Pregnancy urinary phenol biomarker concentrations in relation to serum levels of inflammatory cytokines: Results from the EARTH study

Abstract

Background:

Evidence on the association between maternal phenol exposure and inflammation during pregnancy is limited and inconsistent.

Objective:

To evaluate associations between urinary phenol biomarkers and serum inflammatory cytokines across pregnancy, and to examine whether associations vary by trimesters.

Methods:

We included 175 pregnant women from the Massachusetts General Hospital Fertility Center and participating in the Environment and Reproductive Health (EARTH) Study (2005–2017), with available data on urinary concentrations of eight phenol biomarkers and serum inflammatory biomarkers, high-sensitivity C-reactive protein (hsCRP) and interleukin-6 (IL-6). Linear regression models were employed to assess the association between individual phenol biomarker concentrations and log-transformed inflammatory cytokine levels, while Bayesian Kernel Machine Regression (BKMR) models were utilized to evaluate phenol biomarker mixtures. Analyses were further stratified by the trimester of sample collection.

Results:

Overall, detectable urinary ethylparaben was positively associated with serum hsCRP (β: 0.464; 95 % CI: 0.012, 0.917). In trimester-specific analyses, urinary butylparaben was positively associated with hsCRP (β: 0.533; 95 % CI: 0.006, 1.059) in the first trimester, but negatively associated with IL-6 (β: −0.613; 95 % CI: −1.062, −0.164) in the second trimester. Urinary bisphenol A was inversely associated with hsCRP (β: −0.428; 95 % CI: −0.731, −0.125) in the third trimester.

Conclusions:

Our findings suggest that exposure to certain phenols may disrupt inflammatory profiles in pregnancy, with effects varying by trimesters. These novel associations underscore the importance of exposure timing when assessing environmental risk factors for maternal and offspring health outcomes.

1. Introduction

Phenols are widely used in consumer and industrial products and are recognized as endocrine-disrupting chemicals (EDCs) (Cantonwine et al., 2015, Gore et al., 2015, Watkins et al., 2015, Minguez-Alarcon et al., 2016a, Kek et al., 2024). Phenols include parabens, triclosan, and bisphenol A (BPA) (Kek et al., 2024). EDCs, including phenols, are linked to various health issues, such as metabolic, endocrine, and inflammatory disturbances (Cantonwine et al., 2010, Cantonwine et al., 2015, Watkins et al., 2015, Ferguson et al., 2016, James-Todd et al., 2016, Aung et al., 2019, Kelley et al., 2019, Kek et al., 2024).

Pregnancy is a particularly vulnerable period for exposure to these chemicals, with evidence indicating that some phenols can alter physiological processes affecting both the pregnant individual and the developing fetus, potentially impacting offspring’s health (Cantonwine et al., 2010, Kelley et al., 2019, Aker et al., 2020, Kek et al., 2024). Moreover, the inflammatory state is crucial for the progression of pregnancy, influencing its initiation, maintenance, and completion (Dutta et al., 2024), with higher inflammation during pregnancy being associated with adverse outcomes, such as preterm birth, preeclampsia, and fetal growth restriction (Aung et al., 2019, Kelley et al., 2019, Sturla Irizarry et al., 2024). Animal and in vitro studies have demonstrated that phenols can enhance the production and secretion of inflammatory cytokines, including interleukin (IL)-1β, tumor necrosis factor alpha (TNFα), and IL-6 (Zhao et al., 2017, Cho et al., 2018, Qiu et al., 2018, Qiu et al., 2019, Song et al., 2020). These effects are thought to occur via phenol-induced oxidative stress and the activation of several inflammatory signaling pathways, including the toll-like receptor 4 (TLR4)/nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) and mitogen-activated protein kinase (MAPK) cascades, the NOD-, LRR- and pyrin domain-containing protein 3 (NLRP3) inflammasome, and the disruption of antioxidant defense mechanisms regulated by nuclear factor erythroid 2–related factor 2 (Nrf2) (Lee et al., 2008, Meng et al., 2020, Xie et al., 2020, Rousseau-Ralliard et al., 2024). These findings suggested that inflammation may mediate the relationship between these chemicals and adverse pregnancy outcomes (Benachour and Aris 2009, Kek et al., 2024).

Some studies have examined the impact of phenol exposures on inflammatory cytokines, such as C-reactive protein (CRP) and IL-6, which are associated with an increased risk of preterm birth when present in amniotic fluid (Wei et al., 2010). However, findings have been mixed, with some studies reporting significant positive associations with BPA (Sturla Irizarry et al., 2024), triclosan (Watkins et al., 2015, Aung et al., 2019), and 2,5-dichlorophenol (Aung et al., 2019), while others found negative associations with ethylparaben (Aung et al., 2019), benzophenone-3 (Watkins et al., 2015, Aung et al., 2019), and butylparaben (Watkins et al., 2015). A potential explanation for these mixed results lies in the immunological shifts that naturally occur during pregnancy. The maternal immune system undergoes distinct transitions: a pro-inflammatory state in the first trimester to support implantation, an anti-inflammatory state in the second trimester to accommodate fetal growth, and a return to a pro-inflammatory state in the third trimester to facilitate labor (Mor et al., 2011, Mor et al., 2017). These dynamic changes influence circulating inflammatory markers and may modulate maternal responses to environmental exposures across gestation. Therefore, evaluating phenol-inflammation associations in a trimester-specific manner is biologically grounded and may help clarify prior inconsistent findings.

