Association between functional genetic variants in retinoid X receptor-α/γ and the risk of gestational diabetes mellitus in a southern Chinese population

Abstract To clarify the effect of retinoid X receptor-α/γ (RXR-α/γ) genes functional genetic variants (RXR-α rs4842194 G>A, RXR-γ rs100537 A>G and rs2134095 T>C) on the risk of gestational diabetes mellitus (GDM), a case–control study with 573 GDM patients and 740 pregnant women with normal glucose tolerance was performed in Guangxi area of China. An odds ratio (OR) with its corresponding 95% confidence interval (CI) was used to assess the strengths of the association between genetic variation and GDM. After adjustment of age and pre-BMI, the logistic regression analysis showed that the rs2134095 was significantly associated with GDM risk (CC vs. TT/TC: adjusted OR = 0.71, 95% CI = 0.56–0.90) in all subjects, and this result remained highly significant after Bonferroni’s correction for multiple testing (P=0.004). The stratified analysis showed that rs2134095 was significantly associated with the risk of GDM among age > 30 years (adjusted OR = 0.61, 95% CI = 0.39–0.97), BMI > 22 kg/m2 (adjusted OR = 0.46, 95% CI = 0.30–0.70), systolic blood pressure (SBP) > 120 mmHg (adjusted OR = 1.96, 95% CI = 1.14–3.36), glycosylated hemoglobin A1c (HbA1c) < 6.5% (adjusted OR = 1.41, 95% CI = 1.11–1.78), TG ≤ 1.7 mmol/l (adjusted OR = 2.57, 95% CI = 1.45–4.53), TC ≤ 5.18 mmol/l (adjusted OR = 1.58, 95% CI = 1.13–2.22), high-density lipoprotein cholesterol (HDL-c) ≤ 1.5 mmol/l (adjusted OR = 1.70, 95% CI = 1.16–2.49) and low-density lipoprotein cholesterol (LDL-c) > 3.12 mmol/l (adjusted OR = 1.47, 95% CI = 1.08–2.00) subjects, under the recessive genetic model. We also found that rs2134095 interacted with age (Pinteraction=0.039), pre-BMI (Pinteraction=0.040) and TG (Pinteraction=0.025) influencing individual’s genetic susceptibility to GDM. The rs2134095 T>C is significantly associated with the risk of GDM by effect of a single locus and/or complex joint gene–gene and gene–environment interactions. Larger sample-size and different population studies are required to confirm the findings.


Introduction
Gestational diabetes mellitus (GDM) is defined as different degrees of impaired glucose tolerance that is first recognized during pregnancy, often manifested as insufficient islet secretion and insulin resistance, which seriously threatens maternal and child health [1]. Worldwide, it affects approximately 2-20% of all pregnancies [2]. Clinical and epidemiological studies have demonstrated that GDM is associated with maternal and perinatal complications, such as fetal macrosomia, pre-eclampsia, preterm birth, spontaneous abortion, respiratory distress syndrome, small for gestational age (SGA), large for gestational age (LGA), shoulder dystocia, hypoglcemia, polycythemia etc [3][4][5]. As yet, the main risk factors are known as older Abbreviations: F, the forward primer; R, the reverse primer.
was collected for detection of biochemical indicators. Biochemical data consisted of results of OGTT and glycosylated hemoglobin A1c (HbA1c).

Genomic DNA extraction
The genomic DNA was extracted from EDTA-treated whole blood using a DNA extraction kit (Aidlab Biotechnologies Co., Ltd, China) and stored in a refrigerator at −20 • C prior to polymerase chain reaction (PCR).

