Participants from this analysis are a subset of men from the Environment and Reproductive Health (EARTH) study, a larger cohort assessing the impact of environmental agents on reproductive health. Participation and recruitment have been described elsewhere (Meeker et al., 2006). Briefly, men (18–54 years of age) attending the Massachusetts General Hospital fertility clinic between 2005 and 2015 were eligible. Participants originated from couples whose infertility diagnosis was either male factor, female factor, or a combination of both. Prior vasectomy or hormone supplementation were the only exclusion criteria. Informed consent was signed by each participant and Institutional Review Board approval was received by all institutions.

Semen collection and analysis have been previously described (Meeker et al., 2006; Lewis et al., 2017). Briefly, men abstained from ejaculation for 48 hours prior to sample collection into plastic specimen cup. Men provided up to five samples depending on the number of fertility treatments, additional fertility evaluation, or a combination of both. An andrologist quantified sample volume (mL) with a graduated pipet. Sperm concentration (mil/mL) and motility (% motile) was determined using a computer-aided semen analyzer (CASA, version 10 HTM-IVOS; Hamilton Thorne Research, Beverly, MA). Samples (5 μL) were collected on a disposable Leja Slide (Spectrum Technologies, CA, USA) and placed into a pre-warmed (37°C) counting chamber (Sefi-Medical Instruments, Haifa, Israel) before assessing concentration and motility. Among each sample, at least 200 sperm cells were analyzed from four different fields. Progressive motility was graded in accordance to the WHO’s assessment criteria of active movement (linearly or in a large circle), regardless of velocity (WHO, 2010). The product of sperm concentration and sample volume determined sperm count (mil/ejaculate) while progressive motility count (mil/ejaculate) was calculated by multiplying progressive motility and total sperm count. Fresh semen samples were allowed to dry on two prepared slides and prepared for morphology (% normal) assessment with a microscope using an oil-immersion 100× objective (Nikon, Tokyo, Japan). A minimum of 200 cells per slide were analyzed for each specimen. Classification of normal or subnormal morphology was determined using strict Kruger scoring criteria (Kruger et al., 1988). Quality assurance and control procedures in the laboratory were conducted for sperm morphology smears weekly, as well as quarterly and biannual evaluations for technicians.

Urine samples (up to five cycles) were collected in sterile polypropylene cups on the day of oocyte retrieval for each cycle per participant. Prior to being frozen (-80°) and stored, specific gravity (SG) was measured using a handheld refractometer (National Instrument Company, Inc., Austin, TX). For metabolite analysis, samples were shipped overnight on dry ice to Dr. Stapleton’s lab at Duke University (Durham, NC).

Analytic methods for metabolites: BCIPP, BDCIPP, DPHP, ip-PPP, and tb-PPP have been previously described (Butt et al., 2014). Briefly, 5 ml aliquots were thawed and transferred to test tubes and spiked with internal standards (d₁₀-BDCIPP = 80 ng, d₁₀-DPHP = 60 ng) before being acidified (pH <6.5) with formic acid and diluted with 1:1 with water. Solid phase extraction (SPE) was used to concentrate and clean samples before drying via nitrogen stream and spiked with the recovery standard (¹³C₂-DPHP = 81.5 ng). Extracts were analyzed using negative electrospray ionization liquid chromatography tandem mass spectrometry (LC-MS/MS) detailed previously (Butt et al., 2014). Optimal parameters under multiple reaction conditions were used to acquire data. The internal standard used for BCIPP and BDCIPP was d₁₀-BDCIPP, while quantification of DPHP, ip-DPHP, and tb-DPHP was performed using d₁₀-DPHP.

Quality assurance and control procedures for LC-MS/MS have been described previously (Carignan et al., 2017). Briefly, samples were processed in multiple batches including five blanks per batch (5 ml Milli-Q water); each batch providing a distinct method detection limit (MDL). MDLs were designated as three times the standard deviation of laboratory blanks and ranged from: 0.07–0.17 pg/ml for BCIPP, 0.02–0.11 pg/ml for BDCIPP, 0.09–0.18 pg/ml for DPHP, 0.06–0.12 pg/ml for ip-PPP, and 0.04–0.15 pg/ml for tb-PPP. Urine samples from previous studies were pooled to establish a standard reference material (SRM) and routinely analyzed. Duplicates of two-subsamples were analyzed to evaluate precision.