While most studies have focused on single pollutant associations (Aung et al., 2019), research on how chemical mixtures influence inflammation during pregnancy remains limited. Using data from the Environment and Reproductive Health (EARTH) Study of women seeking care at a fertility center, we evaluated the associations of urinary phenols and their mixtures with inflammatory markers, including high-sensitivity CRP (hsCRP) and IL-6, during pregnancy. Women experiencing sub and infertility often exhibit elevated systemic inflammation (Lainez and Coss 2019, Snider and Wood 2019, Fabozzi et al., 2022), potentially making them more susceptible to inflammatory responses triggered by environmental exposures. This population is also clinically relevant for studying early pregnancy health due to their higher risk of adverse reproductive outcomes (Lalani et al., 2018), although findings from this population may not fully generalize to the broader population of pregnant women. We hypothesized that higher urinary phenol concentrations during pregnancy would be associated with elevated serum levels of hsCRP and IL-6. Considering the dynamic inflammatory environment across pregnancy stages (Mor et al., 2011, Mor et al., 2017), we also conducted trimester-specific analyses to examine variations in these associations.

2. Methods

2.1. Study population

The EARTH Study is a prospective cohort that enrolled women and men seeking fertility evaluation and medically assisted reproductive treatment at the Massachusetts General Hospital (MGH) Fertility Center, with the goal of investigating the effects of environmental and dietary factors on fertility (Minguez-Alarcon et al., 2016b). Women aged 18–45 years who planned to use their own gametes for infertility treatment were eligible to participate. Among women referred by physicians, approximately 60 % of those approached by the research nurses agreed to enroll in the study (Chiu et al., 2018, Messerlian et al., 2018). Spot urine and non-fasting blood samples were collected at study visit, which were all collected in the morning.

Of the 991 women enrolled in the EARTH study (2004–2019), 699 became pregnant during the study period. From these, 200 with available blood samples collected during pregnancy (2005–2017) were randomly selected for serum inflammatory biomarker analysis. The present cross-sectional analysis included 175 women with data on both serum inflammtory cytokines and urinary phenol biomarkers. Of these, 61 (34.9 %) collected samples during the first trimester with a median (interquartile ranges, IQR) gestational age of 8 (7, 9) weeks, 47 (26.9 %) during the second trimester with a median (IQR) gestational age of 23 (20, 26) weeks, and 67 (38.3 %) during the third trimester with a median (IQR) gestational age of 34 (33, 36) weeks. Information on sociodemographic characteristics, lifestyle, and medical history, reproductive and occupational history, consumer products use, and physical activity were collected by research staff using questionnaires at enrollment. The participant’s weight and height were measured by trained study staff. Pre-pregnancy body mass index (BMI) was determined at enrollment by dividing weight (in kilograms) by height (in meters) squared. Physicians assigned infertility diagnoses based on the definitions provided by the Society of Assisted Reproductive Technology (SART). Information on pregnancy-related covariates, including clinical diagnoses of gestational diabetes mellitus (GDM) and gestational hypertension, was extracted from electronic medical records.

Dietary intake was assessed at cohort enrollment using a 131-item food frequency questionnaire (FFQ) that has been extensively validated (Salvini et al., 1989, Rimm et al., 1992, Yuan et al., 2017, Yuan et al., 2018). In brief, nutrient intakes, including total energy, were calculated based on the FFQ by summing the products of the intake frequency and the nutrient content of each food (per specified portion size) and supplement (per specified dose) (Minguez-Alarcon et al., 2021). Two dietary pattern scores, the Prudent and Western patterns (Gaskins et al., 2012), were derived to summarize overall food choices using factor analysis of 40 predefined food groups, as previously described (Minguez-Alarcon et al., 2021). Higher scores indicate greater adherence to each pattern, with the Prudent pattern reflecting healthier food choices (e.g., fish, vegetables, legumes), and the Western pattern representing less healthy items (e.g., red/processed meats, refined grains, sweets).

Perceived stress was assessed concurrently with blood and urine collection using the 4-item version of the Perceived Stress Scale (PSS-4), a validated short form of the original PSS-10 and PSS-14. Participants rated four items on a 5-point Likert scale (range: 0–16), with higher scores indicating greater perceived stress (Reddy et al., 2025). The PSS-4 has demonstrated acceptable internal consistency and validity in large population-based studies (Andreou et al., 2011, Vallejo et al., 2018, Ruisoto et al., 2020). The research received approval from the Human Subject Committees at the Harvard T.H. Chan School of Public Health, Massachusetts General Hospital (MGH), and the Centers for Disease Control and Prevention (CDC). Trained research staff explained the study procedures to the participants, who then provided informed consent after having all their questions addressed.

