Extending non-targeted exposure discovery of environmental chemical exposures during pregnancy and their association with pregnancy complications—a cross-sectional study

2.1 Data and Sample Collection

We enrolled pregnant participants in the CIOB2 cohort between 2014 and 2018 from the prenatal clinics at the Zuckerberg San Francisco General Hospital (SFGH) and Mission Bay or Moffit Long (MB/ML) hospitals. Participant recruitment and sample collection is described elsewhere.^4,5,14 Briefly, participants were recruited during the second trimester of pregnancy, with eligibility criteria including 18-40 years of age, English or Spanish speaking, with singleton pregnancies, and no diagnosed pregnancy complications at recruitment. Paired maternal and umbilical cord blood were collected from participants who agreed to bank their samples and be contacted for participation in future studies. Maternal samples were collected at delivery and cord blood was collected after delivery but before clamping whenever possible. Blood was collected in BD Vacutainer Plus serum tubes, spun at 3000 rpm to collect the serum, and stored at -80°C until analysis. Study protocols were approved by the Institutional Review Boards at the University of California, San Francisco and Berkeley (# 13-12160).

Demographic information was collected via interviews conducted during the second trimester of pregnancy and included maternal age, race and ethnicity (white, Latinx, and Non-Hispanic Asian/Pacific Islander, other race—due to low numbers of Black, Native American, and multiracial participants, these groups were collapsed into “other” for model specification), maternal education (some college or less, completed a college degree, >college degree), household income (<40k per year, 40 to 100k, >100k), nativity (born in the U.S. yes or no). Pre-pregnancy body mass index (BMI kg/m2; <18.5, 18.5 to <25, 25 to <35, >=35)¹⁵ and hospital of delivery [Zuckerberg San Francisco General Hospital (SFGH) or Mission Bay/ Moffit Long (MB/ML)] were abstracted from the medical record.

Pregnancy complications were identified by clinical diagnoses and abstracted from the medical record. These included the binary outcomes of: gestational diabetes mellitus (GDM), preeclampsia, and gestational hypertension. Preeclampsia and pregnancy related hypertension were combined into one outcome called hypertensive disorder of pregnancy because they make up part of a larger spectrum of pregnancy complications related to hypertension.^16,17

2.2 Chemical selection

Nine environmental chemicals were measured in maternal and cord samples using targeted approaches described below. Linear and branched isomers of perfluorooctane sulfonate (PFOS); perfluorohexane sulfonate (PFHxS); monoethylhexyl phthalate (MEHP), a metabolite of diethylhexyl phthalate, a common plasticizer¹⁸; 4-nitrophenol, a known endocrine disruptor used in pesticides, dyes, and pharmaceuticals¹⁹; tetraethylene glycol, a plasticizer and solvent used in metal lubricants, printing inks, and cosmetics;²⁰ tridecanedioic acid and octadecanedioic acid which are endogenous fatty acids that also have exogenous sources primarily from plastics synthesis;^21–23 and deoxycholic acid, a bile acid and endogenous biomarker associated with GDM (Table 1).²⁴

The selection and confirmation procedures for selection of these chemicals is described elsewhere.^5,6,13 Briefly, in previous NTA studies, 30 matched maternal and cord samples were scanned using HRMS in positive and negative ionization modes identifying suspect features. Detected features were matched to an in-house database and features were then prioritized for MS/MS fragmentation based on the ubiquity of exposure, feature intensity differences across demographic variables, and correlation between maternal and cord samples. Fifteen chemicals were confirmed at level 1 confidence via comparison to analytical standards.²⁸ Level 1 confirmed features included monoethylhexyl phthalate, 4-nitrophenol, tridecanedioic acid, and octadecanedioic acid. Subsequent generalized suspect screening analysis of this cohort was expanded to include 295 matched maternal and cord samples and a more comprehensive database including both endogenous and exogenous compounds. This resulted in nineteen chemicals confirmed with analytical standards to level 1 confidence, including those previously identified, and adding PFHxS, PFOS, and deoxycholic acid which were tentatively identified through MS/MS fragmentation (confidence level 2). We selected chemicals for targeted quantification if the chemical was identified in previous NTA with level 1 or level 2 confidence, is a widely used industrial compound with high potential for exposure, is associated with pregnancy complications, or is not regularly biomonitored by the Center for Disease Control or other large biomonitoring studies.⁵

2.3 Targeted chemical analysis

Quantitation analysis of the nine selected chemicals was carried out using Agilent MassHunter Quantitative Analysis software package (Agilent Technologies Inc., Version 10.0). Analytical grade standards were purchased from suppliers (Sigma Aldrich and Thermo Fisher) and serial dilutions of the standards were prepared in methanol. Calibration curve of each of the specific standard was acquired under the same condition as serum sample analysis, using the LC/QTOF-MS (Agilent 1290 UPLC interfaced with Agilent iFunnel 6550 QTOF-MS system) as described previously.^5,6,13 M2-PFOA (Wellington Laboratories, Ontario, Canada) was used as internal standard for negative ionization chemicals, while D15-TPP (Cambridge Isotope Laboratories, Tewksbury, MA) for positive ionization chemicals. Particularly, PFOS in serum samples were separated chromatically, thus both total and linear PFOS quantitation were calculated, and branched PFOS concentrations were obtained by subtracting linear portion from the total PFOS.

