Introduction

Gestational diabetes mellitus (GDM)—glucose intolerance of varying degrees of severity that is first recognised during pregnancy [1]—affects 2–14% of all pregnancies [24]. GDM may have the same genetic background as type 2 diabetes mellitus as there is evidence for the clustering of type 2 diabetes and impaired glucose tolerance in families with women with GDM [5] and for a higher prevalence of type 2 diabetes in mothers of women with GDM [6]. Furthermore, it is well known that women with GDM are at a greater risk of developing type 2 diabetes later in life [79]. However, knowledge regarding the genetics of GDM is limited [7, 10]. Several studies have examined candidate genes in women with and without GDM. Positive associations were shown for genes encoding glucokinase [11], calpain-10 [12], sulfonylurea receptor 1 [13], potassium inwardly rectifying channel, subfamily J, member 11 (KCNJ11) [14], β3 adrenergic receptor [15], plasminogen activator inhibitor 1 [16] and transcription factor 7-like 2 (TCF7L2) [17, 18]. Except for the effects of TCF7L2 in Scandinavian women [17], no robust associations of genetic variants with GDM have been demonstrated.

In addition to those in KCNJ11, PPARG and TCF7L2, the recent genome-wide association (GWA) studies have identified new genetic variants with reproducible associations with susceptibility to type 2 diabetes [1923], the majority of which were found in genes that were not even considered candidates [24]. Furthermore, robust signals (p < 5 × 10−7) were identified in certain gene regions (i.e. the HHEXIDE and CDKN2A–CDKN2B regions) [1923]. We recently found that the diabetogenic genetic variants reported by the large-scale GWA studies in Europids [1923] were also associated with the risk of type 2 diabetes in Asian populations, including Koreans [25].

If GDM and type 2 diabetes share a common genetic background, the genetic variants determining the risk of type 2 diabetes may also be associated with GDM. In the present study, we compared the genotype frequencies of the single nucleotide polymorphisms (SNPs) in the diabetogenic genes in GDM patients with those in non-diabetic controls.

Methods

Patients with GDM

This study included 869 Korean women diagnosed with GDM at the Samsung Cheil Hospital (Seoul, Korea) between January 1996 and February 2003. During the study period, 39,190 consecutive women underwent screening for GDM. We followed a previously described protocol for the screening and diagnosis of GDM [3, 26]. In brief, all pregnant women without a previous diagnosis of glucose intolerance were screened for GDM between 24 and 28 weeks of gestation by using the 50 g, 1 h glucose challenge test as recommended by the Third International Workshop-Conference on GDM [1]. A plasma glucose concentration of 7.2 mmol/l or more was considered positive for GDM and was followed by a 100 g OGTT. GDM was diagnosed according to the criteria of the Third International Workshop-Conference on GDM [1]. The threshold glucose values were as follows: fasting ≥5.8 mmol/l, 1 h  ≥ 10.5 mmol/l, 2 h  ≥ 9.2 mmol/l and 3 h  ≥ 8.0 mmol/l.

Gestational age at the time of the screening test was 27.9 ± 2.9 weeks (mean ± SD). Plasma glucose concentration was measured by the glucose oxidase method using an YSI 2300 STAT analyser (Yellow Springs Instrument Company, Yellow Springs, OH, USA). Serum insulin concentration was measured using insulin-specific radioimmunoassay kits (Linco Research, St Louis, MO, USA). Homeostasis model assessment of insulin resistance (HOMA-IR) was calculated using the formula developed by Matthews et al. [27]. GAD antibodies were measured using a radioimmunoassay method (RSR, Cardiff, UK).

The clinical characteristics of women with GDM at 6 weeks postpartum are shown in Table 1. Height and body weight were measured to the nearest 0.1 cm and 0.1 kg, respectively, with the patient standing barefoot and in light clothing. BMI was calculated as body weight in kilograms divided by the square of height in metres. Blood pressure was measured after the participant had remained seated for 10 min. Measurements were taken twice, 5 min apart. The frequency of a positive family history of type 2 diabetes in women in the GDM group was 41.3%. Women who tested positive for GAD antibodies were excluded from the study.

Table 1 Clinical characteristics of the study participants

Non-diabetic controls

We recruited 632 non-diabetic controls (345 women, 287 men) who visited Seoul National University Hospital for a routine health check-up. The following selection criteria were used: age ≥60 years, no history of type 2 diabetes, no first-degree relatives with type 2 diabetes, fasting plasma glucose level <6.1 mmol/l and HbA1c level <5.8%. Therefore, controls were expected to be at a very low risk of type 2 diabetes. Table 1 shows the clinical characteristics of the controls in detail. It was not confirmed whether the non-diabetic female participants had experienced pregnancy without GDM. However, the ascertainment bias should be minimal given that the prevalence of GDM is estimated to be very low in Korea, at a rate of 2.2 cases per 100 pregnant women [3].

