Prenatal Exposure to Bisphenols and Phthalates and Postpartum Depression: The Role of Neurosteroid Hormone Disruption

Abstract

Context

Postpartum depression (PPD) is a serious psychiatric disorder. While causes remain poorly understood, perinatal sex hormone fluctuations are an important factor, and allopregnanolone in particular has emerged as a key determinant. Although synthetic environmental chemicals such as bisphenols and phthalates are known to affect sex hormones, no studies have measured allopregnanolone and the consequences of these hormonal changes on PPD have not been interrogated.

Objective

To investigate associations of repeated measures of urinary bisphenols and phthalates in early and midpregnancy with serum pregnenolone, progesterone, allopregnanolone, and pregnanolone concentrations in midpregnancy and PPD symptoms at 4 months postpartum.

Methods

Prospective cohort study of 139 pregnant women recruited between 2016 and 2018. Bisphenols and phthalates were measured in early and midpregnancy urine samples. Serum sex steroid hormone concentrations were measured in midpregnancy. PPD was assessed at 4 months postpartum using the Edinburgh Postnatal Depression Scale (EPDS). Multiple informant models were fit using generalized estimating equations. Serum levels of allopregnanolone, progesterone, pregnanolone, and pregnenolone were examined as log-transformed continuous variables. PPD symptoms were examined as continuous EPDS scores and dichotomously with scores ≥10 defined as PPD.

Results

Di-n-octyl phthalate (DnOP) and diisononyl phthalate (DiNP) metabolites were associated with reduced progesterone concentrations. Log-unit increases in ∑DnOP and ∑DiNP predicted 8.1% (95% CI –15.2%, –0.4%) and 7.7% (95% CI –13.3%, –1.7%) lower progesterone, respectively. ∑DnOP was associated with increased odds of PPD (odds ratio 1.48; 95% CI 1.04, 2.11).

Conclusion

Endocrine disrupting chemicals may influence hormonal shifts during pregnancy as well as contribute to PPD.

Postpartum depression, phthalate, bisphenol, progesterone, allopregnanolone

Postpartum depression (PPD) is a serious psychiatric disorder that affects up to 20% of childbearing women (1). While its causes remain largely unknown, it has long been hypothesized to have an endocrine basis, with hormonal fluctuations throughout pregnancy and postpartum identified as an important factor (2). However, the literature linking sex steroid hormones such as estrogen and progesterone with PPD remains inconsistent. One promising new direction is allopregnanolone, a neuroactive metabolite of progesterone, which has been found to prospectively predict PPD. As a potent allosteric modulator of gamma-aminobutyric acid (GABA)_A receptors, allopregnanolone has anxiolytic and sedative effects (3, 4). Furthermore, a synthetic version of allopregnanolone, brexanolone, has recently been approved as the first medication to treat PPD (5, 6).

Pregnancy, childbirth, and postpartum constitute periods of dramatic hormonal and physiologic changes that heighten susceptibility to external factors like environmental chemicals (7). In particular, environmental chemicals that are endocrine disrupting chemicals (EDCs) can impact endogenous hormone levels in animals and humans (8). Bisphenols, used in polycarbonate plastics and aluminum can lining, and phthalates, used as plasticizers and in personal care products, are 2 classes of EDCs that are ubiquitous in the environment and detectable in nearly all US pregnant women (9). In experimental studies, bisphenol A (BPA) was associated with decreased estradiol and progesterone levels (10, 11) and di-2-ethylhexyl phthalate and dibutyl phthalate inhibited estradiol and progesterone production both in cycling and pregnant rats (12-14), suggesting that these chemicals alter steroidogenesis (15, 16). The few existing epidemiological studies have been consistent in associating BPA with lower estradiol among pregnant women (17) and women undergoing in vitro fertilization (18-20), as well as monoethyl phthalate and mono(carboxy-isooctyl) phthalate with reduced progesterone (21, 22). Although evidence supports effects on estrogen and progesterone, no studies have examined EDCs in relation to allopregnanolone.

Data from experimental studies in animals are consistent with the hypothesis that exposure to bisphenols and phthalates contributes to PPD (23). Studies have shown that BPA and phthalate administration during gestation affects maternal behavior and care after parturition, including poor maternal–pup attachment, suboptimal nest construction, less frequent nursing, more time spent outside the nest, and less licking and grooming of pups (24-26). In humans, few studies have linked chemical exposures with PPD. Although a recent study reported associations between prenatal levels of polybrominated diphenyl ethers and maternal depression, outcomes were considered up to 8 years postpartum, and hormones were not measured (27). In addition, 2 studies found associations between mono-n-butyl phthalate and adult depression, but they were cross-sectional and examined major depression in adults and elderly people, not PPD (28, 29). Therefore, 2 key gaps exist. First, while epidemiologic studies have evaluated associations between chemical exposures and sex steroid hormone levels (8), the allopregnanolone pathway has not been measured. Second, the consequences of these hormonal changes on perinatal mental health have not been interrogated.

The purpose of this study was to examine the associations between repeated measures of urinary bisphenols and phthalates in early and midpregnancy and sex steroid hormones on the allopregnanolone pathway assessed in midpregnancy as well as PPD symptoms at 4 months after delivery.

