figure b

Introduction

Gestational diabetes mellitus (GDM) is an increasing health problem worldwide and is associated with short- and long-term health risks for women and their offspring [1]. GDM is a state of impaired glucose tolerance and/or increased fasting blood glucose concentrations first recognised in pregnancy [2]. Normal pregnancy is characterised by increasing peripheral insulin resistance, thereby promoting transfer of glucose and fatty acids from the mother to the fetus. These changes are exaggerated in women with GDM; raised postprandial glucose concentration and/or fasting hyperglycaemia ensue when pancreatic insulin secretion is insufficient to achieve adequate glucose clearance and suppression of hepatic glucose output, respectively [2]. This may reflect an underlying defect in insulin signalling or pancreatic function [3].

Although there is ongoing debate about optimal methods for screening and diagnosis of GDM, the International Association of Diabetes and Pregnancy Study Groups (IADPSG) has recommended universal testing of pregnant women using an OGTT at 24–28 weeks of gestation [4]. However, the logistical challenges of completing an OGTT, with the required overnight fasting, blood sampling over 2 h and frequent nausea, impacts on uptake and adequacy of testing and GDM detection, while the timing of the OGTT results in a narrow window of opportunity for intervention in the third trimester. Further, there is increasing evidence that women with GDM may have metabolic derangement from early in pregnancy prior to clinically detected glucose intolerance [5]. Thus, an evidence-based strategy that enables risk stratification for GDM in early pregnancy or even prior to conception would be beneficial.

The genetic contribution to GDM is an emerging area of research that could potentially inform future risk stratification. Numerous genome wide association studies (GWAS) have identified strong reproducible susceptibility variants for type 2 diabetes across different populations [6,7,8]. In Korean women, GWAS have identified variants in CDKAL1 and near MTNR1B that are associated with GDM [9]. Other studies have found that several variants (including CDKAL1 and near MTNR1B) are associated with increased risk of both type 2 diabetes and GDM, suggesting that these conditions may have a shared genetic background [10, 11]. This is further supported by the fact that women affected by GDM have up to a sevenfold increased risk of subsequently developing type 2 diabetes [12].

Recently, a missense variant (rs373863828, Arg457Gln, c.1370G>A) in the CREBRF gene was identified as being strongly associated with higher BMI (+1.4 kg/m2) and waist circumference (+3 cm), but lower risk of type 2 diabetes (OR 0.59) among adults of Polynesian [13, 14] and Micronesian [15] ancestry. The minor (A) allele is prevalent in New Zealand Māori and Pacific people (allelic frequency 10–27%) but exceedingly rare in other ethnic groups (0.01% in East Asians and 0.004% in Europeans) in the Genome Aggregation Database (http://gnomad.broadinstitute.org, accessed 1 October 2017) [14]. It is currently not known whether the CREBRF rs373863828 (A) variant is associated with reduced risk of GDM. Thus, our aim was to examine the association of the CREBRF missense variant with risk of GDM in Māori and Pacific women with obesity and to investigate whether the CREBRF variant genotype improves clinical risk prediction for GDM in early pregnancy within this subgroup of Māori and Pacific women.

Methods

Study population

This study was undertaken among women recruited to a nutritional intervention study known as the Healthy Mums and Babies (HUMBA) trial (www.anzctr.org.au registration no. ACTRN12615000400561) in South Auckland, New Zealand, where more than half of the maternity population is of Māori and Pacific descent [16]. The HUMBA trial is a 2 × 2 factorial randomised controlled trial that investigated whether excessive gestational weight gain in pregnant women with obesity and birthweight in their infants could be reduced by the following interventions: (1) a multi-faceted dietary intervention provided by community health workers that included text messaging compared with routine dietary advice; and/or (2) probiotics compared with placebo. The dietary intervention consisted of four home-based education sessions, reinforced with behaviour change techniques, personalised pregnancy weight gain targets and motivational text messaging three times per week. Women allocated to routine dietary advice received routine pamphlets on healthy eating in pregnancy available to all women in New Zealand. Women allocated to probiotics received a capsule of Lactobacillus rhamnosus GG and Bifidobacterium lactis BB12 (Chr.Hansen, Denmark) at a dose of 7 × 109 colony forming units daily until birth.

Women were eligible for this genetic substudy if they had a grandparent of New Zealand Māori or Pacific (Polynesian) ethnicity as determined by maternal self-report. If women had both Māori and Pacific ancestry, ethnicity was prioritised as Māori, in line with Statistics New Zealand guidelines. Women with a singleton pregnancy and BMI ≥30 kg/m2 were recruited to the HUMBA trial between 12+0 and 17+6 weeks of gestation (gestational age is given as weeksdays). Women with pre-existing diabetes (HbA1c ≥ 50 mmol/mol [≥6.7%]) in early pregnancy were excluded [17].