Ethical considerations

The Institutional Review Board of the Clinical Research Institute of Seoul National University Hospital and the Research and Ethics Committee of the Samsung Cheil Hospital approved the study protocol, and informed consent was obtained from each study participant. The study was carried out in accordance with the principles of the Declaration of Helsinki as revised in 2000.

Gene and SNP selection

SNPs from six novel genetic loci, which were identified through the recent GWA studies [1923] and showed reproducible associations with type 2 diabetes in Europeans, were selected for this study. Specifically, these were rs7756992 and rs7754840 in CDKAL1; rs564398 and rs10811661 in CDKN2A–CDKN2B; rs8050136 in FTO; rs1111875, rs5015480 and rs7923837 in HHEX; rs4402960 in IGF2BP2; and rs13266634 in SLC30A8. Two representative SNPs (rs1333040 and rs10757278) close to CDKN2A–CDKN2B that were associated with coronary heart disease and myocardial infarction were also selected [2830]. In addition, rs7903146 and rs12255372 in TCF7L2 [31]; rs5215 and rs5219 in KCNJ11 [32]; and rs3856806 and rs1801282 in PPARG [33] were also genotyped.

Genotyping

We genotyped 18 SNPs in nine genes from genomic DNA. An allelic discrimination assay was performed in 5 μl of 1× TaqMan Universal PCR Master Mix (Applied Biosystems, Foster City, CA, USA) containing 20 ng of genomic DNA and 0.125 μl of 40× Assays-on-Demand SNP genotyping product (Applied Biosystems), according to the manufacturer’s instructions. (A list of the assay ID numbers and public ID numbers for the Assays-on-Demand SNP genotyping products used in this study is presented in Electronic supplementary material [ESM] Table 1.) Next, the plate was placed in a thermal cycler (PE 9700; Applied Biosystems) and heated for 2 min at 50°C and 10 min at 95°C, followed by 40 cycles at 95°C for 15 s and 60°C for 1 min. The TaqMan assay plate was then transferred to a Prism 7900HT instrument (Applied Biosystems) and the fluorescence intensity of each well was read. The fluorescence data files for each plate were analysed using Sequence Detection System version 2.1 (Applied Biosystems). Forty-four blind duplicates and eight blank samples were used as positive and negative controls, respectively. The overall genotyping success rate was 99.4% (range 97.4−99.8%; ESM Table 2) and the concordance rate based on blind duplicate comparisons was 99.6%.

Statistical analyses

All continuous variables are expressed as mean±SD. Statistical analyses were conducted using SPSS for Windows, version 11.0 (SPSS, Chicago, IL, USA). To determine whether individual polymorphisms were in Hardy–Weinberg equilibrium, χ2 tests were used. The result of the test for Hardy–Weinberg equilibrium and the statistical power for each variant are shown in ESM Table 2. Since there were no differences in the genotype frequencies between male and female non-diabetic controls (ESM Table 3), we initially compared the genotype frequencies of the 869 GDM patients with those of the 632 controls (i.e. both men and women), which was considered to be a non-diabetic gene pool. In addition, we compared the genotype frequencies of the 869 GDM patients with those of the 345 female controls. Logistic regression analyses were used for calculating the ORs, 95% CIs and the corresponding p values with regard to the number of risk alleles using an additive model. Genotypes were given codes of 0, 1 and 2, and the OR was expressed per difference in the number of risk alleles. For the comparison of representative SNPs for type 2 diabetes and GDM in European and Korean populations, the heterogeneity of ORs among studies or populations was assessed by Cochran’s Q statistic, as shown elsewhere [25]. Insulin resistance was evaluated using the HOMA-IR, and insulin secretory function was assessed by determination of the insulin AUC during a 100 g OGTT performed at the time of diagnosis of GDM. Multivariate linear regressions, adjusted for age and BMI, were used for comparing insulin resistance and insulin secretory function in GDM patients according to genotype. A p value of less than 0.05 was considered statistically significant.

