Introduction

Gestational diabetes (GDM) has an onset during pregnancy and complicates approximately 2% of all pregnancies in Scandinavia [1]. Genetic factors have been implicated in the pathogenesis of GDM, yet specific genes have not been identified [1]. HLA genes of the MHC have been related to GDM in several studies. Some have demonstrated a positive association between GDM and HLA class II alleles [2], while others have not [3]. If GDM is to be properly classified, it is necessary to genetically characterise it and make distinctions from other types of diabetes. The aim of the present work was to investigate whether there is an association between GDM and HLA-DQ genotypes associated with type 1 diabetes.

Methods

Population

Mothers and children were prospectively recruited from the Diabetes Prediction in Skåne (DiPiS study) [4]. Between September 2000 and August 2004, cord blood from newborns and venous blood from mothers were obtained as dried blood spots in the delivery room after informed consent had been obtained.

All pregnant women in Skåne were invited to undergo a 75 g OGTT at 25–28 weeks of pregnancy or at 10–12 weeks of pregnancy for those who had had GDM in earlier pregnancies or a parental history of diabetes. GDM was defined as a capillary 2 h glucose concentration ≥ 9.0 mmol/l. The midwife recorded the health status of the mother (diabetes or gestational diabetes) on the remittance for dried blood spots.

Study group

From more than 35,000 deliveries, 902 mothers were classified as having GDM at least once. Of the GDM mothers, 70 gave birth to a child with the presence of autoantibodies to GAD65, insulinoma-associated protein 2 (IA-2) (>96th percentile of newborns) or insulin (>99th percentile of newborns). A further three children could not be analysed for all three autoantibodies. These 73 mothers with GDM were excluded. HLA genotypes were available in 764 (92%) of the remaining 829 mothers. As controls, 1191 additional unique mothers were randomly selected and included on the basis that they had no diabetes or autoantibodies during any of their pregnancies and could be analysed for HLA. The HLA distribution of the controls was compared with that of the newborn population. The Lund University Ethics Committee approved the DiPiS study.

HLA genotyping

HLA-DQB1 genotypes were analysed and divided into five groups based on the association with type 1 diabetes, as previously described [4].

Autoantibody measurement

Analysis of autoantibodies to GAD65 or IA-2 and antibodies against insulin were determined in cord blood of newborns [5]. Autoantibodies were determined in a first combined screen in which eluates from dried blood spots were incubated overnight in duplicate with labelled antigen. Antibody-bound antigen was precipitated with protein A–Sepharose and unbound antigen was removed by washing. The radioactivity of antibody-bound GAD65 and/or IA-2 was counted in a beta-counter. The combined screen (COMB) compared a positive reference with two negative reference samples in order to determine high levels. The diagnostic sensitivity for GAD65 autoantibodies using a cut-off at 31 radioactivity units (RU)/ml was 68% (1,317/1,950 consecutively diagnosed [during May 2005 to September 2008] type 1 diabetes patients from all over Sweden). In the IA-2 autoantibody assay using a cut-off of 5 RU/ml, the diagnostic sensitivity was 75% (1,465/1,950 patients). The GAD65 and IA-2 autoantibody assays showed mean inter-assay and intra-assay coefficients of variation of 14% and 8% respectively. Samples greater than the 99th percentile of COMB were individually analysed for each antibody. In children born after 2001 with high individual GAD65 or IA-2 autoantibodies (above the 99th percentile), the dried blood spot sample from the mother was also analysed for the two autoantibodies. The 99th percentile was defined on the entire population.

Insulin autoantibodies were analysed in cord blood serum samples, which were incubated in duplicate wells in 96-well plates. Antibody-bound and free-labelled insulin were separated with 40% protein A–Sepharose and the radioactivity was measured in a beta-counter. Results were expressed in arbitrary units and all samples above the 99th percentile were reanalysed in duplicate wells with 8 U/ml cold insulin to identify serum samples with non-specific binding. The data were expressed in relative units based on the degree of blocking 125I-labelled insulin to the highest positive reference standard by cold insulin, and the inter- and intra-assay coefficients of variations were 6.8–7.8% and 5.2–7.8% respectively. Samples above the 99th percentile were considered to have high levels.

Statistical analysis

Differences in proportions between categorical groups were examined for significance using the exact χ 2 test. Logistic regression on diabetes status was used to examine whether frequencies of type 1 diabetes-associated HLA-DQB1 alleles differed between controls and GDM mothers. Our study was capable of detecting odds ratios below 0.7 and above 1.35 with 80% power. Maternal age, country of birth, number of pregnancies, number of siblings and weight gain during pregnancy were considered as confounders and were adjusted in a multiple logistic regression. Bonferroni correction was used in order to adjust for multiple comparisons.