Descriptive statistics for PFR metabolites, semen parameters, and demographic factors were calculated. Values below MDL for metabolites were imputed as MDL/√2. Metabolites were presented as wet-weight and adjusted for SG as: C_SG = C*[(SG_M-1)/(SG_i -1)], where C_SG = SG-adjusted urinary metabolite concentration, C = urinary metabolite concentration, SG_M = mean SG for the population, and SG_i = SG for an individual sample (Boeniger et al., 1993). We evaluated bivariate associations among PFR metabolites, semen parameters, and demographic factors using Spearman correlation coefficients, Wilcoxon rank-sum tests, and Kruskal-Wallis tests as appropriate. A sum variable (ΣPFR) for BDCIPP, DPHP, and ip-PPP was created by combining all three metabolites per sample. Intraclass correlation coefficients (ICC) and 95% confidence intervals for metabolites (wet-weight and SG corrected) and semen parameters were calculated to assess variability between samples of each participant. All metabolites and semen parameters presenting as right-skewed were transformed by natural logarithm for further statistical modeling. PFR metabolites were evaluated as continuous variables and quartiles except for tb-PPP with low detection rate (11.34 %) was modeled as detect/non-detect. Sperm parameters were evaluated both continuously and dichotomized using WHO reference level for sperm: count (< 39mil/ejaculate), concentration (< 15 mil/mL), motility (< 40%), progressive motility (<32%), and morphology (< 4% normal) (WHO, 2010). Initially, crude associations were calculated among PFR metabolites and semen parameters (Supplemental Table 1). Bivariate tests for possible covariates: age, BMI, abstinence period, race, smoking status, education, and season of collection along with priori knowledge were used to select covariates for modeling (Supplemental Table 2). Although season of sample collection was associated with PFR metabolites, it was not associated with semen parameters and not included in final models. Multivariable regression models, adjusted for SG, age, BMI, and abstinence period, were used to test associations using only the first urine and semen sample for all participants. Multivariable generalized mixed models using continuous, dichotomous, and quartiles for PFRs were used to assess associations with repeated exposures and/or semen parameters. To test for trends, quartiles of each metabolite were treated as a continuous variable. We conducted a sensitivity analysis excluding SG measurements below 1.01 and above 1.03 for multivariable models (Supplemental Table 3) to examine any effect of extreme urine concentrations. Missing data were excluded from models. All statistical analyses were carried out using SAS 9.4 (SAS Institute Inc., Cary, NC).

Demographic characteristics of our subsample from the EARTH study are displayed in Table 1. Demographics from this sample are similar to previous studies in similar cohorts (Meeker et al., 2006) as well as national trends of men undergoing IVF (Hotaling et al., 2012) regarding age (mean=36.66 ± 5.07), BMI (27.18 ± 4.03), race or ethnicity (89% white), and education (80% college graduates).

The distribution of semen parameters among our sample is presented in Table 2. All parameters met WHO guidelines for normal sperm among more than half of participants (WHO, 2010). Total sperm count and concentration exceeded the guideline for normal sperm (43.42 and 15.9, respectively) in 90% of samples. Conversely, less than half of participants had above average motility (45%) while 75% exceeded the guideline for morphology (4%). Participants provided semen (n=269) and urine (n=355) samples from up to five clinic visits, with the majority of men providing one to three. Repeated measures of semen parameters had moderately strong intraclass correlations (0.51–0.58), except motility (ICC=0.79) and progressive motility (ICC=0.71) which had stronger ICCs (Supplemental Table 4).

Distributions of PFR metabolites, as both wet-weight and SG-corrected are displayed in Table 3. Metabolites BDCIPP, DPHP, and ip-PPP were detected in a high proportion of urine samples (85%, 86%, and 66% respectively). We identified weak (r < 0.30) yet significant (p < 0.01) correlations among BDCIPP, DPHP, and ip-PPP, and moderate (r=0.43) correlations between DPHP and tb-DPHP (p=0.01) (Data not shown). Similarly, temporal stability between metabolite measurements (Table 4) were weak-to-moderate (ICC <0.35) and decreased further when excluding non-detects and adjusting for SG. Concentrations varied by season of collection, particularly for BDCIPP where samples collected in the summer had the highest concentrations (p<0.0001) (Supplemental Table 2).

When modeling PFR metabolites and semen parameters as continuous variables, there were no significant effect estimates from repeated measure models (Table 5). Whereas when semen parameters were dichotomized, elevated BDCIPP was associated with a decreased odds of low sperm count (OR=0.79, 95% CI=0.64, 0.99; p=0.04). Results were similar in a sensitivity analysis excluding extreme urine dilution concentrations (0.01 ≤ SG ≤ 1.03) (Supplemental Table 3). When modeling PFR metabolites as quartiles, we identified several negative associations among individual quartiles. DPHP (Quartile 2, p=0.04) and ΣPFR (Quartile 3, p=0.03) (Table 6) concentrations were inversely associated with sample volume, while concentrations of DPHP (Quartile 3, p= 0.02) increased the odds abnormal semen morphology (Supplemental Table 5). However, overall p-values were not statistically significant. When the semen parameters were modeled continuously (Table 6), concentrations in the third quartile for metabolites BDCIPP, ip-PPP, and Σ PFR had the strongest decrease in sample volume, yet when they were dichotomized (Supplemental Table 5), results were mixed across all parameters.

To date, there are limited studies examining the potential for adverse health effects related to PFR exposure despite their high detection in various environmental media and respective metabolites in urine. Parent compounds TDCPP and TPHP were detected in nearly all samples of house dust from a previous sample of 50 men from the EARTH cohort (Meeker and Stapleton, 2009). Similarly, a study in Durham, North Carolina (n=40 adults) detected parent compounds to BDCIPP, BCIPP, and DPHP ( TDCIPP, TCIPP, and TPHP, respectively) in 100% of samples using silicone wrist bands and >95% of hand wipes (Hammel et al., 2016). Concentrations are showing temporal progression; a recent study combining several cohorts from various parts of the U.S. found a 15-fold increase in BDCIPP samples collected in 2015 compared to those collected in 2002 (Hoffman et al., 2017a).