2.4 Statistical analysis

To address chemical levels below the limit of detection we used machine reported values when available or, if not available, we imputed levels below the LOD with LOD/sqrt(2). We assessed the detection frequency (percent above the limit of detection) and limited our statistical analysis to chemicals with a detection frequency of 67% or higher. We described the distribution of chemicals in maternal and cord blood samples, and calculated Spearman correlation coefficients for chemicals within and between maternal and cord samples. We natural log-transformed the data for use in statistical models and applied logistic regression models to assess the relationship between maternal chemical exposure and pregnancy complications. Potential confounders were identified a priori (Supplemental Figure 1), and our final models were adjusted for maternal age, hospital of delivery, and race/ethnicity. We considered hospital of delivery as a cumulative indicator of socioeconomic status (including income, educational attainment, and insurance status). Patients in both hospitals are racially and ethnically diverse. However, SFGH is a safety net hospital that serves a large proportion of patients without private health insurance, while MB/ML serves patients of higher socioeconomic status and with private health insurance. Race and ethnicity are included in models as an indicator of the exposure to structural racism, that is independently associated with chemical exposure and adverse pregnancy complications. We calculated the interquartile range (IQR) of each natural log transformed chemical. Beta coefficients and 95% confidence intervals (CI) were multiplied IQR and exponentiated to identify the odds ratio (OR) of each pregnancy complication for each IQR increase in exposure level.

Participants, demographics, and medical record data

The analysis included 302 participants (Table 2). The average age at delivery was 33 years of age. Forty-three percent of participants identified as white and 32% as Latinx or Hispanic. Most participants completed college or graduate degrees (>65%), were born in the U.S. (54%), and had a household income of over $100,000 per year (56%). For the outcomes, 21% of participants had a diagnosis of GDM, 13% gestational hypertension, and 7% preeclampsia.

3.1 Distribution and correlation of chemical levels in maternal samples

Participants had at least four chemicals in maternal samples and three chemicals in cord samples. PFHxS, linear PFOS, deoxycholic acid, and octadecanedioic acid were all detected in at least 97% of maternal samples. Deoxycholic acid, tridecanedioic acid, PFHxS, were detected in at least 87% of cord blood samples (Table 3). Since we estimated branched PFOS levels by subtracting linear PFOS from the total PFOS we did not calculate a detection frequency for this chemical. Chemical detection frequency varied between maternal and cord blood samples. Linear PFOS, and 4-nitrophenol were detected more often maternal samples while monoethylhexyl phthalate, and tridecanedioic acid were detected more frequently in cord blood. Similarly, the average chemical levels varied across the different matrices with branched PFOS, linear PFOS, deoxycholic acid, and octadecanedioic acid found at higher levels in maternal samples compared to cord samples, while tridecanedioic acid was higher in cord than in maternal samples.

Monoethylhexyl phthalate, tetraethylene glycol, and 4-nitrophenol were each detected in fewer than 50% of maternal samples, however 4-nitrophenol was detected in more maternal serum samples than cord samples, while monoethylhexyl phthalate was found in more cord blood samples. Due to their low detection frequency, monoethylhexyl phthalate, tetraethylene glycol, and 4-nitrophenol, were excluded from further analyses.

When assessing the Spearman correlation coefficients (Figure 2) of the same chemical across maternal and cord samples, we found that linear PFOS, and branched PFOS and PFHxS were highly correlated in the maternal and cord samples (correlation coefficients of 0.8, 0.8, and 0.9 respectively). When looking at the correlations of different chemicals in the maternal and cord samples we found moderate correlations between cord-linear PFOS and maternal-branched PFOS, as well as between cord-linear PFOS and maternal-PFHxS (coefficients 0.7 and 0.5 respectively). In the cord blood samples, linear PFOS was highly correlated with branched PFOS and moderately with PFHxS (coefficients of 0.8 to 0.6 respectively), and octadecanedioic acid was moderately correlated with tridecanedioic acid (coefficient of 0.4). In the maternal samples, the relationship between linear and branched PFOS had a correlation coefficient of 0.8, branched PFOS and PFHxS a correlation of 0.6, and between linear PFOS and PFHxS of 0.6. The PFAS chemicals were weakly correlated with the fatty acids—octadecanedioic acid, tridecanedioic acid—and neither PFAS, nor the fatty acids were correlated with deoxycholic acid, within or between the maternal serum and cord blood samples.

• Download figure
• Open in new tab

Figure 1.

Summary of NTA procedures leading to the selection of nine chemical for quantification.

• Download figure
• Open in new tab

Figure 2:

Spearman correlation coefficients for chemicals detected in maternal and cord samples, chemical have at least 65% DF in maternal and cord samples (n = 302).

3.2 Association between chemical exposure with pregnancy complications

From our logistic models we observed a trend of increased odds of GDM for each IQR increase in levels of exposure to environmental chemicals and increased odds of pregnancy hypertensive disorders with tridecanedioic acid (Table 4). For the relationship between chemical exposures and GDM, all the chemicals included in this analysis had an OR above 1. Stronger associations were observed for branched PFOS, linear PFOS, and octadecanedioic acid; The unadjusted relationship between linear PFOS and branched PFOS with GDM was statistically significant. The direction and measures of association we observed between chemical levels and hypertensive disorders of pregnancy was less consistent than for GDM. Exposure to tridecanedioic acid was associated with an increase in the odds of hypertensive disorders of pregnancy [OR (95%CI): 1.28(0.90, 1.86)]. Exposure to tridecanedioic acid was also associated in increased odds of the individual outcomes that made up hypertensive disorders of pregnancy—preeclampsia and gestational hypertension (Supplemental Table 1). The ORs for PFHxS, octadecanedioic acid, and deoxycholic acid with gestational hypertension approximated 1. Linear and branched PFOS had an OR of <1 but the confidence intervals included the null value.