DPB1 Alleles Are Associated With Type 1 Diabetes Susceptibility in Multiple Ethnic Groups

Mexican and Puerto Rican sample sets.

Table 1 shows the results of allele frequency determination for the DPB1 locus in 91 patients and 82 control subjects from Puerto Rico and in 84 Mexican-American simplex families. In the Puerto Rican data, significant association was seen for DPB1*0301 (P = 0.024). The data were suggestive, although not statistically significant, of a positive disease association with the allele DPB1*0202 (P = 0.058). DPB1*1701 also appeared positively associated with type 1 diabetes in these data (P = 0.009); however, negative type 1 diabetes association for DPB1*1701 was previously reported (10) for Caucasian families. This difference may be attributable to differences in LD patterns between populations. For Mexican-American families, expected associations were observed: positive association for DPB1*0202 (P = 0.027) and DPB1*0301 (P = 0.002) and negative association for DPB1*0402 (P = 0.017).

Because of the strong LD within the HLA region, some apparent associations may simply be “bystander” effects caused by LD with highly predisposing or protective DR- and DQ-encoding loci. DRB1 and DQB1 genotypes were determined for all individuals in both datasets. Values were calculated for LD between DRB1-DQB1 haplotypes and DPB1 alleles (Table 2).

In both the Mexican-American and the Puerto Rican populations, DPB1*0401 was in positive LD with DRB1*0101-DQB1*0501. Other LD patterns differed between the two populations. DPB1*0402 was in positive LD with DRB1*0701 in the Puerto Rican data but in negative LD with DRB1*0701 in the Mexican-American data. DPB1*0402 had a clear protective effect among Mexican Americans, but the reduced frequency in patients was not significant in Puerto Ricans. The very high positive LD with DRB1*0701 (type 1 diabetes protective) may reduce the power to detect any additional protective effect of DPB1*0402 in Puerto Ricans. No significant LD was observed for DPB1*0202 in either population, although this may be attributable to a lack of power resulting from small sample size rather than the absence of LD. DPB1*0202 is found primarily on DR3 haplotypes in Caucasians (10); it is found on multiple haplotypes in Filipinos, where it shows a strong type 1 diabetes association (8). DPB1*0301 exhibited LD with DRB1*1302-DQB1*0604 (neutral to slightly predisposing) in the Puerto Rican data but showed no significant LD in the Mexican-American data. This suggests that DPB1 associations observed in these data are not attributable solely to LD of DPB1 alleles with DRB1-DQB1 haplotypes.

Revised expected values for all DPB1 alleles were calculated based on the LD results. Deviation from the null hypothesis was tested (Table 3). For both populations, results for DPB1*0301 and DPB1*0202 still suggest a positive type 1 diabetes association after correction for LD. The novel negative association of DPB1*1701 with type 1 diabetes in the Puerto Rican data also remains significant (P = 0.011) but may be spurious or unique to this population. The protective effect of DPB1*0402 in the Mexican-American data remains (P = 0.012) after LD correction.

DAISY samples.

Table 4 shows a comparison of DPB1 allele frequencies for the two groups of DAISY samples. High type 1 diabetes risk individuals (DR3/4-DQB1*0302; DR4≠DRB1*0403) were selected from both the General Population Cohort (GPC) and the Sibling-Offspring Cohort (SOC) (see research design and methods). This study design differs from the others in that the comparison is not between patients and control subjects but between HLA-matched individuals with or without a type 1 diabetic relative. The observed higher type 1 diabetes risk for high-risk SOC subjects than for high-risk GPC subjects leads to the testable hypothesis that the SOC subjects are enriched in non–DR-DQ susceptibility alleles.

The data show that DPB1*0301 is significantly overrepresented in the SOC (P = 0.04) compared with the GPC, whereas DPB1*0402 is underrepresented in the SOC (P = 0.002) compared with the GPC, consistent with previous observations for these alleles. DPB1*0202 was rare and did not differ significantly between the two populations. To address issues of ethnic differences between the GPC and SOC, the Caucasian subset of these samples was analyzed separately. In the Caucasian-only sample, DPB1*0301 remained overrepresented in the SOC, although not significantly (P = 0.087), suggestive of a positive type 1 diabetes association. The underrepresentation of DPB1*0402 in the SOC remained significant (P = 0.01) in the Caucasian set, supporting a protective effect on type 1 diabetes susceptibility.

The HLA class II DR- and DQ-encoding loci are strongly associated with type 1 diabetes susceptibility, but even the highest-risk DR-DQ genotype (DR3/DR4-DQB1*0302) only predicts an ∼7% risk of disease in the general population. Increasing the predictive value of genetics for type 1 diabetes susceptibility requires identification of other risk factors. Because effects of other genetic susceptibility loci are expected to be modest, associations of such loci are difficult to identify and replicate and can be confounded by DR-DQ heterogeneity. The frequency of the DR3/DR4 heterozygous genotype is too low in the Mexican-American and Puerto Rican control groups to make any risk estimates based on DPB1 subgroups; the frequency of DR4/DRX genotypes was sufficient. The data suggest that for DR4/DRX individuals (X ≠ DR3 or DR4), the added risk of carrying at least one copy of either DPB1*0202 or DPB1*0301 increases the probability of developing type 1 diabetes by 60% in the Puerto Ricans and 104% in the Mexican Americans. In addition, not carrying DPB1*0402 increases the probability of developing type 1 diabetes by 16% in the Puerto Ricans and 83% in the Mexican Americans (data not shown).

