, and HLA-DR9 (35.8% [44 of 123] vs. 16.9% [23/136], P = 0.0006).

Although the frequency of HLA-DR2 also differed among the three subgroups of patients (P = 0.0027), it was higher in patients with fulminant type 1 diabetes than in those with acute-onset type 1 diabetes (P = 0.0005) or slow-onset type 1 diabetes (P = 0.0058) (Table 1), whereas there was no significant difference from the frequency in normal control subjects (45.5% [5 of 11] vs. 30.9% [42 of 136], P = 0.33). The haplotype of DRB1*1501-DQA1*0102-DQB1*0602 also showed the same trends (Table 1).

To identify variables making an independent contribution to complete β-cell loss in type 1 diabetes, Cox’s proportional hazards model was applied to data from the longitudinal study of residual β-cell function in 196 patients. In this model, the response variable was the time until the occurrence of complete β-cell loss or the time until the final measurement of residual β-cell function. The covariates examined included the sex and possession of the combination of HLA-A24, HLA-DQA1*03, and HLA-DR9 as categorical variables and age at onset (range 8–73 years), BMI at onset (range 13.2–31.3 kg/m²), and time from the onset of diabetes to initiation of insulin therapy (range 0–300 months) as continuous variables. In this model, significant independent risk factors were male sex (hazard ratio 1.87 [95% CI 1.08–3.24]; P = 0.026), age at onset (1.03 per 1-year increase [1.01–1.05]; P = 0.0009), the time from the onset of diabetes to initiation of insulin therapy (0.97 per 1-month increase [0.96–0.99]; P = 0.0006), and possession of the HLA-A24, HLA-DQA1*03, and HLA-DR9 combination (1.79 [1.04–3.08]; P = 0.036). When HLA-A24 plus HLA-DQA1*03, HLA-A24 alone, HLA-DR9 alone, or HLA-DR2 alone was incorporated into this model instead of the three-allele combination (HLA-A24, HLA-DQA1*03, and HLA-DR9), neither the two-allele combination nor the individual HLA alleles were a significant risk factor (P = 0.061, 0.33, 0.059, and 0.11, respectively).

In the present study, the HLA alleles that contributed to both early complete β-cell destruction and acute onset of diabetes were the two-allele combination of HLA-A24 and HLA-DQA1*03 and the three-allele combination of HLA-A24, HLA-DQA1*03, and HLA-DR9. The three-allele combination had a stronger influence than the two-allele combination. Multivariate analysis showed that only the three-allele combination of HLA-A24, HLA-DQA1*03, and HLA-DR9 was a significant independent risk factor for complete β-cell loss among the HLA alleles tested as covariates. Each HLA allele in this three-allele combination is already known as a risk allele for type 1 diabetes or for β-cell destruction. HLA-A24 (A*2402) is associated with complete β-cell destruction in type 1 diabetic patients who have other susceptible HLA alleles (2), whereas HLA-A24 and HLA-DQ haplotypes involving HLA-DQA1*03 accelerate β-cell destruction in an additive manner (13). In addition, not only HLA-DQ haplotypes involving HLA-DQA1*03 but also HLA-A24 itself contribute to susceptibility to type 1 diabetes (13–16). Besides the contribution of HLA-DR9 to susceptibility to type 1 diabetes (11,12), HLA-DR9-DQ9 was associated with lower serum C-peptide levels in type 1 diabetic patients compared with HLA-DR4-DQ4 over a relatively short disease duration (∼3 years) (31). Although HLA-DR9, especially in homozygotes, contributed to early complete β-cell destruction in our longitudinal study, it did not discriminate acute-onset from slow-onset type 1 diabetes despite an increased frequency in acute-onset patients compared with normal control subjects, as reported previously (32). Taking these findings into consideration, our results indicate that a combination of HLA class I and class II alleles, each of which contributes to susceptibility to type 1 diabetes or promotes β-cell destruction, results in the further potentiation of β-cell destruction in type 1 diabetes.

