Discussion
Within our nested case–control study of mothers giving birth in the Boston area, we found no significant associations between averaged or cross-sectional urinary BPA levels and PTB when treating PTB as a single, homogeneous outcome. After further classifying PTB by either spontaneous or placental etiology, no significant relationships with BPA exposure developed, except for significantly elevated odds of spontaneous preterm birth in association with levels measured at visit 4 [median, 35.1 weeks (range, 33.1–38.3 weeks)]. However, these associations should be interpreted cautiously because of the low number of cases in each subgroup. After additional stratification on infant sex, we observed significantly elevated odds of overall PTB for female but not male infants in association with averaged BPA exposure. As far as we are aware, this is the first study to use a more clinical and biological classification of PTB, and to assess BPA exposure at multiple time points during pregnancy with regards to the risk of PTB.
Only one other study has assessed the odds of delivering a PTB in relation to BPA exposure. In a small nested case–control study (n = 30 cases, n = 30 controls) of PTB in Mexico City, the unadjusted OR of delivering at < 37 weeks in relation to specific gravity–corrected third-trimester urinary BPA concentration was 2.5 (95% CI: 1.1, 6.0) (Cantonwine et al. 2010). It is interesting to note that although the overall results from the present study differ from the one conducted in Mexico City, our analysis from late-pregnancy (visit 4) BPA concentrations with regard to spontaneous PTB closely parallels what was found by Cantonwine et al. (2010). Although type of preterm delivery was not reported by the study authors, the timing of BPA measurements between our study [median, 35.1 weeks (range, 33.1–38.3 weeks)] and the Mexico City study [median, 33.4 weeks (range, 30.6–37.7 weeks)] was very similar. However, only 66 of our 130 cases provided a urine sample at visit 4, and the final gestational age for those cases ranged from 35.0 to 36.9 weeks, reflecting only late preterm births.
Other studies have found inconsistent evidence between BPA and gestational length. In a small study of 40 pregnant women living in southeastern Michigan, there were no differences in gestational length between women with plasma BPA concentrations > 5 and ≤ 5 ng/mL (Padmanabhan et al. 2008). Wolff et al. (2008) found no significant associations between urinary BPA concentration during the third trimester and gestational length among 367 infants living in New York City (NY, USA). Conversely, Weinberger et al. (2014) reported a 1.1-day decrease (95% CI: –2.0, –0.1) in gestational length associated with an inter-quartile range (180.1 ng/mL) increase in BPA urinary concentrations exposure during pregnancy among 72 women living in New Jersey. Reasons for the conflicting evidence between BPA exposure and either risk of PTB or gestational length may include differences in study size and design, differences in populations, use of differing biological media for exposure assessment, or other factors.
In our secondary analysis, we further analyzed preterm birth by differing clinical presentation. Our intention was not only to explore potential mechanistic links between BPA exposure and PTB but also to address potential bias that could arise from including medically indicated protocol-driven PTBs. These preterm births, usually after 36 weeks, are commonly delivered for reasons associated with obstetrical practice rather than actual underlying pathobiology, and without intervention they would have likely proceeded to term. By excluding certain types of protocol-driven PTBs in our data set, we observed a strengthening of our overall associations (i.e., spontaneous + placental PTBs; see Supplemental Material, Table S2 http://ehp.niehs.nih.gov/wp-content/uploads/123/9/ehp.1408126.s001.acco.pdf), suggesting that inclusion of these PTBs biases our results toward the null.
Placental preterm birth comprised preterm birth following preeclampsia or intrauterine growth restriction (IUGR), which can result from impaired placentation early in pregnancy. The intrauterine environment in early stages of placentation is highly sensitive, and increases in oxidative stress can lead to apoptosis and altered cytotrophoblast turnover rate in the developing placenta (Burton et al. 2009; Heazell and Crocker 2008). Mechanistic evidence exists for the role of BPA in inducing oxidative stress and effecting trophoblast turnover (Babu et al. 2013; Benachour and Aris 2009; Jin and Audus 2005; Morice et al. 2011; Tachibana et al. 2007; Yang et al. 2009). Relevant to placental development, Benachour and Aris (2009) reported dose-dependent apoptosis in isolated primary cytotrophoblast cells from term placentas. Moreover, Morice et al. (2011) demonstrated in vitro that BPA exposure at environmentally relevant concentrations can affect the proliferative process in trophoblastic cells through the ERRγ1. Additionally, that study demonstrated the presence of ERRγ1 in several trophoblastic cell lines and isolated extravillous and villous cytotrophoblasts from first-trimester placenta, further suggesting another potential mechanistic link to disruption of normal placental development by exposure to BPA (Morice et al. 2011). Even though our study found no significant relationships between BPA exposure and risk of placental PTB, these results should be interpreted cautiously because we had very few (n = 35) placental preterm births in our population, and calculated odds ratios, though not statistically significant, were somewhat elevated.