2.2. Exposure biomarker assessment

Urine samples were stored at −80 °C and shipped frozen overnight on dry ice to the CDC for analysis. Following the protocol described previously (Zhou et al., 2014), we employed online solid-phase extraction with isotope dilution-high-performance liquid chromatography-tandem mass spectrometry for the quantification in urine of: BPA, benzophenone-3, triclosan, methylparaben, propylparaben, butylparaben, ethylparaben, bisphenol S (BPS), and bisphenol F (BPF). Because the phenol panel expanded to include more compounds only in the later years, the number of participants with BPF data was limited (N = 19). Therefore, we did not include urinary BPF in our analysis. Limits of detection (LOD) ranged from 0.1 to 2.3 μg/L, depending on the phenol (Table 2). For each analytical batch, we included a set of calibrators, reagent blanks, and quality control (QC) materials. The QC concentrations were assessed using established statistical probability rules (Caudill et al., 2008). Should the QC samples not meet these statistical standards, all study samples in that batch were re-extracted. Phenol concentrations were measured using the CDC’s analytical method (CDC, 2021, 2022). Additional details about urinary phenol assessments have been described previously (Minguez-Alarcon et al., 2022). Specific gravity was measured at room temperature using a handheld refractometer (National Instrument Company, Inc., Baltimore, MD, USA). The device was calibrated with deionized water prior to each measurement. To avoid bias, uncorrected urinary phenol biomarker concentrations were used, and specific gravity was incorporated as a covariate in the analysis (Barr et al., 2005, Schisterman et al., 2005).

Table 2.

Associations and 95 %CIs of phenols with inflammatory markers from the linear regression model.

	N	Crude model		Adjusted model
		β (95 %CI)	p-value	β (95 %CI)	p-value
hsCRP
Bisphenol A	175	−0.088 (−0.238, 0.061)	0.245	−0.106 (−0.281, 0.069)	0.232
Bisphenol S
Non-detectable	33	Ref.		Ref.
Detectable	50	−0.060 (−0.509, 0.389)	0.791	−0.066 (−0.588, 0.456)	0.801
Benzophenone-3	133	0.003 (−0.172, 0.178)	0.972	0.013 (−0.170, 0.195)	0.891
Triclosan	133	−0.158 (−0.331, 0.015)	0.073	−0.163 (−0.338, 0.012)	0.068
Methylparaben	175	0.006 (−0.144, 0.156)	0.938	0.003 (−0.158, 0.163)	0.975
Propylparaben	175	0.017 (−0.133, 0.167)	0.825	−0.022 (−0.181, 0.138)	0.789
Butylparaben
Non-detectable	74	Ref.		Ref.
Detectable	101	0.283 (−0.017, 0.583)	0.064	0.281 (−0.008, 0.570)	0.057
Ethylparaben
Non-detectable	36	Ref.		Ref.
Detectable	47	0.353 (−0.084, 0.790)	0.112	0.464 (0.012, 0.917)	0.045
IL-6
Bisphenol A	175	−0.127 (−0.276, 0.022)	0.095	−0.095 (−0.254, 0.064)	0.239
Bisphenol S
Non-detectable	33	Ref.		Ref.
Detectable	50	−0.209 (−0.658, 0.239)	0.356	−0.207 (−0.694, 0.280)	0.399
Benzophenone-3	133	−0.120 (−0.296, 0.056)	0.179	−0.009 (−0.175, 0.157)	0.918
Triclosan	133	−0.045 (−0.221, 0.132)	0.619	−0.013 (−0.174, 0.149)	0.876
Methylparaben	175	−0.105 (−0.254, 0.044)	0.167	−0.068 (−0.213, 0.078)	0.359
Propylparaben	175	−0.064 (−0.214, 0.085)	0.397	−0.020 (−0.164, 0.125)	0.790
Butylparaben
Non-detectable	74	Ref.		Ref.
Detectable	101	−0.084 (−0.386, 0.219)	0.586	−0.051 (−0.316, 0.214)	0.702
Ethylparaben
Non-detectable	36	Ref.		Ref.
Detectable	47	0.179 (−0.265, 0.623)	0.424	0.164 (−0.272, 0.599)	0.456

Effect estimates are shown as per 1-SD change in inflammatory marker in phenol exposure. The crude model did not include any covariate adjustments. The adjusted model accounted for age at pregnancy, pre-pregnancy BMI, education level, race, infertility diagnosis, cycle type, number of babies, trimester, and specific gravity. Abbreviations: Confidence intervals (CI), high sensitivity C-reactive protein (hsCRP), interleukin-6 (IL-6), standard deviation (SD).

2.3. Outcome assessment

We centrifuged the blood samples at 3000 RPM for 20 min after the blood samples clotted. The resulting serum was then aliquoted, frozen, and stored at −80 °C until transferred to the Clinical and Epidemiologic Research Laboratory at Boston Children’s Hospital in Boston, MA. Two inflammatory cytokines, high hsCRP and IL-6, were measured in serum. The concentration of hsCRP was measured using an immunoturbidimetric assay on the Roche Cobas 6000 system (Roche Diagnostics – Indianapolis, IN). This FDA-approved, high-sensitivity assay had a detection threshold of 0.15 mg/L and exhibited day-to-day variability rates of 8.4 % at 0.53 mg/L and 2.1 % at 13.3 mg/L. The concentration of serum IL-6 was determined using an ultra-sensitive ELISA assay developed by R & D Systems, based in Minneapolis, MN. The assay’s sensitivity was as low as 0.09 pg/mL, with day-to-day variability of 10.8 %, 4.92 %, and 3.9 % at concentrations of 0.53, 2.75, and 5.58 pg/mL, respectively.