Results

We compared the genotype frequencies in the 869 GDM patients with those in the 632 non-diabetic controls (345 women, 287 men) who were expected to be at very low risk of type 2 diabetes. All SNPs were in Hardy–Weinberg equilibrium (ESM Table 2). Compared with the 632 controls, we found that GDM was significantly associated with rs7756992 (p = 9.14 × 10−9) and rs7754840 (p = 4.17 × 10−9) in CDKAL1; rs10811661 (p = 1.05 × 10−7) in CDKN2A–CDKN2B; rs1111875 (p = 0.003), rs5015480 (p = 0.035), and rs7923837 (p = 0.011) in HHEX; rs4402960 (p = 0.034) in IGF2BP2; rs13266634 (p = 0.005) in SLC30A8; and rs7903146 (p = 0.038) in TCF7L2 (Table 2). In this analysis, three SNPs (rs7756992 and rs7754840 in CDKAL1 and rs10811661 in CDKN2A–CDKN2B) met the stringent criterion for robust association (p < 5 × 10−7) [19], and their ORs ranged from 1.39 to 1.55. When we compared the genotype frequencies in the GDM patients with those in the 345 female controls, rs7756992 (p = 0.001) and rs7754840 (p = 1.36 × 10−5) in CDKAL1; rs10811661 (p = 1.42 × 10–7) in CDKN2A–CDKN2B; and rs13266634 (p = 0.044) in SLC30A8 showed significant associations with GDM (ESM Table 4).

Table 2 Comparison of genotype frequencies between GDM patients and non-diabetic controls

Table 3 compares the ORs for the associations between the variants and GDM in our study with those previously reported for the associations of these variants with type 2 diabetes in European and Korean populations [20, 22, 23, 25, 47]. There were no significant differences in ORs between type 2 diabetes and GDM in Koreans. Although the effects on the risk of diabetes were in the same direction, the effect sizes of rs7754840 and rs10811661 were slightly greater for GDM in Koreans than for type 2 diabetes in Europeans.

Table 3 Comparison of significant genetic variants associated with type 2 diabetes and GDM in European and Korean populations

Next, we examined the associations between the risk alleles and insulin resistance (i.e. HOMA-IR) and insulin secretory capacity (i.e. AUC of insulin during a 100 g OGTT; Table 4). No SNPs were significantly associated with HOMA-IR, with the exception of rs7754840 in CDKAL1 and rs1111875 in HHEX, which showed modest associations (adjusted p = 0.049 and 0.026, respectively). The risk alleles of the SNPs rs7756992 and rs7754840 in CDKAL1 and rs10811661 in CDKN2A–CDKN2B, which, as mentioned above, showed robust associations with GDM, were associated with a marked decrease in AUC of insulin during a 100 g OGTT. Interestingly, the risk alleles of rs1111875, rs5015480 and rs7923837 in HHEX were also significantly associated with a reduced insulin AUC (adjusted p = 0.0000002, 0.0002, and 0.006, respectively). In contrast, the risk allele of rs8050136 in the FTO gene was associated with an increased AUC of insulin (adjusted p = 0.006) but was not associated with HOMA-IR.

Table 4 Associations between risk alleles and insulin resistance and insulin secretory function

Discussion

We found that some of the SNPs recently identified as genetic determinants of type 2 diabetes by GWA studies [1923] were also associated with GDM in Koreans. Three SNPs in particular (rs7756992 and rs7754840 in CDKAL1 and rs10811661 in CDKN2A–CDKN2B) were very strongly associated with GDM. These same SNPs were significantly associated with insulin secretory capacity as assessed by the insulin AUC during a 100 g OGTT performed at the time of the diagnosis of GDM. However, they did not show any robust associations with insulin resistance. Their association with pancreatic beta cell function is consistent with the results obtained in a study on type 2 diabetes in the Japanese population, which showed that risk alleles at CDKAL1 (rs7756992) and CDKN2A–CDKN2B (rs10811661) were associated with impaired beta cell function [34].

It is well known that normal pregnancy is accompanied by a marked increase in insulin resistance, which may be the result of both increased maternal adiposity and the insulin-antagonising effects of several placental hormones [7]. Therefore, maternal pancreatic beta cell compensation is crucial for overcoming the insulin resistance provoked by pregnancy and for maintaining the metabolic balance during pregnancy. It has recently been shown that prolactin represses islet menin levels and stimulates beta cell proliferation during pregnancy in mice. It was found that the transgenic expression of the gene encoding menin in maternal beta cells inhibited islet expansion and led to the development of GDM phenotypes [35]. In human studies, inadequate compensatory insulin secretion in the face of increased insulin resistance has consistently been observed in patients with GDM [3638]. In this regard, the risk alleles of rs7756992 and rs7754840 in CDKAL1 and rs10811661 in CDKN2A–CDKN2B may play crucial roles in the pathogenesis of GDM through impaired compensatory insulin secretion by pancreatic beta cells.

