2.1 Study Recruitment and Sample Collection
Non-smoking pregnant women (n = 138) with uncomplicated pregnancies undergoing elective terminations during mid-gestation of pregnancy (gestational week [GW] 15-24) were recruited at the Women’s Options Center (WOC) in the San Francisco Bay Area from 2014–16. The WOC serves a racially/ethnically diverse low-income population of pregnant women in Northern and Central California who tend to rely on public health insurance coverage for prenatal care. Written and verbal consents were obtained during the clinical visit, followed by administration of a survey questionnaire to each study participant. All study protocols were approved by the UCSF Institutional Review Board prior to the clinical visit.
We collected the following biological samples—maternal serum, placenta, and fetal liver (n = 141, due to three sets of twins)—from study participants during or immediately following the clinical procedure. Chemical PBDE analyses were performed on n = 135 placental samples, with n = 130 matched samples (excluding twins) of all three biomatrices (Figure 2). For the chemical analyses, maternal serum samples were centrifuged at 3000 RPM for 10 min at 4°C prior to aliquoting and transfer of the serum with glass pipettes into pre-screened (to confirm absence of PBDE) sterilized amber vials (storage at −80°C). A subset of placental samples (n = 62) was further processed and evaluated for three molecular and morphological features (see below). The placental subset was selected based on the highest and lowest PBDE exposure levels to maximize our ability to detect differences, which we expected would reduce selection bias and enhance generalizability across the whole population. Population characteristics between the subset assessed for placental biomarkers (n = 62) and the whole population (n = 130) were similar. Placental biopsies contained a section of the basal plate (which represents the maternal-fetal interface) and includes placental regions covered in our analysis (chorionic villi and decidua).
Placental tissues were submerged in medium (DME/H-21 [Gibco], 12.5% fetal bovine serum [Hyclone], 1% glutamine plus [Atlanta Biologicals], 1% penicillin/streptomycin [Invitrogen], and 0.1% gentamicin [Gibco]); dissected into 1 x cm3 pieces; fixed with 3% paraformaldehyde (PFA); and frozen in Optimal Cutting Temperature (OCT) medium (Sakura Finetek, SA62550-01) in a cryomold at -80⁰C [42]. Placental biopsies were later sectioned (5µM thickness) using a cryostat (Leica) and placed on glass slides for molecular and morphological assessments.
2.2 Chemical Exposure Assessment
Nineteen PBDE congeners (including BDE-17, -28, -47, -66, -85, -99, -100, -153, -154, -183, -196, -197, -201, -202, -203, -206, -207, -208, and -209) were analyzed in maternal serum, placenta, and fetal liver by the Environmental Chemistry Laboratory at California’s Department of Toxic Substances Control in Berkeley, CA, using gas chromatography–high-resolution mass spectrometry (GC–HRMS, DFS, ThermoFisher, Bremen, Germany). Isotopically-labeled internal surrogate mix standards (IS) were used for quantitation (Wellington Laboratories, Inc., Guelph, Ontario, Canada).
Thawed serum samples (1 mL) were spiked with carbon-labeled internal standards (13C12-BDE-28, 47, 99, 153, 154, 183, 197, 207, and 209), with 4 mL each of formic acid and water added. Serum samples were vortexed and loaded into an automated sample extraction system (RapidTrace, Biotage; Uppsala Sweden). Oasis HLB cartridges (3 cc, 500 mg, Waters Corp.; Milford, MA) were used for sample extraction and acidified silica (500 °C prebaked, manually packed, 3 cc) for sample extraction cleanup. Final eluates were concentrated 10-fold using an automated nitrogen evaporation system (TurboVap LV, Biotage; Uppsala, Sweden) and spiked with a 13C12-PCB-209 recovery standard [43,44]. Total cholesterol and triglycerides were measured enzymatically by Boston Children’s Hospital (Boston, MA) and subsequently used to calculate the total serum lipid level for each study participant using the Phillips formula [45]. Fetal liver and placental samples were analyzed using our liver analytical method with slight modification [25]. Before sample extraction, only placenta samples were lyophilized. Briefly, samples were homogenized and spiked with the same internal standards (listed above). Samples were then denatured with hydrochloric acid and extracted with 1:1 hexane:methyl tert-butyl ether (MTBE). Aqueous potassium chloride solution was added to each sample extract to remove potentially co-extracted aqueous compounds and the organic layer was re-extracted and dried in a pre-weighed, pre-baked aluminum weighing dish for lipid content determination via gravimetric analysis. Samples were then reconstituted in hexane and lipids were removed using concentrated sulfuric acid, followed by cleanup with acidified silica (500 °C prebaked, manually packed, 3 cc) on automated SPE system (RapidTrace, Biotage; Uppsala Sweden).