Materials and Methods

Study Population

This study was embedded within the New York University (NYU) Children’s Health and Environment Study (CHES). CHES is an ongoing, clinically enrolled, longitudinal cohort study of pregnant women and their children that is part of the Environmental Influences on Child Health Outcomes program (30). Pregnant women ≥18 years old, <18 weeks of gestation, and with nonmedically threatened pregnancies were recruited and enrolled through their routine prenatal care at 1 of 3 hospitals: NYU Langone Hospital—Manhattan, NYU Langone Hospital—Brooklyn, and Bellevue Hospital Center. Maternal biospecimens including urine and serum were collected at <18 weeks of gestation (ie, early pregnancy) and ≥18 to <25 weeks of gestation (ie, midpregnancy). Biopecimens were processed, aliquoted, and stored at –80°C in polycarbonate-free tubes until laboratory analysis.

This study comprised 139 pregnant women recruited between 2016 and 2018 with early (≥5-<18 weeks) and midpregnancy (≥18-<25 weeks) urine samples assessed for bisphenols and phthalate metabolites, midpregnancy serum samples assessed for sex steroid hormones, and data on postpartum depression at 4 months after delivery.

Measures

Questionnaires were administered to women in each trimester of pregnancy and included details on sociodemographic information, psychosocial factors, medical history, and health behaviors and experiences. Depressive symptoms during pregnancy were assessed using the Patient Health Questionnaire (PHQ)-9 in each trimester of pregnancy. Scores ≥10 indicated depression (31). Information on material hardship was collected during pregnancy through the US Census Survey of Income and Program Participation and constructed as a sum of experiences (32, 33). At 4 months postpartum, mothers completed an additional questionnaire detailing pregnancy complications, birth outcomes, infant health, and their health behaviors and social conditions. This questionnaire included the Edinburgh Postnatal Depression Scale (EPDS), which was used to measure PPD symptoms. Scores ≥10 were considered PPD cases (34-36). This threshold for PPD has been implemented by the international perinatal psychiatry consortium, Action Towards Causes and Treatment (PACT) (35); it represents the most sensitive threshold for PPD (37) and reflects a range of severity. Clinical data such as prepregnancy body mass index (BMI) and perinatal psychotropic medication use according to Anatomical Therapeutic Chemical codes (38) were abstracted from electronic health records.

Urinary cotinine was measured using high-performance liquid chromatography (HPLC) with tandem mass spectrometry (MS/MS) in the same samples in which bisphenols and phthalates were analyzed. This allowed for the quantification of nicotine exposure through either active or passive smoking. A cut-off value of ≥0.31 μg/g creatinine (Cr) was used to distinguish at least passive exposure to tobacco smoke from nonexposure (39).

Exposure Assessment

Bisphenol and phthalate concentrations were measured in early (<18 weeks) and midpregnancy (≥18-<25 weeks) urine samples at the NYU Human and Environmental Exposure Analysis Laboratory. Specifically, 8 bisphenols (BPA, bisphenol AF, bisphenol AP, bisphenol B, bisphenol F, bisphenol P, bisphenol S, and bisphenol PZ) and 22 phthalate metabolites were quantified (Table 1).

Table 1.

Bisphenol and phthalate metabolite concentration (ng/mL) distributions across pregnancy among 139 New York University Children’s Health and Environment Study participants, 2016-2018

		Early pregnancy (≥5–<18 weeks)			Mid pregnancy (≥18–<25 weeks)			ICC
Compounds	Average LOD	%<LOD	Median	IQR	%<LOD	Median	IQR
Bisphenols
Bisphenol A	0.15	30	0.68	<LOD, 1.5	26	0.77	0.11, 1.7	0.44
Bisphenol AF	0.04	71	<LOD	<LOD, 0.04	78	<LOD	<LOD, <LOD	—
Bisphenol AP	0.35	93	<LOD	<LOD, <LOD	94	<LOD	<LOD, <LOD	—
Bisphenol B	0.33	96	<LOD	<LOD, <LOD	100	<LOD	<LOD, <LOD	—
Bisphenol F	0.36	77	<LOD	<LOD, <LOD	73	<LOD	<LOD, 0.40	—
Bisphenol P	0.13	94	<LOD	<LOD, <LOD	96	<LOD	<LOD, <LOD	—
Bisphenol S	0.08	24	0.44	0.08, 1.2	17	0.62	0.27, 1.30	0.39
Bisphenol Z	0.30	99	<LOD	<LOD, <LOD	100	<LOD	<LOD, <LOD	—
Phthalates
Phthalic acid	0.12	1	18.2	9.3, 42.5	0	18.9	9.6, 39.0	0.30
Low molecular weight phthalate metabolites
Monomethylphthalate	0.04	40	1.6	<LOD, 7.1	41	1.4	<LOD, 6.9	0.53
Monoethylphthalate	0.04	1	38.4	17.9, 97.6	0	35.4	17.5, 102.0	0.43
Monobutyl phthalate	0.04	1	12.8	6.4, 28.5	1	16.3	5.9, 41.7	0.45
Mono-isobutyl phthalate	0.02	4	8.1	3.7, 16.3	3	9.5	3.6, 20.2	0.44
Monoisopropyl phthalate	0.15	82	<LOD	<LOD, <LOD	81	<LOD	<LOD, <LOD	—
Mono-n-pentenyl phthalate	0.19	99	<LOD	<LOD, <LOD	99	<LOD	<LOD, <LOD	—
Monocyclohexyl phthalate	0.04	93	<LOD	<LOD, <LOD	96	<LOD	<LOD, <LOD	—
High molecular weight phthalate metabolites
Di-2-ethylhexyl phthalate metabolites
Mono-(2-ethyl-5-carboxypentyl) phthalate	0.05	1	7.9	3.6, 13.6	1	6.9	3.9, 15.5	0.54
Mono-ethylhexyl phthalate	0.30	28	1.2	<LOD, 3.9	32	1.7	<LOD, 4.3	0.45
Mono-(2-ethyl-5-oxohexyl) phthalate	0.02	1	4.1	2.2, 7.7	0	4.4	2.0, 8.2	0.50
Mono-(2-ethyl-5-hydroxyhexyl) phthalate	0.04	1	7.4	3.6, 13.9	0	7.8	3.5, 12.6	0.48
Mono-(2-carboxymethylhexyl) phthalate	0.17	4	2.9	1.5, 8.0	1	3.3	1.5, 6.4	0.46
Di-n-octyl phthalate metabolites
Mono-n-octyl phthalate	0.04	96	<LOD	<LOD, <LOD	96	<LOD	<LOD, <LOD	—
Mono-(3-carboxypropyl) phthalate	0.22	6	0.94	0.52, 1.6	3	0.95	0.60, 1.8	0.33
Mono-(7-carboxy-n-heptyl) phthalate	0.11	46	0.17	<LOD, 1.4	44	0.33	<LOD, 1.4	0.32
Diisononyl phthalate metabolites
Mono(carboxyisooctyl) phthalate	0.15	0	1.5	0.79, 3.6	0	1.8	0.85, 4.5	0.23
Monoisononyl phthalate	0.02	44	1.0	<LOD, 4.4	42	1.0	<LOD, 6.5	0.92
Other high molecular weight metabolites
Mono(carboxy-isononyl) phthalate	0.17	22	0.97	0.23, 2.2	28	0.84	<LOD, 2.0	0.34
Mono-hexyl phthalate	0.02	65	<LOD	<LOD, 0.07	55	<LOD	<LOD, 0.08	—
Mono(2-heptyl) phthalate	0.03	58	<LOD	<LOD, 0.18	52	<LOD	<LOD, 0.19	—
Monobenzyl phthalate	0.03	11	3.2	1.1, 8.2	8	3.04	0.87, 10.4	0.57