Ethics approval for this study was obtained from the Southern Health and Disability ethics committee, New Zealand (14/STH/205). All participants provided written informed consent for trial participation, the collection of samples and subsequent genetic analysis.

Measurement of variables

Data obtained at the recruitment visit included demographic data, family history of diabetes, maternal anthropometric measures (height, weight, waist circumference and mid-arm circumference) and BP. Anthropometric measures were repeated at 28–30 weeks’ and 36 weeks’ gestation. Gestational weight gain (GWG) was defined as mean weekly weight gain between recruitment to 28 and 36 weeks’ gestation by Institute of Medicine criteria [18]. Finger-prick non-fasting blood lipid and HbA1c testing was conducted at each study visit (recruitment,) using the Roche Cobas b 101 point-of-care system [19]. A non-fasting blood (buffy coat) specimen was collected for genetic testing in consenting women.

Participants underwent a 75 g OGTT at 24–28 weeks’ gestation before the second trial visit, including fasting, 1 h and 2 h venous blood glucose measurements. In this study, GDM was defined by IADPSG criteria (glucose concentrations: fasting ≥5.1 mmol/l; 1 h ≥10 mmol/l; or 2 h ≥8.5 mmol/l) [4]. It is unlikely that the women developed GDM after the 24–28 week timepoint, as pregnancy-related increases in insulin resistance peak by 24 weeks’ gestation. Neither of the trial interventions (dietary intervention vs routine dietary advice nor probiotics vs placebo) altered the incidence rate of GDM, and so groups were combined for analysis.

SNP design and genotyping

DNA extraction was conducted according to the manufacturer’s recommendation using the PureLink Genomic DNA Mini Kit (Invitrogen, USA). A custom designed TaqMan probe-set (Applied Biosystems, USA) was created for rs373863828 using a custom Python script (snp_design; DOI: https://doi.org/10.5281/zenodo.56250) to annotate the human genome build 37 reference sequence (ftp://ftp.ensembl.org/pub/grch37, accessed 1 August 2016) with rs373863828 and any surrounding SNPs (obtained from the NCBI dbSNP build 147 common SNP list; ftp://ftp.ncbi.nlm.nih.gov/snp): forward primer: CAAGAGAGGATGCTGAGACCAT; reverse primer: ACCATGATGTAAGCCATTTTTCTGATACA; probe 1 (VIC): TGAGTGGAACCGAGATAC probe 2 (FAM): AGTGGAACCAAGATAC. Genotyping was performed using the LightCycler 480 Real-Time PCR System in 384-well plates (Roche Applied Science, USA). Quality control measures included genotyping of non-template controls (to ensure the absence of cross-contamination and primer cross-reactivity) and genotyping of samples set as technical replicates to evaluate consistency. There was 100% successful genotyping call rate. Re-genotyping of 25% of the samples demonstrated 100% concordance.

Analysis

Statistical analyses were performed using the R v3.3.2 statistical software (within RStudio v0.99.902; www.rstudio.com). Univariable analysis was undertaken to compare clinical and biochemical risk factors among women with and without GDM and by CREBRF status (G/G vs G/A or A/A), using χ2 test for categorical data and t test for continuous data. Logistic regression was used to test for an association between rs373863828 minor allele (c.1370A, p.457Gln) and GDM. Māori and Pacific women were analysed separately and the effects were combined using an inverse-variance-weighted fixed-effect meta-analysis. Heterogeneity among sample sets was assessed using Cochran’s heterogeneity (Q) statistics. A p value <0.05 in meta-analyses was considered statistically significant.

Results are presented as allelic ORs with 95% CIs, representing the estimated effect of each copy of the minor A allele on likelihood of GDM. Multivariable analysis was carried out to adjust for potential confounding (maternal age, BMI and family history of diabetes) and mediating factors (weekly weight change from recruitment to 36 weeks and HbA1c, HDL-cholesterol and triacylglycerol concentrations at recruitment). None of the models adjusted for the HUMBA trial treatments. The predictive value of early pregnancy risk factors (maternal age, BMI and family history of diabetes) for GDM with and without CREBRF rs373863828 genotype status (A/G or A/A) was evaluated by comparing the area under the corresponding receiver operating characteristic (ROC) curves. Specificity, sensitivity, positive and negative predictive values and the negative likelihood ratio were calculated using the DAG stats software package (v2000, https://biostats.com.au/DAG_Stat) [20].