2.3 Assessment of Stage-Specific Antigens and Morphological Endpoints
Molecular and morphological placental features were evaluated independently by two investigators who were blinded to PBDE exposure levels. We assessed molecular immunoreactivity of CTBs at the maternal-fetal interface with antibodies specific for integrin alpha 1 (ITGA1), VE-cadherin (CDH5) or metalloproteinase 1 (MMP1) in five stages of trophoblast differentiation: Zone I) CTBs resident in FV; Zone II) CTBs in the proximal (p) regions of AV cell columns; Zone III) CTBs within the distal (d) regions of AV cell columns; Zone IV) interstitial invasive/extravillous CTBs (iCTB); and Zone V) endovascular CTBs (eCTB) (Figure 1). We selected ITGA1 and CDH5 based on previous in vitro and in vivo literature demonstrating the importance of these adhesion receptor molecules in placental development and function (specifically regarding invasive and endovascular CTB differentiation pathways that are critical for vascular remodeling) as well as their potential disease associations. MMP1 is a metalloproteinase matrix-degrading enzyme that facilitates CTB migration/invasion during placentation [7,9,46] (Table S1 and Figures S3–S5). We immunolocalized ITGA1, CDH5, and MMP1 in each placental zone using published methods [47]. The antibodies and dilutions that were used for this purpose are described in (Table S2). CTBs were identified by reactivity with CK7. Batch effects which influence immunofluorescence (IF) intensity were continuously monitored by including serial sections from the same two placental samples in each assessment. Slides were imaged using an upright Leica DFC450 microscope equipped with a camera and Leica Advanced Fluorescence Application Suite Ver. 3.2 (Leica Microsystems). Immunoreactivity was scored using a previously established semi-quantitative approach in which two independent reviewers categorized immunofluorescence into three broad pre-defined categories (rather than continuous measures) based on the percentage of CTBs that reacted with each antibody: 1) < 25% (-); 2) 25–75% (-/+); and 3) > 75% (+) [47].
Placental biopsies were also stained with Hematoxylin and Eosin (H&E) to evaluate morphological features: (1) average total number of white blood cells (WBC) present in basal plate (BP) per 10X field; (2) percent of FV with perivillous fibrinoid deposits (~ 60 villi evaluated per field); 3) fibrinoid deposition at the utero-placental junction; and (4) total number of modulated uterine spiral arteries which contained CTBs (defined as presence of > 50% endovascular CTBs in the lining of uterine spiral arteries) (Table S1 and Figure S5). All values were based on examination of the entire placental section at 10X resolution (~6-10 independent fields). In total, we evaluated 62 placentas; however, the sample size for each molecular and morphological endpoint varied depending on whether tissue sections captured the key structures of interest, most notably uterine arteries. For example, eCTB were confined to placental tissue sections that contained uterine arteries (n = 29–42); pAV and dAV CTBs were observed in 51–53 samples, and iCTB were found in 60 samples (Table S3). For morphological assessments, one sample was lost while processing (n = 61 placentas evaluated). Images were acquired using a bright field Leica DFC450 microscope equipped with a camera (Figure S5).
2.4 Statistical Analysis
We calculated detection frequencies for all 19 PBDE congeners in matched samples of maternal serum, fetal liver, and placental tissues obtained in this study (2014–16; n = 130) (Table S4). We normalized wet-weight PBDE concentrations to total lipid levels in order to account for possible measurement error related to estimating PBDE exposure from biomonitoring data, since biological proxies of exposure to PBDEs and other lipophilic compounds can vary systematically with population characteristics (e.g., lipid content) rather than the PBDE exposures they represent [48,49]. For five PBDE congeners (BDE-28, -47, -99, -100, and -153) detected in > 50% of placental samples (n = 135), we further calculated wet-weight and lipid-adjusted descriptive statistics, including geometric mean, interquartile range (25th–75th percentile), and range (min–max). We also computed a summary PBDE metric by adding four of congeners (ΣPBDE4 = BDE-47, -99, -100, and -153) with detection frequencies > 50% in all maternal-fetal tissues, in order to facilitate results comparison with maternal serum and fetal liver in secondary analyses. We used maximum likelihood estimation (MLE) assuming a log-normal distribution to account for PBDE levels below the laboratory method detection limit (MDL) [50]. Statistical analyses were performed using R (Version 3.5.1) [51], with significance defined as p < 0.05 and marginal significance as p < 0.10 (two-sided tests).