CDKAL1 is expressed in human pancreatic islets and shows considerable homology with CDK5RAP1, a well-known inhibitor of cyclin-dependent kinase 5 (CDK5) activation [23]. CDK5 has been suggested to downregulate insulin expression through the formation of p35/CDK5 complexes [39, 40]. In addition, CDK5 transduces glucose toxicity signals in pancreatic beta cells [39].

The 5′ sequence upstream of rs10811661 contains CDKN2B and CDKN2A (encoding p15INK4b and p16INK4a, respectively) [23]. p16INK4a is known to inhibit CDK4, a powerful regulator of pancreatic beta cell replication [4144]. CDKN2B overexpression was found to be related to islet hypoplasia and diabetes mellitus in murine models [45]. Furthermore, both CDKN2A and CDKN2B are expressed at high levels in pancreatic islets [23]. It was reported that the SNPs in this region showed a stronger signal as a haplotype [23], but we could not find such a trend in our study (ESM Tables 5 and 6)

In our study, the association of TCF7L2 with GDM was not as strong as that reported by the Scandinavian study [17]. This discrepancy may be explained by the difference in the frequency of the rs7903146 risk allele between the two studies (risk allele frequency in the control group was 0.238 vs 0.025, respectively), even though the effect size was similar (OR 1.49, 95% CI 1.28–1.75, p = 4.9 × 10−7 vs OR 1.58, 95% CI 1.03–2.43, p = 0.038, respectively).

Although HHEX was not robustly associated with GDM, it was strongly associated with insulin secretory capacity. Interestingly, the protective allele of rs1111875 in HHEX was very modestly associated with increased insulin resistance (adjusted p = 0.026) but very strongly associated with increased insulin secretory capacity (adjusted p = 0.0000002). This finding suggests that rs1111875 may be more likely to be one of the genetic factors involved in beta cell compensatory insulin secretion in the face of insulin resistance induced by pregnancy. It is known that HHEX is expressed at high levels in the fetal and adult pancreas [23] and is crucial for the development of the ventral pancreas [46]. In Europids [2023], rs1111875 is found within an extended region of linkage disequilibrium that contains not only HHEX but also KIF11 and IDE. Therefore, the influence of KIF11 and/or IDE cannot be ignored.

Although FTO did not show any significant association with GDM or insulin resistance, the risk allele of rs8050136 in the FTO gene was found to be associated with an increased insulin AUC (age- and BMI-adjusted p = 0.006). In addition, the pre-pregnancy BMI did not differ according to the rs8050136 genotype (data not shown). We suggest that rs8050136 does not increase the risk of GDM but may afford protective by increasing insulin secretory capacity, at least in the Korean population.

We found that the risk allele frequencies of the SNPs in patients with GDM did not differ from those in patients with type 2 diabetes in the Korean population (Table 3). In this regard, some genetic factors (especially, rs7756992 and rs7754840 in CDKAL1 and rs10811661 in CDKN2A–CDKN2B) are associated with GDM in particular, as well as type 2 diabetes in general.

We recently confirmed that these SNPs were associated with the risk of type 2 diabetes in Koreans, showing similar effect sizes to those in Europeans, although the risk allele frequencies of most of these SNPs were different between populations [25]. The effect sizes of rs7754840 and rs10811661 were slightly greater in Korean women with GDM than in Europeans with type 2 diabetes, although the effects were in the same direction.

This study is subject to certain limitations. We compared the genotype frequencies in GDM patients with those in non-pregnant, non-diabetic controls. The results of the current study may therefore be regarded as a comparison of genotype frequency among gene pools of patients with GDM and individuals with a very low risk of type 2 diabetes. A control group consisting of age- and BMI-matched pregnant women without GDM may be more suitable for identification of the GDM susceptibility genes. It was not confirmed whether the non-diabetic female controls who were enrolled in this study had experienced pregnancy without GDM. However, the ascertainment bias should be minimal because the prevalence of GDM is estimated to be very low (2.2 cases per 100 pregnant women in Korea) [3]. Our study was underpowered to detect associations of some of the SNPs with GDM (see ESM Table 2), probably because of their low frequencies, which may have resulted in some associations being overlooked.

In conclusion, some of the type 2 diabetes-associated genetic variants discovered in the recent GWA studies are also associated with GDM in Koreans. Further studies need to be conducted to examine whether these risk variants predict the development of type 2 diabetes later in life.