Results

Of 230 women with obesity in the HUMBA trial, 166 (72%) were of Māori or Pacific ethnicity; 112 (67%) completed both OGTT and CREBRF testing and were included in the analysis (Fig. 1). The CREBRF rs373863828 (A/G or A/A) allele was carried by 31 (28%) women: 28/77 (36%) of the women without GDM and 3/35 (9%) of the women with GDM. GDM was diagnosed in 35 (31%) women: 3/27(11%) women with the A/G genotype and 32/81 (40%) women with the G/G genotype (Table 1); none of the four women with the A/A genotype developed GDM. In univariable analyses, women who developed GDM had higher BMI (p = 0.02), waist circumference (p = 0.03), and HbA1c level at recruitment (p < 0.001) and higher fasting glucose concentration (p < 0.001) and 1 h and 2 h glucose concentrations (both p < 0.001) on OGTT (Table 1). There were no statistically significant differences in clinical and biochemical risk factors between women with and without the CREBRF rs373863828 (A) allele; carriers had lower fasting and 2 h glucose concentrations on OGTT (Table 1) but the differences did not reach statistical significance (p = 0.05 and p = 0.07, respectively).

Fig. 1
figure 1

Participant flow in the CREBRF study

Full size image
Table 1 Cohort characteristics
Full size table

The CREBRF rs373863828 (A) allele was associated with a statistically significant decrease in the likelihood of developing GDM in both univariable analysis (OR 0.19 [95% CI 0.05, 0.69], p = 0.01) and after adjustment for potential confounding by maternal age, BMI and family history of diabetes (adjusted OR 0.13 [95% CI 0.03, 0.53], p = 0.004) (Table 2). Adjustment for potential mediators, including GWG and early gestational HbA1c, HDL-cholesterol and triacylglycerol concentrations, did not alter the association between CREBRF rs373863828 (A) allele and GDM (adjusted OR 0.18 [95% CI 0.04, 0.74], p = 0.02).

Table 2 Association between CREBRF rs373863828 (A) allele and risk of GDM in Māori and Pacific women
Full size table

In ROC analysis, the AUC for clinical risk factors (maternal age, BMI and family history of diabetes) was 0.67 (95% CI 0.57, 0.78) (Fig. 2). Addition of rs373863828 status (A/G or A/A) increased the predictive value of the model, giving an AUC of 0.76 (95% CI 0.67, 0.85), an absolute increase of 8.8% (95% CI 1.0, 16.6, p = 0.03) (Fig. 2). Table 3 provides estimates of the specificity, sensitivity and positive and negative predictive value and negative likelihood ratio of rs373863828 status for GDM. Overall, the sensitivity and specificity of rs373863828 status for GDM was 91% and 36%, respectively, with a positive predictive value (G/G) of 40%, a negative predictive value (A/G or A/A) of 90% and a negative likelihood ratio of 0.24.

Fig. 2
figure 2

ROC curves for prediction of GDM. Clinical risk factors include maternal age, BMI and family history of diabetes

Full size image
Table 3 Predictive value of the CREBRF status for GDM
Full size table

Discussion

We have identified a novel protective biomarker, the minor (A) allele of CREBRF rs373863828, that is associated with a fivefold reduction in the likelihood of GDM (defined by IADPSG criteria) in Māori and Pacific women with obesity. This is an important finding, given the high rates of GDM and obesity in these ethnic groups, and the fact that GDM is a major risk factor for subsequent development of type 2 diabetes in women [12].

This decreased rate of GDM in women carrying the CREBRF rs373863828 minor (A) allele appeared to be due to a reduction in both fasting and 2 h OGTT glucose concentrations. Adjustment for potential confounding strengthened this association, suggesting that the CREBRF rs373863828 minor (A) allele is associated with reduced likelihood of GDM independent of maternal age, BMI and family history of diabetes. Adjustment for GWG and HbA1c, HDL-cholesterol and triacylglycerol concentrations in early gestation did not alter the association between the CREBRF rs373863828 (A) minor allele and GDM, suggesting that these are not mediating factors.

The CREBRF gene is found on chromosome 5 and encodes a negative regulatory factor of the cyclic AMP-responsive element-binding protein 3 (CREB3), which in turn is involved in regulation of protein translation. The rs373863828 (A) allele variant was first associated with higher BMI (1.5 kg/m2) and waist circumference (3 cm) but lower odds of type 2 diabetes (OR 0.59) among people of Polynesian ancestry [13, 14, 21]. The minor (A) allele was also weakly associated with increased insulin sensitivity by HOMA-IR in the Sāmoan and American Sāmoan populations [13]. The protective effect of the minor (A) allele for GDM is concordant with the findings of previous association studies among non-pregnant women, which suggested a strong genetic relationship between the CREBRF A allele and lower risk of both GDM and type 2 diabetes.