Spontaneous preterm birth was clinically defined as resulting from spontaneous preterm labor or preterm premature rupture of membranes, both of which primarily arise from inappropriate initiation of an intrauterine inflammatory cascade (Challis et al. 2009). BPA has been shown to stimulate the production of pro-inflammatory cytokines (Lee et al. 2003; O'Brien et al. 2014; Tian et al. 2003) and can induce T-helper (Th)–1 type cytokines while simultaneously suppressing Th-2 cytokines (Youn et al. 2002). Additionally, it has been shown in human populations that BPA concentrations are associated with increased serum C-reactive protein levels (Lang et al. 2008; Yang et al. 2009). It is interesting to speculate that our finding that BPA concentrations measured late in pregnancy (mean gestational age, 35 ± 1 weeks) were associated with significantly increased odds of delivering a spontaneous PTB may be representing an enhanced maternal response to inflammation via BPA exposure leading to either PPROM or spontaneous preterm labor. However, these results should also be interpreted cautiously due to limited numbers of preterm subtypes.
Our results suggest that female infants may be more sensitive to being delivered preterm in relation to gestational BPA exposure than males. Endocrine-disrupting effects of BPA are well studied, and multiple animal and human studies have reported evidence of sex-specific adverse health effects resulting from BPA exposure (Braun et al. 2011b; Kubo et al. 2003; Perera et al. 2012; Rubin et al. 2006). Only one other study has looked at sex-specific relationships with association with BPA exposure and gestational length, but not risk of PTB (Weinberger et al. 2014). Past research has demonstrated that female fetuses are more sensitive to the changes in inflammatory stressors (Cankar et al. 2004; Clifton and Murphy 2004), though further research is needed to understand the potential mechanisms of our findings.
Our study had several strengths, including a repeated time point assessment of BPA exposure, ultrasound dating of gestational age, physician-validated clinical outcomes, and a large number of subjects and preterm cases, which allowed for exploring the heterogeneous nature of PTB. Still, results from our secondary analyses of subtypes of PTB and stratification on infant sex should be interpreted cautiously, given that we were limited in our number of placental and spontaneous PTB cases and are likely to be underpowered to detect subtle relationships. There was also no control for multiple comparisons, which may lead to an inflated type 1 error rate. We acknowledge that the few significant associations found in our analysis may be attributable to chance alone, and larger follow-up studies to replicate the findings are warranted. We substituted BPA values below the LOD with LOD divided by the square root of two in the present study. Although several authors have shown that using a maximum-likelihood estimate for LOD can reduce the inherent bias in using a substitution method for LOD, the improvements are minimal (1–2% reduction in bias) (Cole et al. 2009; Dinse et al. 2014; Guo et al. 2010).
Additionally, this study was limited in our understanding of other residual confounders, such as diet, which may be associated with BPA levels and PTB. It is worth noting that the generalizability of our population may be restricted because Brigham and Women's Hospital is a major referral center for high-risk pregnancies; thus, our population might have a higher than normal proportion of women with prior conditions that could lead to medically indicated protocol-driven PTB (e.g., placement of abdominal cerclage for cervical insufficiency). Regardless, this population was originally collected in a prospective fashion from a major regional hospital in a large urban center whose overall demographic breakdown of delivery patients is reflective of the general U.S. population. Our choice of a nested case–control design stemmed from financial considerations for exposure assessment and may have some inherent limitations. Analyzing continuous gestational age in a time-to-event framework in the entire cohort would yield more informative inference if we could obtain exposure measures on a larger cohort; however, given the resources, the case–control design maximizes our power to detect exposure–outcome associations. Although, to our knowledge, this was the first study of preterm birth or gestation length to use multiple urinary BPA concentrations from each woman, the low temporal reliability of BPA concentrations across pregnancy may indicate that even with three to four repeated measures, there may still be substantial nondifferential exposure measurement error, which would further limit statistical power to detect associations.