SNP selection
The potential functional SNPs of candidate RXR-α and RXR-γ genes were screened by using NCBI dbSNP database (http://www.ncbi.nlm.nih.gov/projects/SNP), SNPinfo Web Server (http://snpinfo.niehs.nih.gov/) and SHEsis online tool (http://analysis.bio-x.cn/) [29], and should meet the following criteria: (1) SNPs with minor allele frequency (MAF) reported in HapMap based Han Chinese in Beijing (CHB) was greater than 0.05; (2) located at the regulatory region of studied genes (i.e., the 5 near gene, 5 untranslated regions [UTRs], 3 UTR, 3 near gene or splice sites); (3) might affect the microRNA-mRNA binding sites (MMBSs) activity, transcription factor-binding sites (TFBSs) activity in the putative promoter region or the post-transcriptional splicing process; (4) the linkage disequilibrium (LD) coefficient D or r 2 < 0.8 between candidate SNPs. According to the selection strategy, three SNPs including rs4842194 G>A in the RXR-α gene, and rs100537 A>G and rs2134095 T>C in the RXR-γ gene were identified for further study. Among them, rs4842194 G>A located at the 3 -UTR region of RXR-α gene may affect the MMBS activity. The SNP rs100537 A>G located in the 5 near gene may affect the TFBS function, and rs2134095 T>C in the splice sites may affect the post-transcriptional splicing process of RXR-γ gene.

Genotyping
The PCR and restriction fragment length polymorphism (RFLP) methods were used to analyze the genotypes of the studied loci. The specific primers were designed and synthesized according to the sequence of candidate SNPs by Suzhou Hongxun Biotechnology Co., Ltd, China, in Table 1.

Statistical analysis
The chi-square (χ 2 ) test was used to assess the differences of selected demographic data, environmental exposure factors and genotypes of studied variants between cases and controls. A χ 2 goodness-of-fit test was performed to determine whether the distribution about genotypes of SNPs in control group conformed to Hardy-Weinberg equilibrium (HWE). Unconditional logistic regression analysis was used to estimate the associations between the genotypes of SNPs and the risk of GDM by computing the odds ratios (ORs) and their 95% confidence intervals (CIs). In the present study, the two-sided test was used, and P≤0.05 was considered statistically significant. Bonferroni's correction of the significance level (0.05/3 = 0.017) according to the number of comparisons was applied to account for multiple testing. The statistical analysis was performed with SPSS 17.0 (SPSS, Chicago, IL, U.S.A.). In addition, false positive report probability (FPRP) was estimated to assess the robustness of findings statistically significant association by using the method described by Wacholder et al. [30]. The FPRP threshold was set to 0.2, and the prior probability was set to 0.1 to detect the noteworthiness for OR of 1.5 or 0.67, with an α level equal to the observed P-value.
The multiple dimension reduction (MDR) software (version 3.0.2) was used to investigate the joint effect of SNPs. A 10-fold cross-validation and 1000-fold permutation testing were adopted under the null hypothesis of no association. The cross-validation consistency (CVC) and the testing balanced accuracy was help to identify the best model among all possibilities considered [31,32].

Functional prediction of studied variants
The online bioinformatics tool SNPinfo Web Server (https://snpinfo.niehs.nih.gov/cgi-bin/snpinfo/snpfunc.cgi) and Human Splicing Finder 3.1 [33] were conducted to investigate the putative functional effect of the variants and the gene alternative splicing regulatory.

Characteristics of research objects
In the study, 1313 pregnant women in Guilin were recruited, including 573 patients in the GDM case group, with an average age of 31.48 + − 4.74 years; 740 patients in the normal pregnancy group, with an average age of 28.91 + − 4.18 years. The comparison of general data showed that age, pre-pregnancy BMI, SBP, DBP, 0/60/120 min OGTT and HbA1c in the cases were much higher than that of controls (P<0.05) as shown in Table 2.

Relationship between studied variants and GDM risk
The genotypes of candidate SNPs were detected by a PCR-RFLP method. The results showed that the genotype frequencies of rs4842194 G>A, rs100537 A>G and rs2134095 T>C were in agreement with HWE in the controls (P>0.05), and the P-values were 0.727, 0.668 and 0.418, respectively. The D and r 2 values were calculated to evaluate the LD of studied SNPs from their genotyping data by using the SHEsis online tool, and showed that there was no significant degree of LD detected among three SNPs (data not shown). A multivariable logistic regression analysis adjusted for age and BMI showed that the rs2134095 T>C in RXR-γ gene was significantly associated with the risk of GDM, individuals with CT genotype were 1.32-times higher than those with TT genotype (adjusted OR = 1.32, 95% CI = 1.01-1.73, P=0.043). However, the significance did not exist after Bonferroni's correction. Further, under recessive genetic model, the CC genotype of rs2134095 could significantly reduce the risk of GDM compared with TT/CT genotypes (adjusted OR = 0.71, 95% CI = 0.56-0.90, P=0.004). We failed to find a significant association between the RXR-α rs4842194 G>A and RXR-γ rs100537 A>G and susceptibility to GDM in present study (P>0.05) as shown in Table 3.