Abbreviations: LOD, limit of detection; IQR, interquartile range.

To quantify bisphenols, urine extraction utilized enzymatic deconjugation followed by liquid–liquid extraction with ethyl acetate (40). Analysis was performed with HPLC coupled with electrospray MS/MS under negative mode of ionization. The isotopically labeled internal standards (IS) of target chemicals were spiked into every sample and used for quantification by isotope dilution method. The limits of detection (LODs) varied from 0.04 to 0.4 ng/mL.

To quantify phthalate metabolites, urine samples were processed using enzymatic deconjugation followed by off-line solid phase extraction with reversed phase HPLC electrospray MS/MS (41). Assay precision was enhanced by incorporating the IS for each of the phthalate metabolites, allowing for LODs in the range of 0.02 to 0.3 ng/mL. Urinary Cr, used to adjust for urinary dilution, was analyzed using HPLC-MS/MS as has been previously described (42).

Hormone Quantification

The end-products of the biosynthetic pathway for allopregnanolone were measured in stored midpregnancy (≥18-<25 weeks) serum samples at the Psychiatric Institute at the University of Illinois at Chicago College of Medicine. This included allopregnanolone, its GABAergic equipotent stereoisomer pregnanolone, progesterone, and their precursor, pregnenolone (43). This was done using gas chromatography mass spectrometry performed after separation of the steroids by HPLC as previously published (44, 45). Briefly, serum samples were extracted in ethylacetate and lyophilized. Target sex steroid hormones were then purified and separated using HPLC. Tritiated hormones were used to monitor retention time through the HPLC, while deuterated IS were added for the quantification of each target compound. Gas chromatography with mass spectrometry analysis was conducted in the electron impact ionization mode for all hormones.

Statistical Analyses

Bisphenols and phthalate metabolites were analyzed in groups as molar sums of metabolites based on their use in product categories and/or parent compounds (Table 2) including ∑bisphenols, ∑low molecular weight (LMW) phthalates comprising metabolites <250 Da and typically used in personal care products (46), ∑high molecular weight (HMW) phthalates comprising metabolites ≥250 Da and typically used in vinyl plastics (47), and ∑di (2-ethylhexyl) phthalate (DEHP), ∑di-n-octyl phthalate (DnOP), and ∑diisononyl phthalate (DiNP) metabolites. Individual metabolites that were infrequently detected (in <50% of samples) were excluded from molar sums. Phthalic acid was examined separately as an individual exposure as it is an end product metabolite for all phthalates. Measures below the LOD were imputed by the LOD/√2 (48).

Table 2.

Bisphenol and phthalate metabolite group concentration (nmol/mL) distributions according to molecular weight and/or parent compounds among 139 New York University Children’s Health and Environment Study Participants, 2016-2018

	Early pregnancy (≥5-<18 weeks)		Mid pregnancy (≥18-<25 weeks)
Compounds	Median	IQR	Median	IQR	ICC
∑Bisphenols	0.01	0.00, 0.01	0.01	0.00, 0.01	0.37
Bisphenol A
Bisphenol S
∑Low molecular weight phthalates	0.39	0.18, 0.93	0.43	0.17, 0.90	0.48
Monomethylphthalate
Monoethylphthalate
Monobutyl phthalate
Mono-isobutyl phthalate
∑High molecular weight phthalates	0.14	0.07, 0.28	0.14	0.07, 0.28	0.47
Mono(carboxy-isononyl) phthalate
Monobenzyl phthalate
ΣDi-2-ethylhexyl phthalate metabolites	0.08	0.04, 0.16	0.09	0.04, 0.16	0.51
Mono-(2-ethyl-5-carboxypentyl) phthalate
Mono-ethylhexyl phthalate
Mono-(2-ethyl-5-oxohexyl) phthalate
Mono-(2-ethyl-5-hydroxyhexyl) phthalate
Mono(2-carboxymethylhexyl) phthalate
ΣDi-n-octyl phthalate metabolites	0.01	0.00, 0.01	0.01	0.00, 0.01	0.28
Mono-(3-carboxypropyl) phthalate
Mono-(7-carboxy-n-heptyl) phthalate
ΣDiisononyl phthalate metabolites	0.01	0.00, 0.03	0.02	0.00, 0.03	0.54
Mono(carboxyisooctyl) phthalate
Monoisononyl phthalate

Abbreviations: ICC, intraclass correlation coefficient; IQR, interquartile range.