Results

Maternal characteristics

Mothers with GDM were older (p < 0.0001), more often born outside Sweden (p < 0.0001), had had fewer pregnancies (p = 0.0002), had had more pregnancies before the screening period (p < 0.0001) and gained less than 15 kg in body weight during at least one pregnancy (p < 0.0001) than control mothers (see Electronic supplementary material [ESM] Table 1)

HLA distribution

The frequency of HLA-DQ genotypes containing DQB1*0302 or DQA1*0501-B1*0201, together with either DQB1*0604 or any other allele except *0301,*0302,*0201,*0602,*0603 or *0604, did not differ between GDM and control mothers (ESM Table 2).

The DQB1*0602 allele was less prevalent (OR 0.64, 95% CI 0.51–0.80, p = 0.0006) in GDM than in control mothers (Table 1). The frequency of the DQB1*0301 allele was increased in GDM mothers compared with controls, but after adjusting for maternal age, country of birth, number of pregnancies/siblings and maternal weight gain there was no significant association between GDM and DQB1*0301 (Table 2). Since this allele is present in several haplotypes (DQA1*0302, DQA1*0501), the positive association did not remain after correcting for the number of DQB1*0301-containing haplotypes. In contrast, the difference in DQB1*0602 frequency between GDM mothers and controls remained after multiple logistic regression analysis (OR 0.67, 95% CI 0.51–0.88, p = 0.009) (Table 2).

Table 1 Comparison of the distribution of HLA-DQB1 alleles associated with type 1 diabetes in mothers with GDM and control mothers
Full size table
Table 2 Association between HLA-DQ genotypes and mothers with GDM after adjusting for number of pregnancies/siblings, age of mother in 2000, country of birth and pregnancy weight gain
Full size table

Discussion

The present 4 year study of more than 700 GDM mothers showed that DQB1*0602 was negatively associated with GDM. As in a previous report [6], we interpret the result to mean that DQB1*0602-positive mothers would be less likely than DQB1*0602-negative mothers to develop GDM. In type 1 diabetes it is known that the DQB1*0602 allele is protective, although this protection decreases with increasing age [7]. The loss of protection from DQB1*0602 is important to the risk of developing mostly type 2, but also type 1 diabetes, in post-partum GDM mothers.

The strength of our study is that all participants were ascertained in one region of Sweden where we have access to all mothers giving birth and where there is a mandatory screening programme for GDM. The HLA distribution of controls was similar to that of 34,710 newborns analysed for HLA. The HLA distribution of controls can therefore be considered as representing the general population. Our study may therefore be viewed as population-based because we ascertained that approximately 2% of all mothers had GDM.

Adjusting for number of pregnancies, maternal age, country of birth and weight gain during pregnancy did not change the results for DQB1*0602. The provisional association between GDM and DQB1*0301 did not remain after multiple logistic regression, supporting the idea that DQB1*0301 may represent different haplotypes in addition to the confounding factors. In contrast, the DQB1*0602 association was still significant after adjustment, supporting the idea that this particular allele is under-represented in women with GDM.

Previous studies have investigated the association between GDM and HLA. The same negative association between DR2, which is in linkage disequilibrium with DQB1*0602, and GDM has been reported in some studies [6, 8]. Most previous studies, however, have reported an association with the high-risk HLA-DR3 and -DR4. A higher frequency of the type 1 diabetes high-risk HLA genotypes was observed in patients with GDM, but only in those women that were found to be positive for islet autoantibody at delivery [9]. When the GDM population was classified according to the presence or absence of islet cell autoantibodies, a higher frequency of HLA-DR3/DR4 in the autoimmune group and of HLA-DR7-DQ2/γ, -DR9-DQ9/γ and -DR14-DQ5/γ were observed in the non-autoimmune group [10]. In our study, we excluded antibody-positive women with GDM and antibody-positive control mothers in order to exclude women who would be considered at risk of developing type 1 diabetes in the future.

In conclusion, the frequency of type 1 diabetes high-risk HLA-DQ alleles (DQB1*0201, B1*0302) did not differ between GDM mothers and controls. The low-risk DQB1*0602 allele was less prevalent. Maternal age, country of birth, pregnancy weight gain and number of pregnancies/siblings did not explain the difference in DQB1*0602 frequency between GDM mothers and controls.