In this cohort of pregnant women with obesity, when comparing those with and without the minor (A) allele there were no statistically significant differences in BMI, waist circumference or GWG. We speculate that the influence of the CREBRF rs373863828 variant may be associated with increased lean mass rather than any changes in adiposity manifested as GWG. CREBRF rs373863828 variant carriers have been shown to be taller [22, 23] and hence height-related increase in muscle mass and pancreatic beta cell mass may decrease the risk of GDM and type 2 diabetes without influencing GWG.

It remains unclear how the rs373863828 minor (A) allele contributes to higher BMI yet lower risk of type 2 diabetes. Cellular models have suggested that the minor (A) allele promotes lipid storage at a reduced energy cost in the adipocytes [13], although this does not explain the lower risk of type 2 diabetes or GDM. CREBRF may also have a role in modifying fat distribution (subcutaneous vs abdominal fat), which may influence flux of NEFA and disposal of glucose [24]. Interestingly, CREBRF-knockout mice have shown markedly reduced prolactin secretion and augmented glucocorticoid receptor signalling [25]. Given the key role of prolactin in the adaptation of insulin-secreting beta cells during pregnancy [26], and the known diabetogenic effects of glucocorticoids [27], this suggests that CREBRF has a role in adaptive changes in the pancreas and in the regulation of insulin resistance during pregnancy. However, further investigation of beta cell function, insulin sensitivity and glucose disposal levels are needed to fully understand the effect of the CREBRF rs373863828 minor (A) allele on glucose metabolism during pregnancy. Elucidating the underlying molecular mechanisms may provide opportunity for development of novel therapeutic interventions for GDM, particularly as CREBRF has been considered as a target gene in other conditions, such as gastric cancer and glioblastoma [28, 29].

In this population, detection of the CREBRF rs373863828 minor (A) allele had high (90%) negative predictive value for GDM, giving a negative likelihood ratio of 0.24. Although not sufficiently predictive to replace diagnostic testing for GDM [30], the CREBRF rs373863828 genotype could be used in Māori and Pacific women for risk stratification in early pregnancy, and there is increasing evidence that risk prediction models need to incorporate ethnic-specific biomarkers [31]. Māori and Pacific women carrying the minor allele exhibited a low incidence of GDM, despite clinically increased risk due to obesity. Given that 60% of Māori and Pacific women in the Counties Manukau region in New Zealand have obesity and nearly 30% carry the protective minor (A) allele, genotyping may have clinical utility by allowing targeting of resources for women at higher risk of GDM (absence of minor [A] allele). This could include preventative dietary interventions and pharmacotherapy, and earlier testing for GDM by OGTT, rather than by the current two-step GDM screening used in New Zealand (OGTT is reserved for women with an abnormal non-fasting polycose test). These data have relevance not only for Aotearoa/New Zealand but also for the Pacific Islands, which have even higher rates of type 2 diabetes and GDM [32].

Although the magnitude of association that we observed between the CREBRF rs373863828 minor (A) allele and GDM was large and not obviously confounded, the fact that our study was conducted only in women with obesity raises the possibility of selection bias. However, typical collider conditioning is unlikely as this requires both the exposure (CREBRF rs373863828 genotype) and outcome (GDM) to be causally linked to sample selection (obesity), and this does not seem plausible in the case of GDM. Nevertheless, replication of our findings in a broader population is warranted, particularly as we used meta-analysis of self-reported ethnicity to estimate exposure effects rather than genetic population structure estimates.

Overall, the combination of CREBRF rs373863828 genotype and clinical risk factors had greater predictive value for GDM than clinical risk factors alone. We were unable to explore whether there is an interaction between CREBRF rs373863828 genotype and clinical risk factors for prediction of GDM due to the limited sample size of our study. It is possible that clinical risk prediction may perform better in Māori and Pacific women with the rs373863828 GG genotype, allowing further optimisation of risk stratification, but this would require evaluation in larger cohorts, including women without obesity. Identifying ways to prevent development of GDM among Māori and Pacific women is a health priority [33].

In summary, we have shown that the CREBRF rs373863828 minor (A) allele is associated with substantially reduced likelihood of GDM in Māori and Pacific women with obesity. The CREBRF rs373863828 genotype could potentially improve risk stratification for GDM in early pregnancy in Māori and Pacific women.