Studies characterizing PFRs in male populations are insufficient compared to those among women and children. Yet, analogous to high detection rate in environmental media, metabolites BDICPP and DPHP were detected in >90% of individuals (Meeker et al., 2013a) and >95% of pooled samples (Van den Eede et al., 2015). Concentrations of BDCIPP in our samples were six-fold higher compared to a prior study (n=16) of adults in California (Median=0.09 ng/mL) (Dodson et al., 2014). Our samples of DPHP were also twice as high (Median= 0.44 ng/mL), yet both were similar in having low detection of BCIPP. A small sample (n=29) of office workers in Boston, MA also had slightly lower concentration of BDCIPP (SG-adjusted Mean=408 pg/mil) (Carignan et al., 2013). However, distributions of BDCIPP, DPHP, and ip-PPP were similar to a recent study of 211 females (n=563 samples) from the EARTH cohort (SG-adjusted Mean= 0.66, 0.78, 0.22 μg/L respectively) (Carignan et al., 2017). Weak to moderate stability in repeated measurements in our sample were somewhat lower than reported from a previous study for BDICPP (ICC=0.55–0.72) and DPHP (ICC=0.35–0.51), although the sample period was considerably shorter (3 months) (Meeker et al., 2013a). We found an unexpected relationship with PFR concentrations and season of sample, where concentrations of BDCIPP (p<0.0001) were highest in summer (June–August), while DPHP (p=0.05) concentrations were highest in the winter (December–February). A sample of adults spanning the US observed a similar seasonal relationship as BDCIPP concentrations in summer were 4.13 times higher than winter samples and contrary to our observations, DPHP concentrations were also highest in summer (Hoffman et al., 2017a). Similar results were found among a sample of pregnant women, where summer (June–August) concentrations of BDCIPP and DPHP were almost 4-fold and 60% higher, respectively compared to winter samples (Hoffman et al., 2017b).

Limited research has been conducted on PFR metabolites and male reproductive health. However, we previously reported a decrease in sperm morphology (36%), straight-line velocity (18%), and curvilinear velocity (14%) in association with BDCIPP in a previous study (n=33) men from the EARTH cohort. The same study also reported decreased sperm concentration (57%) and straight-line velocity (19%) in association with urinary DPHP (Meeker et al., 2013b). Similar relationships were detected in a study (n=50) of their parent compounds in house dust where concentrations of TDCPP and TPHP were inversely associated with sperm concentration, motility, and morphology, although only the relationship between TPHP and sperm concentration was statistically significant (p=0.01) (Meeker and Stapleton, 2009). In this more robust analysis we observed suggestive declining trends in our adjusted models among BDCIPP and DPHP with total sperm count and sample volume when modeled as continuous variables. Our observations are inconsistent with previous work, possibly as a result of substantial sample size differences.

Laboratory studies assessing the reproductive impacts of PFRs are also limited, yet suggest PFRs act as endocrine disruptors and induce oxidative stress. Several in vitro models found TDCPP to be an estrogen agonist (Kojima et al., 2013; Krivoshiev et al., 2016) while another found the hydroxylated metabolite of TPHP to have stronger estrogenic activity than the parent compound (Kojima et al., 2016). A study of mouse Leydig cells concluded TPHP failed to disrupt steroidogenesis, although increased TPHP concentrations resulted in a 1.7 fold increase in superoxide production (Schang et al., 2016). However, another study of mice found TCPP and TECP to alter antioxidant enzymes and testosterone levels (Chen et al., 2015).

Although novel, our study is not without limitations. While the largest study to date, our sample size is somewhat modest. Men from a fertility clinic are a selective population that potentially limit their generalizability to the men from general population (Hotaling et al., 2012). However, the semen quality of these men is comparable with the semen quality of men from the general population. Due to limited studies characterizing PFRs in male cohorts, we are unable to conclude the PFR concentrations found in our study do not reflect levels in the general population or that men from a fertility clinic would respond differently to PFR exposure. Concentrations of PFRs in our sample are similar to those measured among the female partners of the EARTH study cohort (Carignan et al., 2017), yet considerably lower compared to more recent samples of pregnant women in Durham, North Carolina (Hoffman et al., 2017b) and Shanghai, Chania (Feng et al., 2016). Contrary to PBDEs which have a long half-life, PFRs are less-persistent and samples are subject to exposure misclassification since urinary metabolite levels may reflect exposure only hours or days prior to sample collection. However, we attempted to reduce this source of error by collecting up to five urine samples per participant and previously reported levels remain moderately stable over a three month period (Meeker et al., 2013a). Finally, while we analyzed five commonly used PFR metabolites, there are other PFRs in use that should be the focus of future investigations. Thus, our results cannot conclusively determine a lack of association with all PFR metabolites.