2.4. Statistical analysis

Demographic and reproductive features along with serum inflammatory biomarker levels were displayed as medians (IQRs) for continuous variables and counts (percentages) for categorical ones. Additionally, we described the distribution of urinary phenol concentrations in terms of percentiles and geometric means with standard deviations (SDs). We imputed phenol concentrations below the LOD using multiple imputation with the R package “mice” (Buuren and Groothuis-Oudshoorn, 2011). The concentrations of urinary phenols and serum inflammatory biomarkers were normalized using log-transformation and standardized as z-scores for association analysis. We fitted linear regression models to assess the association between these z-scores for urinary phenol concentrations and serum inflammatory biomarkers. Given their lower detection frequency compared to other phenol biomarkers, urinary BPS, butylparaben, and ethylparaben concentrations were categorized as binary variables (detectable vs. non-detectable) to assess their associations with inflammatory biomarkers.

Adjusted models included covariates such as age at pregnancy (continuous), pre-pregnancy BMI (continuous), education level (graduate degree vs. other), race (non-Hispanic white vs. other), infertility diagnosis (yes/no), cycle type [no medical treatment vs. in vitro fertilization (IVF) vs. intra-uterine insemination (IUI)], number of babies (singleton vs. twins/triplets), trimester (first vs. second vs. third), and specific gravity (continuous). We included specific gravity in the multivariable models to adjust for urinary dilution rather than creatinine, as specific gravity is less influenced by demographic and physiological factors (e.g., age, sex, muscle mass) (Barr et al., 2005, Kuiper et al., 2021), provides a more direct proxy for hydration status, and has demonstrated better performance and greater practicality in previous studies (Sauve et al., 2015). We examined model diagnostics to ensure that the assumptions of linear regression were met, including linearity, normality of residuals, homoscedasticity, and multicollinearity. Model assumptions were evaluated in representative regression models. Residual plots and Q–Q plots showed no major deviations from linearity, homoscedasticity, or normality assumptions. Variance inflation factor (VIF) values were all below 2.5, suggesting no evidence of multicollinearity.

Additionally, we stratified the participants into three subgroups based on the trimester during which their urine samples were collected. Within each subgroup, associations between urinary phenols and serum inflammatory biomarkers were analyzed using a linear regression model adjusted for the same confounders but excluding trimester. Sensitivity analyses were conducted: 1) by additionally adjusting for PSS-4 score (continuous), total energy intake (kcal/day), Prudent pattern score (continuous), Western pattern score (continuous), total physical activity (hour/wk), and adverse pregnancy outcome (yes/no); and 2) by excluding participants with extreme hsCRP or IL-6 values. Adverse pregnancy outcome was defined as a clinical diagnosis of GDM, gestational hypertension, or both during pregnancy. Extreme values were defined as values greater than the 75th percentile plus 1.5 times the IQR.

To analyze the combined effect of all phenols examined, we employed Bayesian Kernel Machine Regression (BKMR). We excluded BPS and ethylparaben from the BKMR analysis due to their relatively small sample sizes (N < 100), as these phenols were added to the measurement panel in later years. The BKMR approach is nonparametric and accommodates non-linear exposure–response relationships and complex interactions among exposures without preset assumptions about their nature. In BKMR, effects are quantified by averaging changes in the outcome associated with shifts in exposure biomarker concentrations across specific quantiles, while adjusting for the same confounders as in the adjusted model mentioned earlier. This method allows presentation of mixture associations as mean differences in hsCRP and IL-6 levels when there is an increase in the urinary concentrations of all phenols in the mixture from the 25th to the 75th percentiles, including corresponding 95 % credible intervals. We also present posterior inclusion probabilities (PIPs), which indicate the posterior probability of a significant association between an exposure and the outcome. To help visualize these findings, we provide several graphical outputs, (i) the overall mixture association resulting from a simultaneous quantile rise in all mixture components compared to median concentrations, (ii) exposure–response relationships for each biomarker with other biomarkers held at median levels, and (iii) single-exposure associations (mean differences from the 25th to 75th percentile) when all other biomarkers are set at the 25th, 50th, and 75th percentiles. Statistical analyses were performed with R (version 4.0.2).

3. Results

Our analysis included 175 women, with median age of 35 years (32, 38) and pre-pregnancy BMI of 22.9 kg/m² (21.2, 25.7) at enrollment (Table 1). Over half of the women (60 %) had a graduate degree and only 29 % reported a history of smoking. The vast majority conceived through fertility treatments (83 % using IUI or IVF), and 82 % experienced singleton pregnancies. Median (IQR) serum hsCRP and IL-6 were 4.2 mg/dL (1.8, 6.3) and 1.8 mg/dL (1.1, 2.6), respectively (Table 1).

Table 1.

Demographic and reproductive characteristics and serum inflammatory cytokine levels [median (IQR) or N (%)] among 175 pregnant women in the Environment and Reproductive Health (EARTH) Study (2005–2017).

Characteristics
Age, years	35.0 (32.0, 38.0)
Race, N (%)
White	154 (88)
Black	5 (3)
Asian	8 (5)
Other	8 (5)
Body Mass Index, kg/m2	22.9 (21.2, 25.7)
Ever smoked, N (%)	50 (29)
Graduate degree, N (%)	105 (60)
Primary Infertility diagnosis, N (%)
Male factor	58 (33 %)
Female factor	59 (34 %)
Unexplained	58 (33 %)
Cycle type resulting in pregnancy, N (%)
Without medical treatment	29 (17 %)
IUI	45 (26 %)
IVF	101 (57 %)
Trimester of sample collection, N (%)
1st	61 (35)
2nd	47 (27)
3rd	67 (38)
Number of babies, N (%)
Singleton (1)	144 (82)
Twins (2) or triplets (3)	31 (18)
Inflammatory biomarkers, (mg/dL)
hsCRP	4.2 (1.8, 6.3)
IL-6	1.8 (1.1, 2.6)

Abbreviations: Intrauterine insemination (IUI), in vitro fertilization (IVF), low-density lipoprotein (LDL), high sensitivity C-reactive protein (hsCRP), interleukin-6 (IL-6).