Whether the DP molecules themselves directly affect type 1 diabetes susceptibility remains in question. A recent study (12) showed DPB1 associations consistent with those seen in the present study, but the authors attributed the associations to LD of DPB1 with other genetic loci in the HLA region, citing LD of DPB1 with a microsatellite marker, located beyond the HLA region and not in proximity to any candidate functional gene. The protein encoded by the DPB1 locus has a similar function to that of DR and DQ, and DPB1 associations have been replicated in multiple studies, supporting the notion that polymorphism in the DPB1 locus itself may affect type 1 diabetes susceptibility.

The results reported here support previous reports of effects for three DPB1 alleles, DPB1*0202, DPB1*0301, and DPB1*0402, on type 1 diabetes susceptibility (5,7,9–12). The relative consistency among studies suggests that these alleles represent a useful addition to genetic models of determining type 1 diabetes risk. Whether the DPB1 locus is causally involved in or merely a marker for type 1 diabetes susceptibility, the consistent associations of specific DPB1 alleles with type 1 diabetes susceptibility observed in this and other studies suggest that DPB1 genotyping can be a useful tool to increase the predictive power of genetics for type 1 diabetes susceptibility.

RESEARCH DESIGN AND METHODS

Mexican-American families.

A total of 84 Mexican-American families (proband plus one or both parents) are included. Control frequencies were determined from AFBACs (affected family-based control subjects) (14). Families were ascertained by the investigators at the Los Angeles County Hospital/University of Southern California Medical School (LAC/USC) and Children’s Hospital of Los Angeles between 1989 and 1993 (7,15,16). Type 1 diabetes was defined according to the National Diabetes Data Group. To be considered of Mexican ancestry, probands had to have a minimum of three of four grandparents born in Mexico. The study was approved by the institutional review board at LAC/USC. Written informed consent was obtained from all subjects and/or their parents before their participation. DPB1 data were previously reported for 42 of these families (7).

Puerto Rican type 1 diabetic patients and control subjects.

The Puerto Rican registry of children with diabetes was established as part of the World Health Organization’s DiaMond project (17). In 1990, an island-wide registry of new cases of childhood type 1 diabetes was established in Puerto Rico retrospective to 1985. Eligible patients (n = 91) were identified by medical record review at 57 island hospitals that admit pediatric patients. Eligibility criteria included age <15 years at diagnosis, taking insulin at discharge to a Puerto Rican residence, and having a Puerto Rican parent (18). Healthy control subjects (n = 82) from the same area were recruited. Informed consent for sample collection was approved by the institutional review board of the Ponce School of Medicine.

DAISY.

The DAISY (19) is a prospective study of the natural history of type 1 diabetes, including limited genetic screening and extensive epidemiological data collection. DAISY has two main cohorts, the GPC, ascertained as newborns, and the SOC, ascertained based on having a first-degree relative with type 1 diabetes. Individuals ascertained for the GPC, but having a first-degree relative with type 1 diabetes, were reclassified with the SOC for this study. In the GPC and SOC, 30,951 and 888 subjects were screened, respectively. The highest-risk HLA DR-DQ genotype (DR3/4-DQB1*0302; DR4≠DRB1*0403) was found in 2.3% (n = 713) of GPC and 10.1% (n = 90) of SOC subjects. Sample collection and genotyping for DAISY was done with the approval of the Colorado Multiple Institutional Review Board. The ethnic origin of DAISY subjects is as follows: non-Hispanic white, 60.5%; Hispanic, 23.5%; non-Hispanic black, 6.0%; Asian or Pacific Islander, 1.8%; Amerindian, 0.4%; and other or unknown, 7.8%.

In the DAISY study, SOC individuals with the highest-risk HLA DR-DQ genotype had a far higher risk of developing type 1 diabetes autoimmunity (risk ∼40%) than high-risk GPC individuals (risk ∼7%), suggesting that high-risk SOC individuals may be enriched in non–DR-DQ type 1 diabetes genetic risk factors. A subset of the high type 1 diabetes genetic risk GPC (n = 200) and SOC (n = 50) subjects was selected, without regard to autoantibody or disease status, and genotyped for DPB1. Four GPC samples and three SOC samples were classified as cases, indicating either type 1 diabetes or presence of autoantibodies.

Genotyping.

Molecular HLA typing data were generated using a linear array that uses the PCR/sequence-specific oligonucleotide probe (SSOP) methodology. DPB1 genotyping by SSOP and linear array technology have been previously described (6,20,21).

Haplotype estimation.

For Mexican-American families, haplotype phase was determined directly from parent data. Phase was unambiguous in all case subjects. For Puerto Rican samples, maximum likelihood haplotype frequencies were computed using an expectation-maximization algorithm (22) with the Arlequin software program (L. Excoffier, University of Geneva, Switzerland). The SD was computed with 100 bootstrap replicates.

Statistical methods.

Expected allele frequencies were computed, given LD and haplotype relative penetrances. The null hypothesis (H0) is that DPB1 allele frequencies will differ between patients and control subjects due to 1) LD between DPB1 and DRB1-DQB1 and 2) chance (sampling), consistent with a neutral role for DPB1 polymorphism relative to disease predisposition. Under H0, the expected allele frequencies at DPB1 can be computed using the equation derived by Thomson (14).

D_ij denotes pairwise LD between the ith DR-DQ haplotype and the jth DPB1 allele in the control sample, q denotes frequency in patients, and p denotes frequency in control subjects. This method relies on sampling estimates of pairwise LD between DPB1 and DRB1-DQB1 and on patient and control frequencies derived from the study samples. A sampling error will be associated with the computed value for expected DPB1 allele frequencies; larger control samples lead to smaller sampling errors. This has been taken into account in the statistical tests:

For any individual DP_i, N_p is patient sample size, N_c is control sample size, q_DPi is observed frequency of the ith allele in patients, and q′_DPi is derived from Eq. 1.