Study population
The MEC is a large prospective cohort that included 96,810 men and 118,441 women aged 45–75 years from five different racial/ethnic groups (African Americans, Latinos, Native Hawaiians, Japanese Americans, and whites) living in Hawaii and California (primarily from Los Angeles County) at enrollment between 1993 and 1996 [16]. Participants completed a baseline questionnaire which assessed demographics, lifestyle, diet, and anthropometrics, and for women, menstrual and reproductive histories and hormone therapy use. Participants were followed prospectively for diagnosis of incident breast cancer (invasive and in situ) through routine yearly linkage with the California and Hawaii statewide cancer registries and for vital status through yearly linkages to the National Death Index and state death certificate files which was through 2014 for this nested case-control study. Stage at diagnosis and estrogen/progesterone receptor (ER/PR) status were obtained from cancer registries.
In 2001–2006, a prospective biorepository was established by collecting pre-diagnostic urine and blood specimens from 67,594 MEC cohort members [17]. A short questionnaire was administered at biospecimen collection which assessed weight, use of hormone therapy and medications, and other factors. For this nested case-control study, cases (n = 1032) were diagnosed with incident breast cancer from 2001 through 2014 after urine collection; 22% were in situ and 78% were invasive cancers. The mean time between urine collection and breast cancer diagnosis was 5.5 years (standard deviation (SD) = 3.3). For each case, we selected one control, who was alive and free of breast cancer at the age of breast cancer diagnosis, and individually matched controls to cases on area (Hawaii or California), birth year (± 1 year), race and ethnicity (white, Native Hawaiian, African American, Latino, Japanese American), urine type (first morning from California, overnight and first morning from Hawaii), date (± 1 year) of urine collection, hours of fasting (8–10, > 10 h), and time of blood draw (± 2 h). As described previously, the overnight urine collection started between 5:00 pm and 9:00 pm (depending on participant) and included all urine passed during the night and the first morning urine sample covering a period of 12 h [18]. Controls were sampled from the representative pool of subjects with existing data on obesity-related and inflammation biomarkers and genotype array data. In total, 1030 controls were identified and each control was individually matched to one case except that one white and one Native Hawaiian control were each matched to the two cases of the same race/ethnic group. There were 30 controls (19 whites, 7 Japanese Americans, 2 Native Hawaiians, 2 Latinos, 0 African Americans) and 44 cases (14 whites, 16 Japanese Americans, 6 Native Hawaiians, 4 Latinos, 4 African Americans) who had other cancers with a mean of 6.3 years (SD 5.2) and 8.4 years (SD 5.8), before the donation of urine collection, respectively. Results were unchanged without these individuals (data not shown) and they were included in the analyses.
Laboratory measurement of urinary phthalate metabolites
Phthalate metabolites (MEP, MMP, MBP, MiBP, MBzP, MCHP, MEHP, MEHHP, MEOHP, MECPP, MCMHP) and phthalic acid measurements were conducted at the University of Hawaii Cancer Center Analytical Biochemistry Shared Resource. Dr. Adrian Franke supervised the day-to-day activities and quality assurance and quality control of phthalate measurements using state-of-the-art sensitive isotope-dilution orbitrap-based high-resolution accurate-mass liquid chromatography mass spectrometry (LCMS) assay after enzymatic hydrolysis and liquid-liquid extraction [19, 20]. In brief, 0.1 mL urine was mixed with 0.01 mL of a mixture of isotopically labeled analyte that was used as internal standards followed by incubation with a glucuronidase/sulfatase mixture at 37 °C for 90 min, extraction with methyl tertiary butyl ether, and LCMS analysis. Our lower limits of quantitation were 0.5 ng/mL for each analyte. Individual phthalate all share phthalic acid as a common metabolite as all the phthalate diesters can be metabolized to phthalic acid [21, 22]. One metabolite (mono 2-carboxy-hexyl phthalate, MCMHP) was not measured reliably and was excluded from data analysis. Personnel were blinded to case-control status, and matched pairs of cases and controls were assayed in the same batch. Replicate samples of pooled urines (5%) were included in each of 37 batches for quality control measures and coefficients of variation (CV) were calculated. The CV% (SD/mean concentration × 100) within-batch was 26.7% for phthalic acid and was < 25% for eight of the ten metabolites with a median of 22.6% for the individual metabolites (range was 5.9% (MCHP) to 35.0% (MMP)). The larger CVs likely reflect several samples close to the lower limit of detection (LLOD). The mean CV of the non-blinded pool samples was 11.7% (SD 6.9%). All analytes were adjusted to urinary creatinine [19, 20] and are shown as micrograms per gram (μg/g) of creatinine. Analytes below the LLOD were assigned a value half of the LLOD. Eight of the ten metabolites and phthalic acid were detected at levels above the LLOD in over 92% of women; this was lower for MMP (80%) and MCHP (89%) (Supplementary Table 1). We also present the geometric mean concentration for each phthalate metabolite collected in the overnight urine samples (i.e., all were from Hawaii) and first morning urine samples (i.e., all but 16 were from Los Angeles County) (Supplementary Table 1).