Distributions of all chemical exposures were assessed by computing medians and interquartile ranges as well as intraclass correlation coefficients to quantify the degree of variation over early and midpregnancy. Hormone distributions were examined by calculating geometric means and corresponding geometric standard deviations of each hormone across covariate strata.

In order to examine associations of bisphenols and phthalates with sex steroid hormone concentrations and PPD, multiple informant models were fit using generalized estimating equations with logit and linear links, depending on the parameterization of the outcome (ie, dichotomous or continuous, respectively) (49, 50). Multiple informant models accommodate repeated exposure measures, considering different exposure windows (ie, in this case, early and midpregnancy) as “informants” and provides a single, integrated estimate for the association between exposure over time and the outcome. Similar models were fit for PPD as the outcome. EPDS scores were considered continuously as well as a dichotomous variable where scores ≥10 defined PPD (34-36).

Separate models were fit for each exposure–outcome combination. All bisphenol and phthalate measures as well as hormone concentrations were examined as log-transformed continuous variables. All regression models controlled for covariates that were identified a priori based on the literature and our conceptualized directed acyclic graph. Adjustment sets were reduced in favor of parsimonious models because estimates did not change when we included a larger suite of potential confounders. For exposure–hormone models, this included urinary creatinine, gestational age at time of serum hormone sampling, maternal age, and prepregnancy BMI. For exposure-PPD models, this included urinary creatinine, maternal age, prepregnancy BMI, race/ethnicity, and education. In exposure–PPD models, we examined the potential influence of additionally controlling for antenatal depressive symptoms (ie, PHQ-9 scores), marital status, material hardship, and pregnancy complications (ie, preeclampsia and gestational diabetes) and results did not change. Thus, we present results without control for these factors.

To account for multiple testing, the P value was adjusted using a modified Bonferroni approach using the effective number of tests (51). This involved constructing a correlation matrix between all 7 exposures (ie, bisphenol and phthalate groupings) and a subsequent calculation using the resulting eigenvalues. The calculated effective number of comparisons was 4 and the adjusted P value was .0125 (α = .05/4).

In sensitivity analyses, we evaluated the influence of correcting for urinary creatinine using standardization as opposed to covariate adjustment (52). In addition, we ran all analyses excluding women who were taking antidepressants, anxiolytics, or antipsychotic medications during or after pregnancy (n = 7). Similarly, we fit the exposure–hormone models excluding women who were prescribed progesterone-containing medications during pregnancy (eg, hydroxprogesterone caproate) (n = 7) as well as those who were prescribed drugs that could potentially interact with progesterone (eg, antifungal agents) (n = 9). Finally, in post hoc analyses we examined midpregnancy hormone concentrations in relation to EPDS scores in linear regression models to evaluate the plausibility of phthalate-associated hormonal shifts to influence PPD symptoms. This study was approved by the Institutional Review Board at New York University Langone Medical Center.

Results

BPA and bisphenol S were frequently detected over time (70-76% and 74-83% in early and midpregnancy, respectively) (Table 1). Of the 22 phthalate metabolites assessed, 15 were detected in more than 50% of samples. These frequently detected metabolites contributed to the summed group measures (Table 2). The intraclass correlation coefficients of individual and grouped urinary bisphenols and phthalate metabolites varied widely, from 0.23 to 0.92, but the vast majority were in the range of 0.3 to 0.5, indicating fair to moderate reproducibility over time (Tables 1 and 2). Total bisphenols had weak correlations with all phthalate sums (Spearman’s ρ = 0.2-0.3, P < .0001) while correlations within phthalate groups were higher (Spearman’s ρ = 0.4-0.9, P < .0001) (data not shown).

Sex steroid hormone concentrations were log-normally distributed (Table S1, (53)) and steadily increased with gestational age (Fig. 1). Table 3 shows the characteristics of the study population and the geometric mean hormone levels by covariate strata. Most women (56%) were between 26 and 34 years with the mean age being 30.8 years (SD = 5.8). While pregnenolone and progesterone concentrations tended to decrease with age, allopregnanolone and pregnanolone increased. Overweight (32%) and obesity (30%) were common and hormone levels decreased with increasing BMI. The study population was ethnically and socioeconomically diverse, with 65% identifying as Hispanic and the majority having completed only high school or some college (66%). Both Hispanic women and those with less than a college degree tended to have lower hormone levels. Approximately one-quarter of women had urinary cotinine levels suggestive of at least passive exposure. While not all hormones showed variation across cotinine concentrations, allopregnanolone and pregnanolone were lower among those with at least passive exposure. Prenatal depression as defined by PHQ-9 scores ≥10 at any time in pregnancy occurred in 14% of women and hormone levels were consistently lower among these women than in those who were never depressed, although not statistically significantly. Similarly, progesterone and allopregnanolone were lower among the 12 women who went onto develop PPD symptoms, but pregnanolone levels were greater than in those who did not, although, again, these bivariate differences were not statistically significant. Finally, 7 women were prescribed psychotropic medications either during or after pregnancy, but no consistent patterns were observed in hormone levels.