We examined how PBDE levels varied across categorical population characteristics, including maternal age (< 20, 20–24, 25–29, ≥ 30 years), gestational age (< 19, 19–21, ≥ 21 weeks), body mass index (BMI; < 25, 25–30, ≥ 30 kg/m2), parity (0 or ≥ 1 live births), fetal sex (male or female), education (≤ High School or ≥ Some college), type of insurance (public or private/self-pay), race/ethnicity (Latina/Hispanic, Non-Hispanic Black, Non-Hispanic White, Asian/Pacific Islander), birth country (U.S. or foreign born), and sample collection year (2014, 2015, 2016). Although sample size precluded multivariable analysis, we examined population characteristics in separate models with our exposure (PBDE levels) and outcome of interest (placental biomarkers). For example, although fetal sex could conceivably be an effect modifier of PBDE effects on placental development and function, we did not observe variation with exposures or outcomes in this study.
In addition, we computed descriptive statistics for molecular and morphological placental biomarker data. We natural log-transformed average WBC counts prior to statistical analysis based on visual inspection of right-skewed distribution as well as results from the Shapiro-Wilk statistical test. One observation had a WBC count of zero which we set to 0.01 prior to log transformation. We calculated the percent of modulated uterine spiral arteries by dividing the number of uterine spiral arteries with > 50% endovascular CTB modulation by the total number of uterine spiral arteries observed. If the total number of uterine spiral arteries (i.e., the denominator) was zero or less than two, the denominator and percent modulation were re-coded as missing. Then we divided % CTB-modulated uterine blood vessels (mBV) at the median and categorized the variable into two groups (low and high). We modeled molecular immunoreactivity categorically rather than continuously based on the three pre-defined categories that were scored by two independent reviewers in the laboratory (as described in section 2.3): (1) < 25% (-); 2) 25–75% (-/+); and 3) > 75% (+).
Kendall’s Tau Correlation Coefficient [52], a non-parametric measure of correlation that compares concordant and discordant pairs of ordered data to determine the extent of increasing or decreasing covariation between two variables, was used to evaluate monotonic relationships: 1) among molecular and morphological placental biomarkers; and 2) between placental biomarkers (molecular and morphological) and placental PBDE levels (wet-weight and lipid-normalized). We calculated the correlation between placental biomarkers and PBDE levels using the censored version of Kendall’s tau [53]. We adjusted for multiple comparisons by estimating the false discovery rate (FDR) using the Benjamini and Hochberg method [54].
To compare PBDE levels across immunoreactivity groups, we tested group mean differences (assuming a log-normal PBDE distribution) using MLE to account for left-censored PBDE data [50]. From these bivariate censored regression/ANOVA models, we calculated the pairwise percent (%) differences in PBDE levels as (eβ −1) ∗ 100 and the 95% confidence interval (95% CI) as (eβ ± 1.96 × SE −1) ∗ 100, where the referent group comprised placental samples in the < 25% (-) immunoreactivity group. The beta coefficient from these models represents the difference in PBDE levels on the lognormal scale between moderate or high immunoreactivity groups and the referent group, while eβ represents the ratio of average PBDE levels between immunoreactivity groups. We also used the Fisher’s exact test to examine whether molecular immunoreactivity varied by high/low PBDE exposure (where PBDE levels were divided at the median and categorized into high vs. low exposure groups).
In addition to assessing correlation between placental PBDE levels and placental biomarkers, we also re-examined placental biomarker associations with fetal liver and maternal serum PBDE levels, since we previously found moderate correlation between congeners across maternal-fetal tissues during mid- gestation, with cross-tissue correlations ranging from 0.18 (BDE-99) to ≥ 0.50 (BDE-47, -100, and -153) (p < 0.0001) [55].