BPA, benzophenone-3, triclosan, methylparaben, and propylparaben were detected frequently (>80 %), whereas BPS, butylparaben, and ethylparaben had lower detection frequencies (<60 %) (Table S1). Median (IQR) concentrations of BPA and propylparaben were 1.00 (0.40, 1.80) μg/L and 23.2 (3.70, 105) μg/L, respectively. Median urinary concentrations of these phenols during pregnancy were generally lower compared to preconception values among women in the same EARTH study cohort (Minguez-Alarcon et al., 2015, Minguez-Alarcon et al., 2016a, Minguez-Alarcon et al., 2017, Minguez-Alarcon et al., 2019). Compared to female participants from the National Health and Nutrition Examination Survey (CDC, 2022), women in this study had lower urinary concentrations for most phenols, except propylparaben and the sunscreen agent benzophenone-3.

Most correlations between the concentrations of these phenols were statistically significant (p < 0.05), with Spearman correlation co-efficients (r) ranging from −0.11 to 0.86 (Fig. S1). Methylparaben showed a strong correlation with propylparaben (r = 0.86), while moderate correlations were observed between methylparaben and ethylparaben (r = 0.47), and between propylparaben and ethylparaben (r = 0.41). Other correlations were relatively weak. The distribution of benzophenone-3 concentrations varied across trimesters, with the highest levels observed in the 2nd trimester and the lowest levels in the 3rd trimester (Table S2). The median (P25, P75) serum IL-6 concentration was lowest in the 2nd trimester [1.2 mg/dL (1.0, 1.9)] but similar between the 1st trimester [2.0 mg/dL (1.3, 2.8)] and 3rd trimester [2.1 mg/dL (1.6, 2.6)] (Table S2).

In the crude model, no significant relationships were observed between urinary phenols and serum inflammatory biomarkers (Table 2). After adjusting for potential confounders, including age at sample collection, pre-pregnancy BMI, education, race, infertility diagnosis, cycle type, number of fetuses, trimester, and specific gravity, pregnant women with detectable ethylparaben had significantly higher serum hsCRP levels (β: 0.464; 95 % CI: 0.012, 0.917; p-value: 0.045; Table 2). Additionally, detectable butylparaben was also associated with increased hsCRP levels (β: 0.281; 95 % CI: −0.008, 0.570; p-value: 0.057), while triclosan showed a borderline inverse association with hsCRP (β: −0.163; 95 % CI: −0.338, 0.012; p-value: 0.068); however, neither association reached statistical significance (Table 2). In trimester-specific analyses, detectable urinary butylparaben measured in the first trimester was positively associated with serum hsCRP (β: 0.533; 95 % CI: 0.006, 1.059; p-value: 0.047), while detectable butylparaben measured in the second trimester was negatively associated with serum IL-6 (β: −0.613; 95 % CI: −1.062, −0.164; p-value: 0.009; Table 3). Additionally, a 1-SD change in urinary BPA was associated with decreased serum hsCRP levels among pregnant women who provided samples during the third trimester (β: −0.428; 95 % CI: −0.731, −0.125; p-value: 0.006; Table 3).

Table 3.

Associations and 95%CIs of phenols with inflammatory markers from the linear regression model stratified by trimesters of urine sample collection.