Table 3.

Midpregnancy (≥18-<25 weeks) hormone concentrations (pg/mL) by participant characteristics among 139 New York University Children’s Health and Environment Study participants, 2016-2018

	Total	Pregnenolone	Progesterone	Allopregnanolone	Pregnanolone
	N (%)	GM (GSD)	GM (GSD)	GM (GSD)	GM (GSD)
Total	139 (100)	7149 (772)	106 922 (4475)	4010 (247)	9385 (1079)
Age (years)
≥18-≤25	31 (22)	7738 (1996)	112 755 (7881)	3638 (295)	8512 (2634)
≥26-≤34	78 (56)	7408 (1093)	106 744 (5893)	3955 (381)	9507 (1364)
≥35	30 (22)	5986 (1104)	101 805 (11 103)	4581 (476)	9972 (2257)
Prepregnancy body mass index
Normal (≥18.5-24.9)	52 (38)	6774 (1369)	120 946 (7291)	5824 (518)	12 379 (2420)
Overweight (≥25-29.9)	44 (32)	9356 (1685)	109 560 (6533)	3307 (440)	9692 (1910)
Obese (≥30)	42 (30)	5531 (905)	89 577 (8462)	3118 (196)	6487 (1303)
Fetal sex
Male	69 (50)	5966 (850)	114 147 (6625)	3895 (362)	11 620 (1698)
Female	70 (50)	8544 (1370)	100 347 (5986)	4124 (338)	7649 (1330)
Ethnicity
Non-Hispanic	49 (35)	8210 (1559)	125 339 (9390)	5293 (489)	10 725 (1673)
Hispanic	90 (65)	6635 (870)	98 125 (4713)	3441 (263)	8712 (1364)
Education
Less than college	91 (66)	6507 (817)	100 435 (5495)	3496 (268)	8770 (1284)
College or more	47 (34)	8560 (1763)	118 473 (7118)	5130 (487)	10 556 (1984)
Urinary cotinine concentrations (μg/g Cr)a
0-<0.31	105 (76)	7042 (882)	107 725 (5281)	4332 (255)	10 014 (1309)
≥0.31	34 (24)	7483 (1622)	104 501 (8461)	3165 (534)	7648 (1844)
Prenatal depressionb
No	119 (86)	7383 (898)	107 283 (4853)	4143 (229)	10 121 (1140)
Yes	20 (14)	5921 (1217)	104 886 (11 858)	3306 (908)	5870 (2583)
Postpartum depressionc
No	127 (91)	7153 (825)	107 401 (4803)	4036 (267)	9292 (1158)
Yes	12 (9)	7112 (2090)	102 167 (11 507)	3744 (568)	10 406 (2159)
Perinatal psychotropic medication used
None	132 (95)	7122 (808)	107 040 (4535)	3943(252)	9478 (1070)
Prenatal	3 (2)	7547 (2678)	115 405 (32 869)	3691 (751)	4679 (10 318)
Postnatal	4 (3)	7784 (2325)	97 493 (39 347)	7365 (1391)	11 477 (4558)

Abbreviations: Cr, creatinine; GM, geometric mean; GSD, geometric standard deviation.

^aCut-off values for distinguishing passive smokers from non-smokers as per Nishihama et al. (39).

^bAs assessed by the Patient Health Questionnaire-9 in each trimester and defined by scores ≥10 at any point in pregnancy.

^cAs assessed by the Postnatal Edinburgh Postnatal Depression Scale 4 months after delivery and defined by scores ≥10.

^dIncludes antidepressants, anxiolytic, and antipsychotic medications according to Anatomical Therapeutic Chemical codes.

Figure 1.

Geometric mean hormone concentrations (pg/mL) by gestational age in completed weeks across 139 Children’s Health and Environment Study participants, 2016-2018.

Among the 12 women (8.6% out of 139) that had EPDS scores ≥10 indicating PPD, women tended to be older than those with EPDS scores <10 (ie, no PPD) (Table 4). Distributions of race/ethnicity and education level were very similar across PPD strata. Women with PPD symptoms had greater material hardship and were more likely to be single and have experienced depressive symptoms during pregnancy compared with those without PPD. While prepregnancy BMI was similar across PPD strata, those with PPD symptoms were less likely to have gestational diabetes but more likely to have preeclampsia during pregnancy, although numbers were sparse.

Table 4.

Participant characteristics by postpartum depression (defined by Edinburgh Postnatal Depression Scale Scores ≥10) among 139 New York University Children’s Health and Environment Study participants, 2016-2018

	PPD (n = 12)	Non-PPD (n = 127)
	N (%)	N (%)
Age (years)
≥18-≤25	2 (17)	29 (23)
≥26-≤34	6 (50)	72 (57)
≥35	4 (33)	26 (20)
Ethnicity
Non-Hispanic	4 (33)	45 (35)
Hispanic	8 (67)	82 (65)
Education
Less than college	8 (67)	83 (66)
College or more	4 (33)	43 (34)
Material hardship (50th, 25th, 75th percentile)a	1 (0,3)	0 (0, 1)
Marital status
Married or cohabitating	8 (67)	104 (82)
Single	4 (33)	23 (18)
Parity
Primipara	6 (50)	53 (42)
Multipara	6 (50)	74 (58)
Fetal sex
Male	5 (42)	64 (50)
Female	7 (58)	63 (50)
Cotinine concentrations (μg/g Cr)b
0-<0.31	9 (75)	96 (76)
≥0.31	3 (25)	31 (24)
Antenatal depressionc
No	9 (75)	110 (87)
Yes	3 (25)	17 (13)
Prepregnancy body mass index
Normal (≥18.5-24.9)	3 (25)	49 (39)
Overweight (≥25-29.9)	5 (42)	39 (31)
Obese (≥30)	4 (33)	38 (30)
Gestational diabetes
No	11 (92)	101 (80)
Yes	1 (8)	26 (20)
Preeclampsia
No	9 (82)	120 (97)
Yes	2 (18)	4 (3)

Abbreviation: Cr, creatinine; PPD, postpartum depression.

