The potential contribution of environmental factors to obesity is of increasing interest. “Obesogens” describes environmental chemicals hypothesized to promote obesity due to altered regulation of adipogenesis and lipid metabolism. Phthalates are endocrine disrupting chemicals present in many consumer products (e.g. cosmetics, food packaging, medications) and are ubiquitous in the environment. Nearly all U.S. residents have detectable concentrations of phthalate metabolites in their urine, though concentrations vary widely [1]. Limited in vitro data suggest that certain phthalates may alter pathways that promote adipogenesis [2, 3], and thus could impact development of obesity.
Scant research evaluating associations between phthalate exposure and body weight report inconsistent findings. Some cross-sectional studies report positive associations between certain phthalate metabolite concentrations and body mass index (BMI) and obesity among adult women. Specifically, one cross-sectional study using data from the 2007–2010 National Health and Nutrition Examination Survey (NHANES) reported increased prevalence of obesity associated with higher concentrations of mono-carboxyoctyl phthalate (MCOP), mono (2-ethyl-5-carboxypentyl) phthalate (MECPP), mono (2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), and the sum of di-ethylhexyl phthalate metabolites (ΣDEHP) [4]. A separate cross-sectional study using 1999–2004 NHANES data observed increased obesity prevalence associated with mono (2-ethylhexyl) phthalate (MEHP) and mono-butyl phthalate (MBP), yet found borderline significant inverse associations between MECPP, MEHHP, mono (2-ethyl-5-oxohexyl) phthalate (MEOHP), ΣDEHP and BMI category [5]. Also, other studies, including one using 1999–2002 NHANES data [6] and another within the Nurses’ Health Study (NHS) and NHS2 cohorts [7], reported inverse cross-sectional associations with MEHP [6], MBP [6, 7], mono-benzyl phthalate (MBzP) [7], and mono-isobutyl phthalate (MiBP) [7].
Phthalates are rapidly metabolized in the body and excreted in urine, and urinary phthalate metabolite concentrations reflect recent exposures [8]. Therefore, the observed cross-sectional associations may reflect confounding via exposure from sources that are themselves associated with obesity, as opposed to causal associations.
One prior prospective analysis, among 977 women aged 32–79 from the Nurses’ Health Study (NHS) and NHS2 [7], reported positive associations with weight gain for MBzP (+ 0.42 kg/year for 4th vs 1st quartiles) and the sum of butyl phthalate metabolites (MBP and MiBP; + 0.34 kg/year for 4th vs 1st quartiles) over a 10 year follow-up period. Weight change was not associated with concentrations of ΣDEHP or mono-ethyl phthalate (MEP) [7].
Whether phthalates affect weight gain remains an unanswered, yet critically important, question. We prospectively evaluated associations between 13 phthalate metabolites (or their sums) and weight change among 997 postmenopausal women enrolled in the Women’s Health Initiative (WHI).
Subjects and methods
Study population
We included 1257 postmenopausal women selected for a nested case-control study of phthalates and breast cancer risk within the WHI. The design of the WHI has been reported previously [9]. Briefly, from October 1, 1993 to December 21, 1998 a total of 161,808 women aged 50–79 years were enrolled in the WHI. WHI participants who were enrolled at three bone density substudy sites (Birmingham, AL; Pittsburgh, PA; Tucson/Phoenix, AZ) provided first morning void urine samples at baseline. A nested case-control study of breast cancer within the WHI quantified urinary concentrations of phthalate metabolite on 419 incident breast cancer cases and 838 matched controls selected from among these bone density substudy participants. Breast cancer cases were selected as all cases of invasive breast carcinoma that occurred among these participants after the year 3 follow-up clinic visit through 2013; controls were matched on enrollment date, length of follow-up, age at enrollment, and WHI study arm with a 1:2 ratio. This analysis includes 997 participants (337 cases, 660 controls) with complete data available (Fig. 1). The longitudinal analysis included only participants selected as controls (N = 660) in the parent study, given that weight gain is common following breast cancer treatment [10].