FPRP analysis
The FPRP test was adopted to assess the noteworthiness of the observed significant associations between the studied rs2134095 T>C SNP and the risk of GDM, and the prior probability of 0.1 and a relatively stringent FPRP cut-off value of 0.2 were set. The FPRP values were 0.050, 0.031, 0.049, 0.149, 0.176 and 0.178 for the associations of rs2134095 and GDM risk in all subjects, and subgroups of BMI > 22 kg/m 2 , HbA1c ≤ 6.5%, TC ≤ 5.18 mmol/l, HDL-c ≤ 1.5 mmol/l and LDL-c > 3.12 mmol/l, respectively, under the recessive genetic model (CC vs. TT/TC). It suggested that the above associations were noteworthy findings as presented in the present study as shown in Table 5.

High-order interaction with GDM risk by MDR analysis
The results of MDR analysis indicated that rs2134095 was the best one-factor model for GDM with a maximum CVC of 10/10, a testing balanced accuracy of 0.531, and a P-value of statistical test of 0.003. More interestingly, the best interaction model to predict GDM risk for the present study population was a two-factor model of rs2134095 and rs100537 with a maximum CVC of 10/10, a testing balanced accuracy of 0.552, and a P-value <0.001 (Table 6).

Functional prediction of rs2134095
Based on the positive findings identified above, the RXR-γ rs2134095 was further explored of a putative alteration in regulation of gene splicing. It seems that the variant has an impact in splicing regulation according to SNP Function Prediction of SNPinfo Web Server and Human Splicing Finder 3.1 tool. The variant is located in an acceptor splicing

a c a g A G T C C T A A C T G A 1 2
Alteration of an exonic ESE site. Potential alteration of splicing. site, and alteration of the C→T allele of rs2134095 leads to the disruption of an exonic splicing enhancer (ESE) SRp40 ( Figure 1). It is biologically reasonable to hypothesize that the variant has an effect on the splicing of the RXR-γ mRNA, thus altering the function of the gene product.