^aAs assessed by the US Census Survey of Income and Program Participation as per Mayer SE, Jencks C. (32).

^bCut-off values for distinguishing passive smokers from nonsmokers as per Nishihama et al. (39).

^c As assessed by the Patient Health Questionnaire (PHQ)-9 in each trimester and defined by scores ≥10 at any point in pregnancy

Regression results showed that Σbisphenols, ΣHMW phthalate metabolites, ∑DnOP and ∑DiNP metabolites were associated with reduced progesterone in unadjusted models (Table 5). In addition, ΣLMW phthalate metabolites were associated with lower pregnenolone, allopregnanolone, and pregnanolone in crude models. However, most associations attenuated after adjusting for covariates, particularly BMI. Bisphenols and ΣLMW phthalate metabolites were not associated with any of the sex steroid hormones. ΣHMW phthalate metabolites did not have any statistically significant associations with hormones, but the estimate for progesterone remained suggestive (a log unit increase in ΣHMW phthalate metabolites was associated with 6.9% lower progesterone [95% CI –14.1, 1.0]). The inverse associations between ∑DnOP and ∑DiNP metabolites with progesterone were the only associations that persisted after adjustment, showing statistically significant reduced progesterone concentrations. Log-unit increases in ∑DnOP and ∑DiNP were associated with 8.1% (95% CI –15.2%, –0.4%) and 7.7% (95% CI –13.3%, –1.7%) lower progesterone, respectively. When adjusted for multiple comparisons, the estimate for ∑DnOP did not meet the adjusted P-value threshold (P = .040) but the estimate for ∑DiNP did (P = .012).

Table 5.

Associations between early and midpregnancy bisphenol and phthalate concentrations and midpregnancy sex steroid hormone levels presented as percent changes and 95% CI in relation to log-unit increases in chemical exposure levels among 139 Children’s Health and Environment Study participants, 2016-2018

	Pregnenolone (n = 137)	Progesterone (n = 134)	Allopregnanolone (n = 138)	Pregnanolone (n = 137)
	Δ% (95% CI)	Δ% (95% CI)	Δ% (95% CI)	Δ% (95% CI)
∑Bisphenols (nmol/mL)
Unadjusteda	–5.1 (–20.6, 13.5)	–7.8 (–13.6, –1.6)	–4.0 (–11.9, 4.7)	–7.8 (–19.7, 5.7)
Adjustedb	–2.4 (–19.1, 17.8)	–4.8 (–10.4, 1.1)	3.9 (–3.6, 12.0)	0.7 (–12.9, 16.4)
PA (ng/mL)
Unadjusteda	–8.8 (–21.8, 6.3)	–6.6 (–14.0, 1.3)	–3.3 (–9.3, 3.2)	–8.9 (–20.2, 4.0)
Adjustedb	–8.9 (–21.8, 6.2)	–5.4 (–13.0, 2.7)	–0.2 (–5.1, 5.0)	–6.0 (–18.6, 8.7)
∑LMW phthalates (nmol/mL)
Unadjusteda	–14.2 (–26.4, –0.1)	–7.9 (–16.0, 1.1)	–8.1 (–15.7, –0.1)	–14.5 (–26.2, –1.0)
Adjustedb	–11.7 (–25.0, 3.9)	–6.1 (–14.3, 3.0)	–3.4 (–10.6, 4.5)	–10.6 (–24.1, 5.3)
∑HMW phthalates (nmol/mL)
Unadjusteda	–4.2 (–17.5, 11.3)	–9.5 (–16.4, –2.1)	–6.3 (–13.0, 1.0)	–0.5 (–14.5, 18.0)
Adjustedb	–1.8 (–16.1, 14.8)	–6.9 (–14.1, 1.0)	–0.2 (–7.4, 7.6)	7.5 (–9.3, 27.4)
∑DEHP phthalates (nmol/mL)
Unadjusteda	–3.3 (–16.3, 11.7)	–6.9 (–13.6, 0.5)	–5.9 (–13.2, 2.2)	2.7 (–12.2, 20.2)
Adjustedb	–1.0 (–14.3, 14.5)	–4.3 (–11.2, 3.0)	–1.1 (–9.1, 7.5)	7.9 (–8.2, 26.9)
∑DnOP phthalates (nmol/mL)
Unadjusteda	0.3 (–14.8, 18.0)	–10.2 (–16.7, –3.2)	–3.5 (–10.9, 4.4)	–11.0 (–22.4, 2.0)
Adjustedb	1.5 (–14.2, 20.1)	–8.1 (–15.2, –0.4)	1.7 (–5.0, 8.9)	–6.1 (–17.8, 7.3)
∑DiNP phthalates (nmol/mL)
Unadjusteda	–4.1 (–16.3, 9.8)	–9.3 (–14.7, –3.5)	–5.2 (–11.7, 1.7)	–9.1 (–20.7, 4.0)
Adjustedb	–2.8 (–15.4, 11.6)	–7.7 (–13.3, –1.7)	–1.9 (–8.3, 5.0)	–5.9 (–17.3, 7.1)

Abbreviations: DEHP, di (2-ethylhexyl) phthalate; DiNP, diisononyl phthalate; DnOP, di-n-octyl phthalate; HMW, high molecular weight; LMW, low molecular weight; PA, phthalic acid.