All participants provided written informed consent upon enrollment into the WHI. The WHI was approved by institutional review boards (IRB) at each clinical center. Additionally IRB approval for the present study was obtained from the University of Massachusetts Amherst. The involvement of the Centers for Disease Control and Prevention (CDC) laboratory in the analysis of samples did not constitute engagement in human subjects research.
Quantification of urinary phthalate metabolites
WHI followed a standard collection, processing, and storage protocol at the three clinical centers that collected urine samples. First morning void urine samples were collected at home and processed within 30 min after participants arrived at the clinic. WHI recommended, but did not require, the use of phthalate-free polypropylene urine collection containers; one site used the recommended containers while the composition of the containers used at the other two clinical centers is unknown. However, all sites used polypropylene centrifuge tubes and cryovials for long-term storage. Additionally, we measured concentrations of metabolites as opposed to the parent phthalate, which should reflect endogenous exposure as opposed to contamination. Urine samples were centrifuged for 5 min at 1330×g and 1.8 mL aliquots were frozen and shipped, packed in dry ice, via overnight FedEx to McKesson Bioservices where they were stored at − 70 °C.
Thirteen phthalate metabolites were measured in baseline urine samples at the CDC: MEP, MBP, mono-hydroxybutyl phthalate (MHBP), MiBP, mono-hydroxyisobutyl phthalate (MHiBP), MBzP, mono (3-carboxypropyl) phthalate (MCPP), MEHP, MEHHP, MEOHP, MECPP, MCOP, and mono-carboxynonyl phthalate (MCNP). Concentrations of phthalate metabolites were quantified after enzymatic hydrolysis of the conjugated metabolites followed by on-line solid phase extraction coupled to high performance liquid chromatography-electrospray ionization-isotope dilution tandem mass spectrometry. Complete details of the analytical method are published online at https://wwwn.cdc.gov/nchs/data/nhanes/2013-2014/labmethods/PHTHTE_H_MET_Phthalates.pdf. The limits of detection (LODs) were in the low ng/mL range. Study samples were randomly distributed through the analytical batches, with cases and matched controls analyzed together. A blinded 10% quality control sample was included, and estimated CVs were as follows: MBP 5.4%, MBzP 6.1%, MCNP 4.7%, MCOP 6.3%, MCPP 5.8%, MECPP 4.3%, MEHHP 5.4%, MEHP 19.5%, MEOHP 6.0%, MEP 3.1%, MHBP 9.0%, MHiBP 21.9%, MiBP 10.3%; the higher average CVs for MEHP and MHiBP reflect small differences in absolute levels of replicates having very low concentrations. All laboratory staff were masked to the identity, disease status, and demographic and risk factor characteristics of the samples. Creatinine was also measured by using an enzymatic assay at CDC on a Roche Modular P Chemistry Analyzer (Indianapolis, IN). The LOD for creatinine was 10 mg/L and the CV of the blinded quality control sample was 2.5%.
We analyzed concentrations of each phthalate metabolite individually. For phthalates with multiple measured metabolites, we also grouped the data by parent phthalate by dividing each metabolite of a single parent by its molecular weight and then summing across metabolites [11, 12]. For example, we calculated the molar sum of DEHP metabolites (ΣDEHP) by dividing each metabolite concentration by its molar mass and then summing the individual concentrations (μmol/L): [MEHHP × (1/294.35)] + MEHP × (1/278.34)] + [MECPP × (1/308.33)] + [MEOHP × (1/292.33)]. The sum of dibutyl phthalate metabolites (ΣDBP) was calculated as the molar sum of MBP and MHBP, and the sum of di-isobutyl phthalate metabolites (ΣDiBP) was calculated as the molar sum of MiBP and MHiBP.