Statistical analysis
We conducted conditional logistic regression, with the matched sets as strata (1028 pairs and 2 triplets), and modeled phthalate variables as tertiles using selected cutoff points based on the distribution among all controls. Odds ratios (ORtertiles) and 95% confidence intervals (CIs) were the primary statistics of interest, and inference was based on the Wald test. We found no evidence of a nonlinear relationship (on the log odds scale) between phthalates and risk using restricted cubic splines (data not shown). Therefore, log-transformed phthalate variables were used as trend variables to test for dose-response relationships. Models were adjusted for potential confounders that were not matching factors (e.g., established breast cancer risk factors) via indicator variables for tertiles of propensity scores for exposure to phthalates [23, 24] in order to maximize power. Specifically, an ordinal logistic regression for tertiles of each phthalate variable was performed using the following independent variables: age at urine collection, education, number of children, age at menarche, menopausal status, body mass index (BMI) at urine collection, neighborhood socioeconomic status (nSES) [25] at urine collection, smoking, alcohol intake, and Mediterranean diet energy adjusted total score [26]. The nSES index is a composite measure created by principal component analysis (PCA) of US Census data that incorporated census block group data on education, occupation, unemployment, household income, poverty, rent, and house values [25]. PCAs were conducted separately for California and Hawaii. Results across states were similar for Eigen vectors and variance explained with a single component identified. This nSES measure was categorized into quintiles based on the nSES distribution of Los Angeles County and Hawaii block groups for California and Hawaii MEC participants, respectively. The propensity for exposure was determined for each individual as the weighted average = 1 × ρ1 + 2 × ρ2 + 2 × ρ3, where ρi is the model-based probability of exposure to tertile i. Heterogeneity of the associations by race/ethnicity was assessed by a global test of the interaction terms between race and the phthalate trend variable. We repeated subgroup analyses for hormone receptor-positive (HR+, ER+, or PR+) and hormone receptor-negative (HR−, ER−, and PR−) cancer, as well as BMI and use of hormone therapy at urine collection. In addition, waist-hip ratio (WHR) was obtained in a follow-up questionnaire and was available for 333 cases prior to diagnosis and 946 control women; we explored risk association results by WHR. We also conducted a sensitivity analysis by restricting to invasive cases only (n = 798) and by lag time between time of urine collection and breast cancer diagnosis (≤ 5 years versus > 5 years), as well as excluding 187 cases that were diagnosed within 2 years of urine collection to minimize the potential effect of pre-diagnostic breast cancer on phthalate levels. Associations with P < 0.05 were considered statistically significant and with 0.05 ≤ P < 0.10 were considered suggestive. The correlations among phthalate metabolites and phthalic acid were examined using Spearman’s rho.
The short-branched phthalates, DMP and DEP, are excreted in urine as the unconjugated monoesters, MMP and MEP, respectively. The longer-branched phthalates such as DEHP are hydrolyzed first to MEHP and subsequently metabolized to MEHHP, MEOHP, MECPP, and other oxidative metabolites [7, 27, 28] (Fig. 1). Oxidation of MEHP to other secondary metabolites effectively decreases the internal body burden of MEHP to presumably less toxic secondary metabolites [29, 30]. Thus, subjects with higher MEHP% and higher ratios of MEHP to major secondary DEHP metabolites (MEHHP, MEOHP, MECPP) may be at higher risk (MEHP% was calculated by converting DEHP metabolites into nanomoles (nmol) using their respective molecular weight and dividing the molar mass of MEHP by the mass of the sum of all four metabolites and then multiplying by 100). For all women, and separately in African Americans and Latinos combined, Native Hawaiians, whites, and Japanese Americans, we examined risk in relation to the 10 metabolites and phthalic acid, MEHP%, the ratios of MEHP to secondary DEHP metabolites, and summary variables, including the sum of all major DEHP metabolites (∑DEHP), low molecular weight phthalates (∑LMWP: MBP, MiBP, MEP, MMP), high molecular weight phthalates (∑HMWP: MBzP, MCHP, ∑DEHP), and total phthalate represented by the sum of all 10 phthalate metabolites and phthalic acid (∑LMHMPA) and the sum based on molar ratios of phthalate metabolites without phthalic acid (∑LMHMmolar). We divided the concentration of each metabolite by its molecular weight to obtain the molar equivalent (micromoles/liter, μmol/L) and then summed the concentrations to get total μm/L of metabolites per creatinine unit.