^aAdjusts for creatinine only

^bAdjusts for creatinine, gestational age, maternal age, and prepregnancy body mass index.

When PPD symptoms represented by continuous EPDS scores were examined, there were no statistically significant associations with either bisphenols or phthalates (Table 6). However, ∑DnOP and ∑DiNP were the most suggestive. For a log-unit increase in ∑DnOP and ∑DiNP metabolites, 0.41 (95% CI –0.11, 0.93) and 0.30 (95% CI –0.06, 0.67) point increases in EPDS scores were predicted, respectively. The estimate for ∑DnOP metabolites was consistent with this trend when the outcome was dichotomous PPD: a log-unit increase in ∑DnOP was associated with increased odds of PPD (odds ratio [OR] 1.48; 95% CI 1.04, 2.11). Still, this estimate did not meet the adjusted P value threshold (P = .031). The estimate for ∑DiNP was null (OR 1.02; 95% CI 0.72, 1.45), as it was for the other bisphenol and phthalate groupings.

Table 6.

Associations between prenatal bisphenols and phthalate concentrations and Edinburgh Postnatal Depression Scale (EPDS) scores and postpartum depression (PPD) among 139 Children’s Health and Environment Study participants, 2016-2018

	EPDS scorea	PPDa,b
	β (95% CI)	OR (95% CI)
∑Bisphenols (nmol/mL)	0.17 (–0.32, 0.66)	0.98 (0.63, 1.53)
PA (ng/mL)	0.23 (–0.25, 0.70)	1.42 (0.90, 2.23)
∑LMW phthalates (nmol/mL)	0.32 (–0.27, 0.90)	1.36 (0.85, 2.20)
∑HMW phthalates (nmol/mL)	0.12 (–0.31, 0.55)	0.94 (0.68, 1.30)
∑DEHP phthalates (nmol/mL)	–0.01 (–0.48, 0.45)	0.76 (0.53, 1.08)
∑DnOP phthalates (nmol/mL)	0.41 (–0.11, 0.93)	1.48 (1.04, 2.11)
∑DiNP phthalates (nmol/mL)	0.30 (–0.06, 0.67)	1.02 (0.72, 1.45)

Abbreviations: DEHP, di (2-ethylhexyl) phthalate; DiNP, diisononyl phthalate; DnOP, di-n-octyl phthalate; HMW, high molecular weight; LMW, low molecular weight; OR, odds ratio; PA, phthalic acid.

^aAdjusts for creatinine, maternal age, maternal education, race/ethnicity, and pre-pregnancy body mass index.

^bDefined by EPDS scores ≥10.

Analyses of creatinine-corrected bisphenols and phthalates yielded similar results as the primary analyses with covariate adjustment for creatinine (Tables S2 and S3 (53)). ∑DnOP and ∑DiNP metabolites were still associated with lower progesterone and the estimate for ΣHMW phthalate metabolites that was suggestive in primary analyses was statistically significant (–8.7%, 95% CI –14.3%, –2.7%). ∑DnOP was still the only chemical exposure grouping found to be statistically significantly associated with PPD. In addition, excluding women who were prescribed antidepressant, anxiolytic, or antipsychotic medications during or after pregnancy (n = 7) did not measurably change results (Tables S4 and S5 (53)). Similarly, excluding women who were prescribed medications during pregnancy that contained synthetic progesterone or had the potential to interact with progesterone yielded results very similar to primary analyses (Tables S6 and S7 (53)). The inverse associations between ∑DnOP and progesterone persisted. Finally, we did not observe evidence of associations between midpregnancy hormone concentrations and EPDS scores (Table S8 (53)).

Discussion

In this prospective cohort study of pregnant women, prenatal urinary concentrations of ∑DnOP and ∑DiNP metabolites were associated with reduced progesterone concentrations in midpregnancy and ∑DnOP metabolites were associated with increased PPD symptoms assessed 4 months after delivery. However, prenatal bisphenols and other phthalate metabolites were not associated with pregnenolone, allopregnanolone, or pregnanolone concentrations or PPD symptoms. Results were robust to covariate adjustment and remained consistent after excluding women on psychotropic or progesterone-disrupting medications and implementing an alternate method for urinary creatinine adjustment.

This study is the first to examine environmental chemical exposures in relation to PPD with an explicit focus on endocrine mechanisms of action. The robust and specific finding that DnOP metabolites were associated with both lower prenatal progesterone and greater PPD symptoms is consistent with the hypothesis that EDCs could drive hormonal shifts which could affect PPD. However, our analyses of midpregnancy hormone concentrations in relation to EPDS scores did not show significant associations, which suggests that other mechanisms may be at play. Still, our observation that DnOP was associated with PPD suggests that because phthalate exposure can be reduced through dietary and behavioral interventions such as avoiding food packaging, certain cosmetics, and polycarbonate plastic products (54-57), these findings provide preliminary evidence identifying prenatal exposure to phthalates as a potentially modifiable risk factor for PPD. In light of the fact that other established risk factors for PPD include genetic predisposition and socioeconomic status and are less readily alterable (58), these finding could inform a plausible avenue to prevent PPD morbidity. More studies are needed to replicate these findings.