	N	hsCRP		IL-6
		β (95 %CI)	p-value	β (95 %CI)	p-value
1st trimester
Bisphenol A	61	0.144 (−0.171, 0.459)	0.362	0.020 (−0.277, 0.317)	0.893
Bisphenol S
Non-detectable	13	Ref.		Ref.
Detectable	24	−0.206 (−1.362, 0.950)	0.716	−0.146 (−1.112, 0.821)	0.759
Benzophenone-3	48	0.190 (−0.219, 0.600)	0.351	−0.030 (−0.409, 0.349)	0.872
Triclosan	48	−0.015 (−0.354, 0.324)	0.930	0.065 (−0.244, 0.374)	0.673
Methylparaben	61	0.031 (−0.252, 0.313)	0.829	−0.009 (−0.274, 0.255)	0.944
Propylparaben	61	−0.072 (−0.374, 0.230)	0.635	0.052 (−0.230, 0.335)	0.712
Butylparaben
Non-detectable	29	Ref.		Ref.
Detectable	32	0.533 (0.006, 1.059)	0.047	0.306 (−0.199, 0.811)	0.230
Ethylparaben
Non-detectable	12	Ref.		Ref.
Detectable	25	0.159 (−0.830, 1.148)	0.743	0.088 (−0.738, 0.915)	0.827
2nd trimester
Bisphenol A	47	−0.072 (−0.446, 0.301)	0.697	−0.106 (−0.391, 0.180)	0.458
Bisphenol S
Non-detectable	7	Ref.		Ref.
Detectable	10	−0.180 (−1.410, 1.051)	0.723	−1.103 (−3.133, 0.927)	0.221
Benzophenone-3	32	0.102 (−0.210, 0.414)	0.503	0.047 (−0.254, 0.348)	0.749
Triclosan	32	−0.238 (−0.550, 0.074)	0.127	−0.063 (−0.378, 0.252)	0.682
Methylparaben	47	0.048 (−0.285, 0.381)	0.773	−0.098 (−0.352, 0.156)	0.439
Propylparaben	47	0.015 (−0.330, 0.360)	0.930	−0.155 (−0.415, 0.104)	0.233
Butylparaben
Non-detectable	17	Ref.		Ref.
Detectable	30	−0.180 (−0.821, 0.462)	0.573	−0.613 (−1.062, −0.164)	0.009
Ethylparaben
Non-detectable	9	Ref.		Ref.
Detectable	8	0.300 (−0.581, 1.180)	0.422	0.019 (−1.795, 1.832)	0.980
3rd trimester
Bisphenol A	67	−0.428 (−0.731, −0.125)	0.006	−0.216 (−0.532, 0.100)	0.176
Bisphenol S
Non-detectable	13	Ref.		Ref.
Detectable	16	0.196 (−0.704, 1.097)	0.651	0.280 (−0.423, 0.983)	0.412
Benzophenone-3	53	−0.012 (−0.312, 0.287)	0.934	−0.026 (−0.291, 0.239)	0.845
Triclosan	53	−0.183 (−0.462, 0.095)	0.191	−0.088 (−0.338, 0.163)	0.484
Methylparaben	67	−0.062 (−0.348, 0.224)	0.664	−0.157 (−0.437, 0.123)	0.267
Propylparaben	67	0.016 (−0.239, 0.271)	0.902	−0.031 (−0.283, 0.221)	0.807
Butylparaben
Non-detectable	28	Ref.		Ref.
Detectable	39	−0.097 (−0.564, 0.370)	0.679	−0.115 (−0.576, 0.346)	0.620
Ethylparaben
Non-detectable	15	Ref.		Ref.
Detectable	14	0.658 (−0.077, 1.393)	0.076	0.161 (−0.474, 0.796)	0.600

Effect estimates are shown as per 1-SD change in inflammatory marker in phenol exposure stratified by trimesters of urine sample collection. The linear regression model was adjusted for age at pregnancy, pre-pregnancy BMI, education level, race, infertility diagnosis, cycle type, number of babies, and specific gravity. Abbreviations: high sensitivity C-reactive protein (hsCRP), interleukin-6 (IL-6), confidence interval (CI).

Similar trends were observed in a sensitivity analysis that additionally adjusted for total energy intake, Prudent pattern score, Western pattern score, total physical activity, and adverse pregnancy outcome. Specifically, positive associations of detectable ethylparaben (β: 0.461; 95 % CI: −0.003, 0.926; p-value: 0.051) and butylparaben (β: 0.299; 95 % CI: 0.005, 0.592; p-value: 0.046) with hsCRP remained (Table S3). Moreover, triclosan was inversely associated with hsCRP levels (β: −0.189; 95 % CI: −0.368, −0.010; p-value: 0.039). Similar results were also observed in the trimester-specific analyses (Table S3).

After excluding participants with extreme values of hsCRP and IL-6, a positive but non-significant associations between ethylparaben and hsCRP was still observed in the overall population (β: 0.267; 95 % CI: −0.165, 0.699; p-value: 0.221; Table S4). Notably, all statically significant findings were retained in the trimester-specific analyses after excluding extreme values (Table S4).

In the BKMR analysis, we did not observed a significant overall trend between the phenol mixture concentration and serum hsCRP (Fig. 1A) or serum IL-6 (Fig. 2A) when all the phenol biomarkers were at their 10th percentile to 90th percentile, compared to all at their 50th percentile. The PIPs for hsCRP ranged from 0.24 to 0.39, with triclosan having the highest PIP of 0.39 (Table S5). Similarly, for IL-6, the PIP ranged from 0.13 to 0.53, with methylparaben having the highest PIP of 0.53 (Table S5). Analysis of the exposure–response curves also did not show significant associations for any phenol with inflammatory biomarkers, when all other components were held at their median concentrations (Figs. 1B, 2B). Moreover, when examining the relationships between each individual phenol and the inflammatory biomarkers, with all other phenol biomarkers fixed at various quantiles (0.25, 0.50, 0.75), no interaction effects between these phenols were observed (Figs. 1C, 2C).

Fig. 1. Bayesian Kernel Machine Regression mixture associations of urinary phenol biomarkers with serum hsCRP levels (N = 133).

The model was adjusted for age at pregnancy, pre-pregnancy BMI, education level, race, infertility diagnosis, cycle type, number of babies, trimesters, and specific gravity. (A) Overall associations of the phenol biomarker mixture at different concentration percentiles compared to the 50th percentile and their 95 % credible intervals. (B) Exposure-response relationships between each phenol biomarker and hsCRP level while holding all other biomarkers at their median concentrations. (C) Mean difference in hsCRP comparing the 75th to 25th percentile of each phenol biomarker (estimates and 95 % credible intervals) when all the other phenol biomarker concentrations were fixed at the 25th, 50th, and 75th percentiles. Abbreviations: High sensitivity C-reactive protein (hsCRP).

Fig. 2. Bayesian Kernel Machine Regression mixture associations of urinary phenol biomarkers with serum IL-6 levels (N = 133).