The mechanism by which accumulation of specific HLA class I and class II (DR and DQ) alleles causes faster and more severe destruction of β-cells in type 1 diabetes remains unclear. MHC class II molecules on antigen-presenting cells present antigenic peptides to CD4⁺ T-cells, whereas MHC class I molecules on target cells present antigenic peptides to CD8⁺ T-cells (33). In addition, HLA-DR and -DQ molecules show functional complementarity in antigen presentation (34). In fact, the profiles of peptide eluted from HLA-DR and -DQ molecules after pulsing with islet antigens are different (35). Thus, the combination of specific HLA-A, -DR, and -DQ alleles may have synergic and complementary effects on various steps of the immune response. However, the frequency of this three-allele combination was only 36% in acute-onset type 1 diabetes, so there remains the possibility that non-HLA genes (10) or environmental factors such as viral infection (36) also promote β-cell destruction in type 1 diabetes.

Another finding of this study was the contrasting distribution of HLA-DR2 between type 1B (fulminant type 1) diabetes and type 1A (acute-onset and slow-onset type 1) diabetes. In fulminant type 1 diabetes, the frequency of HLA-DR2 was not lower than in normal control subjects and was thus higher than in acute-onset or slow-onset type 1 diabetes. Although our study population only included a small number of patients with fulminant type 1 diabetes, an increased frequency of HLA-DR2 in fulminant type 1 diabetes compared with acute-onset type 1 diabetes was also detected by a nationwide survey performed in Japan (32). This difference in the frequency of HLA-DR2 between fulminant type 1 diabetes and type 1A diabetes may be due to a different pathogenesis of the former (37). In contrast, HLA-DR2 was protective against early complete β-cell destruction in type 1A diabetes, a result that was consistent with the findings of our previous cross-sectional study (21).

As clinical parameters, male sex, an older age at onset, and shorter time until insulin therapy were independent risk factors for early complete β-cell destruction in type 1 diabetes. Male sex was previously shown to be a risk factor for insulin dependence in LADA (38), but the association of an older age at onset with early complete β-cell destruction contradicts the results of a previous large-scale study (3). This may have been due to a wide-ranging age of onset in our study population as a result of including patients with fulminant type 1 diabetes, which develops later than acute-onset type 1 diabetes (39). The higher age of onset in our patients compared with Caucasians may also be due to the inclusion of patients with LADA. A close relationship between the mode of onset of diabetes and the rate of β-cell destruction is consistent with our cross-sectional study that showed preservation of residual β-cell function in LADA compared with acute-onset type 1 diabetes (29). Because our patient population was limited to Japanese subjects, studies performed in other racial groups (including patients with fulminant type 1 diabetes or LADA) will be necessary to confirm the cumulative effect of HLA alleles presented in this study.

Although our study demonstrated the temporal profile of β-cell destruction in type 1 diabetes and detected an HLA allele combination that contributed to early and acute complete β-cell destruction, its retrospective design and one-point determination of serum C-peptide are limitations with respect to delineating longitudinal changes of residual β-cell function. Complete β-cell destruction could have occurred at an earlier time than when it was detected. The potential bias may also be introduced by the exclusion of the patients whose first C-peptide test revealed complete β-cell loss at >5 years after the onset of diabetes from the longitudinal study. Long-term prospective studies on residual β-cell function using the stimulated CPR would be preferable, as suggested by the ADA workshop report (40).

In summary, we demonstrated that a combination of specific HLA class I and class II alleles contributed to acute and early complete β-cell destruction in Japanese patients with type 1 diabetes.

This work was partly supported by the Japanese Ministry of Education, Science, and Culture (Grant 14571117).

We thank Fumie Takano for her secretarial work.

Parts of this work were presented at the 64th scientific session of the American Diabetes Association, Orlando, Florida, 4–8 June 2004 and at the 41st annual meeting of the European Association for the Study of Diabetes, Athens, Greece, 12–15 September 2005.