Measurement of weight and BMI calculation
Height and weight were measured at the baseline, year 3, and year 6 clinic visits and used to calculate BMI as weight (kg)/height2 (m2) grouped as: underweight/normal weight (< 25.0 kg/m2), overweight (25.0– < 30.0 kg/m2), and obese (≥30.0 kg/m2).
Assessment of covariates
Extensive data on demographic, reproductive, medical history, and behavioral characteristics were collected in the WHI using self-administrated questionnaires at baseline. We considered the following variables as covariates: age (continuous), race/ethnicity (Caucasian, African American, Hispanic/Latino, other), education level (<high school, high school/some college, college degree and higher), income (<$10,000, $10,000-$19,999, $20,000-$34,999, $35,000-$49,999, ≥ $50000), health insurance (no insurance, military insurance, Medicare, Medicaid, private insurance), smoking status (never smoker, past smoker, current smoker), alcohol use (non-drinker, past drinkers, current drinkers), Healthy Eating Index-2005 (HEI-2005, [13]) score (continuous), dietary energy intake (kcal per day; continuous), total recreational physical activity (categorized in quartiles of Metabolic Equivalent values per week (METs/wk.; < 1.25 METs/wk., [1.25- < 6.38 METs/wk., 6.38- < 16.5 METs/wk., ≥16.5 METs/wk), oral contraceptive use (ever, never), any hormone therapy use (never, past, current), ever had diabetes (no, yes), ever had cardiovascular disease (no, yes), hypertension (never hypertensive, untreated hypertensive, treated hypertensive), and dyslipidemia (no, yes).
Statistical analyses
Phthalate biomarker concentrations were natural log transformed to improve normality. Baseline characteristics were summarized according to the BMI categories and differences assessed using analysis of variance (ANOVA) or chi square tests, as appropriate. Geometric means were calculated for each creatinine-standardized phthalate biomarker (i.e. individual metabolite or sum of metabolites of a common parent phthalate) with stratification on baseline BMI group, and differences across groups were assessed with ANOVA.
In cross-sectional analyses we included both cases and controls, given that cases were all diagnosed following the year 3 clinic visit and thus were considered “healthy” at baseline. We categorized phthalate metabolite concentrations into quartiles using the distribution among the controls. Linear regression and multinomial logistic regression analyses were used to model the relationship of each individual phthalate biomarker and baseline weight and BMI category, respectively. All models were adjusted for age and urinary creatinine concentration. We built the regression models by 1) fitting univariable linear and multinominal logistic regression models for each variable with weight and BMI, respectively, 2) including all variables with p < 0.25 in the univariable model in a preliminary multivariable model along with the phthalate biomarker, and 3) evaluating the significance of each covariate using backward selection and retaining all covariates with a p value < 0.10 or of known biological importance. A common set of covariates was included in the multivariable models to facilitate comparisons across phthalate biomarkers. Trends in the weight β coefficient and the odds ratio (OR) of overweight and obesity with increasing categories of phthalate biomarker were evaluated by testing the significance of a continuous variable including the median concentration of each biomarker quartile in the regression model. We included 997 participants with complete data on covariates, exposure, and outcomes in our analysis.
We modeled the prospective weight change rate over 3 and 6 years by the quartiles of urinary phthalate biomarker concentrations using mixed-effect models with: a random coefficient, a fixed effect for weight and year of follow up, and including product terms between phthalate biomarkers and year of follow up (i.e. year 3 and year 6). A parsimonious multivariable model was built using the process described above. Analyses were repeated with stratification on baseline BMI to evaluate possible effect modification, and we plotted predicted weight change over time by BMI category for models including an interaction with BMI and a model without this term. We obtained p-values for linear trends by including an interaction term between each year of follow up and the median concentration of each biomarker quartile in the mixed-models as a continuous variable. We considered a P-value < 0.05 as statistically significant. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, North Carolina) and Stata version 15.0 (Stata Corp, College Station, TX).