The model was adjusted for age at pregnancy, pre-pregnancy BMI, education level, race, infertility diagnosis, cycle type, number of babies, trimesters, and specific gravity. (A) Overall association of the phenol biomarker mixture at different concentration percentiles compared to the 50th percentile and their 95 % credible intervals. (B) Exposure-response relationships between each phenol biomarker and IL-6 level while holding all other biomarkers at their median concentrations. (C) Mean difference in IL-6 comparing the 75th to 25th percentile of each phenol biomarker (estimates and 95 % credible intervals) when all the other phenol biomarker concentrations were fixed at the 25th, 50th, and 75th percentiles. Abbreviations: Interleukin-6 (IL-6).

4. Discussion

This study investigated the relationships between urinary phenol biomarker concentrations and serum inflammatory cytokine levels among pregnant women with fertility issues enrolled in the EARTH Study. We applied both linear regression and BKMR analysis to evaluate the associations of individual phenols and phenol mixtures with inflammation. Overall, pregnant women with detectable ethylparaben had higher serum hsCRP levels. In trimester-specific analyses, detectable urinary butylparaben was positively associated with serum hsCRP in the first trimester but inversely associated with serum IL-6 in the second trimester. Additionally, higher urinary BPA was linked to lower serum hsCRP concentrations among pregnant women who provided samples during the third trimester. These results underscore the importance of considering trimester-specific associations, given the dynamic changes in inflammation throughout the course of pregnancy. Further studies are warranted due to the scarce literature on the topic.

To our knowledge, only a few epidemiological studies have investigated the impact of phenols on the levels of inflammatory CRP and IL-6 among pregnant women, showing contradictory results. For example, repeated measurements of certain chemicals during pregnancy, such as BPA (Ferguson et al., 2016, Sturla Irizarry et al., 2024), triclosan (Watkins et al., 2015, Aung et al., 2019), and 2,5-dichlorophenol (Aung et al., 2019), have been associated with increased levels of serum CRP and/or IL-6 during the same period. On the other hand, Kelley et al., observed no significant associations between early pregnancy exposure to phenol mixtures with pregnancy serum IL-6 levels in a cohort of 56 pregnant women from Michigan (Kelley et al., 2019). Similarly, Aung et al., (Aung et al., 2019) found no associations of propylparaben, butylparaben, and ethylparaben with IL-6 levels in pregnant women from Massachusetts. The observed differences among studies may relate to differences in study population demographics (e.g., race, socioeconomic status), specific examined chemicals and methods to correct for urine dilution, and the measured inflammatory biomarkers. In addition to these factors, the inflammatory environment varies across pregnancy stages (Mor et al., 2011, Mor et al., 2017), so the observed discrepancies may also relate to lack of consideration of trimester-specific associations as inflammation changes during pregnancy (Li et al., 2020, Trasande et al., 2024).

A growing number of studies on the harmful effects of human exposure to parabens have highlighted their potential for estrogen-like activity and oxidative properties (McGrath, 2003, Kang et al., 2013, Golestanzadeh et al., 2022). A positive association between urinary ethylparaben and serum hsCRP levels was observed among all participants in our study. Ethylparaben has also been associated with various chronic diseases and conditions that are strongly linked to increased inflammation and oxidative stress, including gestational diabetes mellitus (Liu et al., 2019), hypertension (Zhang et al., 2023), increased gestational weight gain, overweight/obesity (Wen et al., 2020), and asthma risk of offspring (Vernet et al., 2017).

In the present study, we observed that the association between phenols and inflammation varied by pregnancy stage. Notably, we were the first to report that urinary butylparaben was positively associated with serum hsCRP during the first trimester but negatively associated with IL-6 during the second trimester. Interestingly, the exposure–response curve for butylparaben and hsCRP (Fig. 1B) suggests a potential non-linear relationship, which has not been formally evaluated and remains largely unexplored in the literature. Such a non-linear pattern may partially underline the trimester-specific associations we observed, where a significant positive association was seen only in early pregnancy. Additionally, urinary BPA was negatively associated with serum hsCRP exclusively among pregnant women who provided samples during the third trimester. Supporting our findings, a recent study by Sturla Irizarry SM et al., (Sturla Irizarry et al., 2024) demonstrated that the effects of phenols on inflammatory biomarkers were modified by pregnancy stage. For example, BPS was positively associated with the pro-inflammatory cytokines matrix metalloproteinases (MMP) 1, 2, and 9 in early pregnancy, whereas BPA was negatively associated with MMP9 in late pregnancy (Sturla Irizarry et al., 2024). Previous studies also have reported conflicting results regarding the relationship between butylparaben and inflammation or oxidation. Using repeated measurements of urinary butylparaben and inflammation/oxidation biomarkers, Watkins et al., (Watkins et al., 2015) found a positive association with serum isoprostane, an oxidative stress biomarker, but a negative association with serum CRP. Moreover, other studies involving pregnant women have reported significant negative associations of phenols, such as benzophenone-3 (Watkins et al., 2015, Aung et al., 2019) and butylparaben (Watkins et al., 2015), with levels of CRP and/or IL-6.