Discussion
GDM is characterized by β-cell dysfunction, insulin secretory defect and peripheral insulin resistance. It is likely to bring adverse consequences for both mothers and their offspring, for instance, pregnant women with GDM are often at a relatively higher risk of T2DM, obesity, dyslipidemia, preterm birth, spontaneous abortion, respiratory distress syndrome of newborn etc. Studies showed that the RXR-α/γ proteins perform the function of transcription factors by binding as heterodimers to the specific sequences in the promoter of target genes of VD and retinoid pathways, which are closely related to diabetes, islet function, and glucose and lipid metabolism. We believe that it is necessary to explore the relationship between RXR-α/γ and GDM. GWASs and meta-analysis have shown that association genetic loci could provide insight into pathogenetic mechanisms underlying the GDM [20,[34][35][36]. To date, there have been studies on multiple genes and their SNPs of VD pathway with the susceptibility to T2DM and GDM [26,[37][38][39].
Studies have shown that the variants in RXR-α/γ genes might be associated with the risk of T2DM and metabolic syndrome [24][25][26]. We genotyped three potential functional SNPs in RXR-α/γ genes explore the pathogenesis of GDM.
In the single locus analysis, we found that the RXR-γ rs2134095 T>C was significantly associated with the risk of GDM under the recessive genetic model in a Southern Chinese population. In the stratified analysis, the reduced risk of GDM associated with rs2134095 SNP was more pronounced among the elderly (>30 years old) and relatively higher pre-BMI (>22 kg/m 2 ) subjects carrying the CC genotype in the recessive genetic model. Further, the interaction between rs2134095 and age and pre-BMI was detected. The above findings suggest that not only a single environmental or genetic risk factor but also their interaction might affect the individual suffering from GDM under different etiologies. Additionally, recent studies have confirmed that the variants located in RXR-γ were significantly associated with the lipid metabolism of chronic metabolic diseases. These variants are closely related to abnormal lipid metabolism by affecting the levels of TG, LDL-c, apolipoprotein B etc [40][41][42]. In the hierarchical analysis of this study, higher risk effects in HbA1c (≤6.5%), TG (≤1.7 mmol/l), TC (≤5.18 mmol/l) and LDL-c (>3.12 mmol/l) are observed in rs2134095CC carrier subjects. These findings might be helpful to clarify the genetic component of RXR-γ and lipid profile.
Very interesting, results from the MDR analysis also consistently identified RXR-γ rs2134095 T>C as the main single susceptibility locus contributed to the risk of GDM, and some complex gene-gene interaction effects exist among the RXR-γ SNPs. It showed that a two-factor model including the combination of rs2134095 and rs100537, is the best model to predict GDM risk in the study population. These results suggest that RXR-γ gene variants may have a joint effect on the genesis of GDM.
These above findings seem to be biologically plausible. Study confirmed that gene variants might change the splicing pattern of pre-mRNA leading to a wrong transcript product and ultimately influence the gene expression or function [43][44][45]. According to the online prediction database 'snpinfo web server' , the SNP rs2134095 may play the role of exonic splicing enhancer (ESE) or silencer (ESS) performing a function of RXR-γ gene post transcriptional splicing. Grave et al. used a general linear model to evaluate the VD pathway gene-gene interaction and detected a significant effect of the interaction between RXR-γ rs2134095 and GC rs7041 on low-density lipoprotein cholesterol (LDL-c) levels. The further functional analysis indicated that the studied variants were involved in the regulation of the gene expression [42]. Another study of Sentinelli et al. found that the 'at-risk' haplotype of RXR-γ gene variants (rs1128977, rs2651860, rs2134095, rs283696 and rs10918169) were associated with the risk of familial combined hyperlipidemia (FCHL) and a higher LDL-c level [40]. It is speculated that the rs2134095 T>C may affect the function of RXR-γ and alter the activation of retinoid and VD pathway genes, which are closely related to diabetes, islet function, and glucose and lipid metabolism. In addition, maternal plasma levels of vitamin A and vitamin D reduce during pregnancy, thus exacerbating the functional impairment of above pathways caused by rs2134095. The above is likely to explain the mechanism of the changes in the susceptibility of pregnant individual to GDM.
The FPRP analysis is an effective method to verify the noteworthiness of significant associations [46]. In the present study, a relatively stringent FPRP threshold of 0.2 was set. The FPRP values of the observed significant associations between rs2134095 T>C and the risk GDM is much lower than the preset threshold 0.2 in BMI > 22 kg/m 2 , HbA1c ≤ 6.5%, TC ≤ 5.18 mmol/l, HDL-c ≤ 1.5 mmol/l and LDL-c > 3.12 mmol/l subgroups under the recessive genetic model. It suggests that the positive findings are probability authentic and reliable. Meanwhile, the detected FPRP values in some comparisons much greater than 0.2 indicated that these significant findings might be chance observations. Therefore, the conclusion obtained here must be viewed as preliminary and needs to be verified.
To our knowledge, it was the first time to explore the relationship between the studied functional variants of RXR-α/γ genes and the risk of GDM, and some etiological clues of GDM are suggested from the perspective of genetics. The strengths of our study include its good design, relatively large sample size and multiple analysis methods. However, some limitations in this study should not be neglected. First, designed as a hospital-based case-control study, it may be still subject to inherent biases, like selection bias and recall bias. Second, although relatively large samples were recruited, however, the lower minor allele frequency of some tested SNPs might still limit the statistical power to detect significant association in some subgroups. Third, some positive correlations between rs2134095 T>C and GDM risk were observed, however, the specific mechanisms underlying the effect of the above-mentioned positive findings should be investigated.

Conclusion
In summary, the present study support that the RXR-γ gene rs2134095 T>C variant is significant associated with the increased risk of GDM by effect of a single locus and/or complex joint gene-gene and gene-environment interactions. Furthermore, larger sample-size and different population studies are required to confirm our findings.