Study population
The Guangxi Zhuang Birth Cohort (GZBC) is a prospective and ongoing birth cohort conducted in county-level hospitals of six major counties in Guangxi Province of China from June 2015. The baseline design, inclusion, and exclusion criteria of this cohort had been reported (Liang et al. 2020). The present study is a subset of the GZBC. In this study, women with a history of diabetes mellitus and insufficient serum sample size were excluded. A total of 100 pregnant women diagnosed with GDM were selected. For each subject with GDM, four healthy women without GDM were selected as pair-matched controls (maternal age and fetal sex). All subjects had detailed questionnaire data and medical records. All women gave their informed consent, and the study was approved by the Ethical Committee of Guangxi Medical University (No.20140305-001). After the informed consent form was signed, a standardized structured questionnaire was administered to each participant through an in-person interview to collect information.
GDM diagnosis
Oral glucose tolerance test (OGTT; 75 g) was used to screen for GDM at 24–28 weeks of gestation. GDM was diagnosed according to the Diagnostic Criteria for Gestational Diabetes Mellitus (WS311-2011) released by the Ministry of Health of China, which coincided with the recommendations by the International Association of Diabetes and Pregnancy Study Groups. Pregnant women who met one or more of the following criteria were diagnosed with GDM: (1) fasting blood glucose (FBG) was more than 5.1 mmol/L; (2) 1-hour blood glucose > 10.0 mmol/L; or (3) 2-hour blood glucose > 8.5 mmol/L. The subjects were tested in the morning after an overnight fast of at least 8 h. The first blood sample was taken to measure the FBG before 9 a.m.
Exposure assessment
Spot serum samples collected during the study visits were stored in polypropylene containers at −80 °C until further analysis. Serum bisphenol concentrations were quantified using an ultra-high liquid performance chromatography-tandem mass spectrometer (UPLC-MS, Waters, USA) with isotope-labeled internal standards as previously described (Liang et al. 2020). Briefly, 500 µL of the serum were spiked with three isotope-labeled internal standards (i.e., BPA-D16, BPS-13C12, and TBBPA-13C12), sodium dihydrogen phosphate dihydrate buffer (pH 5.4), and β-glucuronidase/sulfatase. The mixtures were hydrolyzed and incubated overnight in the dark at 37 °C. After the enzymatic hydrolysis, the solution was extracted twice with 2 mL of the solvent [n-hexane: acetone (7:3, v/v)] and 2 mL of methyl-tert-butyl ether. The supernatants were combined, evaporated until dry at 40 °C, redissolved in 100 µL of methanol: 0.1% ammonia solution (50:50, v/v), and filtered for instrumental analysis. The chromatographic separation was achieved by using an Acquity UPLC BEH C18 (1.7 mm, 2.1 × 100 mm, Waters, USA) analytical column with a mobile phase gradient of ammonia solution and acetonitrile. The bisphenols were detected by using negative-ion electrospray ionization mass spectrometry and multiple reaction monitoring mode. Procedure blank, solvent blank, and calibration standard samples were measured in each batch sample. As previously reported, the limits of detection (LODs) of BPA, BPB, BPF, BPS, and TBBPA were 0.193, 0.232, 0.507, 0.046, and 0.454 ng/mL, respectively (Liang et al. 2020). The determined concentration below the LOD was calculated as the LOD divided by the square root of 2. The resulting quotients were then used in the subsequent data analyses.
Covariates
For each participant, a face-to-face interview was conducted at the hospital by professionally trained interviewers using a standardized and structured questionnaire to collect information, including demography (e.g., age, ethnicity, employment status, and self-reported pre-pregnancy weight) and lifestyles (e.g., drinking and smoking before pregnancy and passive smoking during pregnancy). Moreover, maternal information (e.g., height, parity, gravidity, and pregnancy complications) and birth characteristics (e.g., sex, gestational age, and anthropometric measures) were obtained from the medical data.
Statistical analysis
Summary statistics were calculated and reported as the means ± standard deviation for continuous variables and n (%) for categorical variables in both the GDM and non-GDM groups. Continuous variables were compared using Mann-Whitney U tests, while categorical variables were compared using χ2 tests. A natural log transformation [ln (X)] was applied to the bisphenols to normalize their distributions. This step stabilized the variances for parametric model assumptions and reduced the influence of extreme values. The distributions of bisphenol concentrations are presented as percentiles and geometric mean. Spearman’s rank correlation analysis was used to explore correlations among the serum bisphenol concentrations in the study population.
We categorized participants into tertiles based on the distribution of serum BPA, BPB, BPS, and TBBPA concentrations among the whole population. Given that a high proportion of the samples was below the LOD, BPF with concentrations < LOD were classified into the low-exposure group (Referent), while those with detected concentrations were divided into the middle- (LOD-median) and high-exposure groups (≥ median) based on the median of the detected concentration levels. Conditional logistic regression analysis was employed to evaluate the risk of GDM by a serum bisphenol level, and individual serum bisphenol concentrations were modeled as continuous and categorical variables. Covariates were selected based on either their biologic plausibility (regardless of statistical significance) or the association with GDM in the bivariate analysis (P < 0.10). The crude model was a basic unadjusted model. In the single-bisphenol models, covariates included maternal age, pre-pregnancy body mass index (BMI), area of residence, passive smoking during pregnancy, gravidity, parity, regular exercise, and infant sex. The multi-bisphenol model was aimed to explore the co-exposure effects of bisphenols by considering other bisphenols in one model. The linear P-values were derived by modeling the median value of each category of bisphenol as a continuous variable in the statistical model.
Given the potential effect of pre-pregnancy BMI and difference in fetal sex for the risk of GDM, stratified analyses were performed using pre-pregnancy BMI (18.5–22.9, ≥23.0 kg/m2) and infant sex to evaluate the potential effect modification on the association between bisphenols and the risk of GDM. The BMI cut-off point of 23.0 kg/m2 was selected for overweight subjects, according to a reported optimal cut-off value of BMI for urban Chinese female adults (Zeng et al. 2014).
Considering the possible nonlinearity and nonadditive effects among mixed bisphenols, BKMR (Bobb et al. 2018, Bobb et al. 2015) was used to assess the joint effect of all bisphenols on the risk of GDM, the impact of an individual bisphenol as part of a bisphenol mixture, the nonlinear dose–response effect of these bisphenols and the risk of GDM, and the possible interaction among different bisphenols. Predictors with a posterior inclusion probability (PIP) greater than or equal to 0.5 were considered meaningful (Lebeaux et al. 2020). Quantile-based g-computation (Keil et al. 2020) was used to corroborate the overall effect to compare with results from the BKMR models. Such computation is an adaptation of a mixture modeling method used in environmental epidemiology known as weighted quantile sum regression (Carrico et al. 2015) (WQSR). Compared with the traditional WQSR, quantile g-computation does not require the directional homogeneity of effect estimates. The covariates adjusted in the BKMR and quantile g-computation were the same as those in the single-bisphenol conditional logistic regression analysis. Serum bisphenol levels were ln-transformed in the BKMR and quantile g-computation.
All data were analyzed using R (version3.6.1; R Foundation for Statistical Computing), and a two-sided P < 0.05 was considered statistically significant.