Recent studies have shown that exposure to EDCs, including phenols, can dysregulate immune signaling pathways, including NF-κB and MAPK activation, and impair maternal-fetal immune tolerance (Rousseau-Ralliard et al., 2024). These effects may be particularly relevant in the context of pregnancy, during which the maternal immune system undergoes tightly regulated, stage-specific adaptations: a pro-inflammatory state in the first trimester to support implantation and placentation, an anti-inflammatory state in the second trimester to facilitate fetal growth, and a return to a pro-inflammatory state in the third trimester to prepare for labor (Mor et al., 2011, Mor et al., 2017). Consistent with previous studies (Jarmund et al., 2021), IL-6 levels were reduced during the second trimester. In contrast, hsCRP appears to be a notable exception, with some studies reporting no consistent pattern (Belo et al., 2005) or even a gradual increase in levels throughout pregnancy (Yu et al., 2019). These immunological shifts shape the baseline levels and responsiveness of inflammatory cytokines, potentially altering susceptibility to exogenous immunomodulatory agents such as phenols (Schjenken et al., 2021). As Schjenken et al., further emphasized, disruption of these temporal immune dynamics by environmental exposures may lead to differential health effects depending on the timing of exposure (Schjenken et al., 2021). Additionally, behavioral factors during pregnancy, such as changes in the use of personal care products, may also influence phenol exposures and their interaction with inflammatory pathways. This dynamic physiological and behavioral context may help explain the trimester-specific associations observed in our study, including the positive association between butylparaben and hsCRP in early pregnancy and the inverse association with IL-6 in mid-pregnancy.

Furthermore, previous research has revealed that parabens can interact with both estrogen and progesterone receptors in experimental studies, influencing immunomodulation and immune tolerance (Kiyama and Wada-Kiyama, 2015, Nair et al., 2017). Progesterone receptor signaling promotes anti-inflammatory cytokines in various immune cells, thereby exerting anti-inflammatory effects (Druckmann and Druckmann, 2005). Similarly, estrogen receptors in uterine natural killer cells stimulate the expression of anti-inflammatory cytokines, also contributing to anti-inflammatory responses (Wilczynski, 2005). The observed negative association between phenolic exposure and inflammatory biomarkers indicates that phenols could potentially influence the inflammatory response.

Several limitations should be considered in this study. Firstly, the generalizability of the findings to the broader pregnant population is limited, as this sample consists of women who attended a fertility center. The relatively high educational attainment and predominantly White racial composition of the study population may further limit the generalizability of our findings. Nevertheless, this study population holds significant public health relevance, given that women with fertility problems often exhibit higher levels of systemic inflammation (Lainez and Coss, 2019, Snider and Wood, 2019, Fabozzi et al., 2022). However, these baseline differences in inflammatory profiles may also confound the observed associations. Secondly, the cross-sectional design limits the ability to draw causal inferences. Thirdly, as with any observational study, residual confounding from other chemical exposures, lifestyle, medication, and reproductive factors cannot be ruled out. Fourth, given the relatively modest sample size in our study, particularly in trimester-specific analyses, some trimester-specific findings should be interpreted with caution. The observed variability across trimesters may reflect true biological differences in immune regulation during pregnancy, but further research is needed to confirm these findings. Although our sample size was comparable to or larger than previous studies reporting significant associations in similar populations (Souter et al., 2013, Welch et al., 2021), the power to detect weaker associations may still have been limited, as reflected in the relatively wide confidence intervals for some estimates. Finally, another limitation is the potential misclassification of phenol exposure because of phenols’ relatively short half-life in humans and the likely episodic nature of the exposures, as we collected urine samples at a single time point. Nonetheless, we have previously shown in the EARTH cohort that a single urine sample can adequately reflect an individual’s exposure to short half-lived chemicals over several months (Braun et al., 2012, Smith et al., 2012). In addition, the time-sensitive nature of inflammatory cytokines, combined with the cross-sectional design of this study, may limit our ability to capture the temporal alignment between exposure and outcome.

Despite these limitations, this study has several strengths. The primary strength is the use of advanced statistical methods to analyze mixtures of exposure biomarkers in relation to inflammatory cytokines. Secondly, the present study was conducted within a well-established cohort of women with fertility issues, who are at high risk for inflammatory infiltration and subsequent adverse pregnancy outcomes (Palomba et al., 2016, Fabozzi et al., 2022). Finally, we adjusted for a range of reproductive and lifestyle factors, theryby reducing the concern about confounding.

In conclusion, our findings suggest potential associations between exposure to certain phenols and serum inflammatory markers during pregnancy, which may vary across gestational stages. Specifically, urinary ethylparaben was positively associated with serum hsCRP levels among overall participants. In trimester-specific analyses, detectable urinary butylparaben was associated with higher serum hsCRP in the first trimester but with lower serum IL-6 in the second trimester. Additionally, pregnant women with higher urinary BPA levels had lower serum hsCRP in the third trimester. The complexity of chemical mixtures and the dynamic nature of the inflammatory environment during pregnancy warrants further research to elucidate the effects of phenols at different stages. Future studies should explore the underlying mechanisms and assess the potential long-term implications for both maternal and offspring health.

These findings have important public health implications, as they highlight the potential for phenol exposure to influence inflammatory pathways during critical periods of pregnancy. Given that many phenols, commonly found in personal care products and consumer goods, are detectable in most pregnant women, our results support the need for additional research to evaluate the potential health impacts of these compounds. Incorporating evidence of trimester-specific susceptibility into risk assessments may help inform more targeted public health guidelines and exposure limits to protect maternal and fetal health.

Supplementary Material

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.envint.2025.109652.

Data availability

The data that has been used is confidential.