While we hypothesized that bisphenols and/or phthalates would be associated with reduced allopregnanolone, this was not apparent after adjusting for covariates. Instead, the most robust finding was that phthalates and particularly DnOP and DiNP metabolites were associated with lower progesterone and these were the same exposures to be correlated with increased PPD symptoms. Taken together, our findings motivate the hypothesis that prenatal exposure to these phthalates leads to increased PPD symptoms and that this may occur at least in part through reductions in progesterone concentrations during pregnancy. However, we did not observe empirical evidence of this when we examined hormone concentrations in relation to EPDS scores. Still, progesterone is known to regulate the production, release, and transport of neurotransmitters (59) and has been shown to upregulate brain-derived neurotrophic factor (60), which is decreased in depressive states (61). Other potential mechanisms linking progesterone with depression are also viable including hyperactivity of the hypothothalamic–pituitary–adrenal axis, impaired stress response circuitry, and hormonally mediated structural plasticity. Withdrawals in progesterone have been found to induce anxiety and depression in experimental animal studies (62, 63) and treatment with progesterone found to reduce anxiety (64), which may be explained by increased amygdala reactivity (65).

This study benefited from several strengths. The data required to test the hypothesis that prenatal exposure to EDCs impact PPD through sex steroid hormone shifts are extensive. By embedding this study within CHES, we were able to leverage early pregnancy recruitment (median gestational age at enrollment = 9 weeks), longitudinal biospecimen collection, a sociodemographically diverse cohort, and high postnatal retention and follow-up (>80%) (30). In addition, this study was conducted among a general population of care-seeking pregnant women. Several other studies on the endocrine determinants of PPD have been conducted among women with previous mood disorder diagnoses (66-68). While this provides statistical power due to a substantial proportion of women developing PPD, the findings may not be generalizable to all pregnant women. Another strength was our ability to assess the potential impact of women using psychotropic medications during or after pregnancy. This is important because antidepressant use can impact both PPD symptoms (ie, assessed EPDS scores) as well as endogenous hormone concentrations (4). However, we did not observe any difference after excluding women using these medications. Finally, embedding this investigation within an already existing pregnancy cohort (ie, NYU CHES) provided a wide array of covariates. We were able to evaluate the influence of adjusting for several covariates and ultimately reduced the adjustment set to only include meaningful confounders. This improved both the precision and power of our estimates.

Limitations of this study included small sample size, the lack of psychiatric evaluation or clinical diagnosis, and nonpersistent exposures assessed in spot urine samples. First, our small sample size limited our statistical power and may have been the reason that DnOP was the only phthalate to be significantly associated with both hormones and PPD. We note that our findings should be interpreted in the context of the small sample size and with caution due to our limited power. Larger studies are needed to both confirm these findings and identify whether other phthalates are associated with these endpoints. Second, PPD was based on the EPDS which is a screening tool rather than diagnostic test. However, the use of the EPDS is beneficial because it captures subclinical symptoms and undiagnosed PPD. Also, as the most widely used validated scale for PPD (36) it facilitates direct comparisons across studies. Given that PPD is under-recognized and undertreated in clinical practice, this further underscores the importance of using a screening tool rather than clinical diagnoses (69). In addition, the EPDS was administered at 4 months postpartum, and while there are no established guidelines for the ideal timing of PPD screening, some studies have suggested that most postpartum mental disorders manifest within 1 to 3 months after delivery (70, 71). Therefore, it is possible that our assessment could have missed any resolved depression. However, this later assessment likely rightly excludes the common and transient “postpartum blues” and captures the majority of incident major depressive disorders (1, 72). Furthermore, our PPD assessment was defined by EPDS scores ≥10. While there are a range of acceptable thresholds (37), this is the most sensitive and includes PPD with a range of severity (35), which enhanced our ability to detect potential signals with this novel exposure. Lastly, spot urine samples were used for exposure assessment and bisphenols and phthalates are nonpersistent and rapidly metabolized. Indeed, several studies have documented large within-person variability for bisphenols and phthalates and poor reliability over time (73, 74). However, the longitudinal design of 2 measures over pregnancy and the statistical approach implemented mitigated this issue.

The results of this study underscore the potential for EDCs to impact hormonally mediated health outcomes and expand the scope of conditions typically considered susceptible to EDCs (8). While PPD remains underdiagnosed, understudied, and with few known modifiable risk factors (75), work in this avenue seeks to understand whether exposure to synthetic environmental chemicals during pregnancy may play a role. This study highlights the biologic plausibility of phthalate-induced sex steroid hormone disruption on the pathway for PPD pathogenesis and the potential for pregnancy as a critical window of exposure to EDCs.

Abbreviations

Abbreviations
 
• BMI

  body mass index

 
• BPA

  bisphenol A

 
• CHES

  Children’s Health and Environment Study

 
• Cr

  creatinine

 
• DiNP

  diisononyl phthalate

 
• DnOP

  di-n-octyl phthalate

 
• EDC

  endocrine disrupting chemical

 
• EPDS

  Edinburgh Postnatal Depression Scale

 
• GABA

  gamma-aminobutyric acid

 
• HMW

  high molecular weight

 
• HPLC

  high-performance liquid chromatography

 
• IS

  internal standard

 
• LOD

  limit of detection

 
• LMW

  low molecular weight

 
• MS/MS

  tandem mass spectrometry

 
• NYU

  New York University

 
• PHQ

  Patient Health Questionnaire

 
• PPD

  postpartum depression

Acknowledgments

Financial Support: This work was funded by the NYU Center for the Investigation of Environmental Hazards pilot project program, the National Institutes of Health Office of the Director grant UG3/UH3OD023305 and the National Institutes of Environmental Health Sciences grant P30ES000260.

Additional Information

Disclosures: The authors have nothing to disclose.

Data Availability

Restrictions apply to the availability of some or all data generated or analyzed during this study to preserve patient confidentiality or because they were used under license. The corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided.