Persistent Environmental Pollutants and Couple Fecundity: The LIFE Study

Background: Evidence suggesting that persistent environmental pollutants may be reproductive toxicants underscores the need for prospective studies of couples for whom exposures are measured. Objectives: We examined the relationship between selected persistent pollutants and couple fecundity as measured by time to pregnancy. Methods: A cohort of 501 couples who discontinued contraception to become pregnant was prospectively followed for 12 months of trying to conceive or until a human chorionic gonadotrophin (hCG) test confirmed pregnancy. Couples completed daily journals on lifestyle and provided biospecimens for the quantification of 9 organochlorine pesticides, 1 polybrominated biphenyl, 10 polybrominated diphenyl ethers, 36 polychlorinated biphenyls (PCBs), and 7 perfluorochemicals (PFCs) in serum. Using Cox models for discrete time, we estimated fecundability odds ratios (FORs) and 95% CIs separately for each partner’s concentrations adjusting for age, body mass index, serum cotinine, serum lipids (except for PFCs), and study site (Michigan or Texas); sensitivity models were further adjusted for left truncation or time off of contraception (≤ 2 months) before enrollment. Results: The adjusted reduction in fecundability associated with standard deviation increases in log-transformed serum concentrations ranged between 18% and 21% for PCB congeners 118, 167, 209, and perfluorooctane sulfonamide in females; and between 17% and 29% for p,p´-DDE and PCB congeners 138, 156, 157, 167, 170, 172, and 209 in males. The strongest associations were observed for PCB 167 (FOR 0.79; 95% CI: 0.64, 0.97) in females and PCB 138 (FOR = 0.71; 95% CI: 0.52, 0.98) in males. Conclusions: In this couple-based prospective cohort study with preconception enrollment and quantification of exposures in both female and male partners, we observed that a subset of persistent environmental chemicals were associated with reduced fecundity.


Research
The impact of persistent environmental chemi cals on human reproduction is a topic of considerable interest. Although several persistent environmental chemicals or their metabolites have been detected in semen, folli cu lar fluid, and genital tract fluid (De Felip et al. 2004;Jirsová et al. 2010;Wagner et al. 1990), questions remain about their bio availability and ability to affect the series of highly inter related and timed processes under lying successful human reproduction. Experimental and human evidence suggests that persistent environ mental contaminants may be associated with reduced follicle count, altered estrous or menstrual cycles, ovulation inhibition, and increased pregnancy loss and resorption in humans and animals, respectively (Buck Louis et al. 2011b;Lione 1988;NicolopoulouStamati and Pitsos 2001;Pocar et al. 2003;Torf et al. 2004).
To our knowledge, only one previous cohort study measured serum persistent organo chlorine pollutants (POPs) in women who were recruited prior to conception and followed through 12 observed menstrual cycles (Buck ). However, there is evidence suggesting a reduction in female fecundity, as measured by a longer time to pregnancy (TTP), associated with persistent environmental chemicals such as 1,1dichloro2,2bis(pchlorophenyl)ethylene (DDE), dioxin, perfluoro chemicals (PFCs), polybrominated diphenyl ethers (PBDEs), and polychlorinated biphenyls (PCBs) (Axmon et al. 2005;Eskenazi et al. 2010;Fei et al. 2009;Gesink Law et al. 2005;Harley et al. 2010) when measured in female partners. In general, previous studies have quantified exposures in women at varying times during pregnancy instead of measuring exposures during the critical pre conception window (Bloom et al. 2009). Studies of pregnant women systematically exclude women who are unable to become pregnant, who may be the most heavily exposed. In addition, studies of pregnant women rely on retrospectively reported TTP, which has been shown to result in both under and overreporting of TTP (Cooney et al. 2009).
Against a background of speculation that human fecundity may be declining (Lutz et al. 2003;Skakkebaek et al. 2001), possibly as a result of effects of environmental factors on both partners of the couple as well as life style changes, we designed the Longitudinal Investigation of Fertility and the Environment (LIFE) Study of persistent environ mental chemicals and couple fecundity. By design, we sought to explore a spectrum of persistent environmental chemicals measured in both partners in relation to couple fecundability, consistent with the coupledependent nature of human reproduction (Buck Louis 2011).

Study design and cohort.
In the LIFE Study we used a prospective cohort design with preconception recruitment of couples who were discontinuing contraception for the purpose of becoming pregnant. The cohort, sampled from an enumerated target population, comprised couples of reproductive age who resided in specific geographic counties in Michigan or Texas with reported environ mental exposure to persistent environ mental chemicals, and who were planning pregnancy in the next 6 months. Given the absence of established sampling frameworks for recruiting couples who were planning pregnancy, we used a commercially available marketing database and a fishing/hunting license regis try to recruit 501 couples during 2005-2007 from the counties in the two Background: Evidence suggesting that persistent environmental pollutants may be reproductive toxicants underscores the need for prospective studies of couples for whom exposures are measured. oBjectives: We examined the relationship between selected persistent pollutants and couple fecundity as measured by time to pregnancy. Methods: A cohort of 501 couples who discontinued contraception to become pregnant was prospectively followed for 12 months of trying to conceive or until a human chorionic gonadotrophin (hCG) test confirmed pregnancy. Couples completed daily journals on lifestyle and provided bio specimens for the quantification of 9 organochlorine pesticides, 1 poly brominated biphenyl, 10 polybrominated diphenyl ethers, 36 polychlorinated biphenyls (PCBs), and 7 perfluoro chemicals (PFCs) in serum. Using Cox models for discrete time, we estimated fecundability odds ratios (FORs) and 95% CIs separately for each partner's concentrations adjusting for age, body mass index, serum cotinine, serum lipids (except for PFCs), and study site (Michigan or Texas); sensitivity models were further adjusted for left truncation or time off of contraception (≤ 2 months) before enrollment. results: The adjusted reduction in fecundability associated with standard deviation increases in log-transformed serum concentrations ranged between 18% and 21% for PCB congeners 118, 167, 209, and perfluorooctane sulfonamide in females; and between 17% and 29% for p,p´-DDE and PCB congeners 138, 156, 157, 167, 170, 172, and 209 in males. The strongest associations were observed for PCB 167 (FOR 0.79; 95% CI: 0.64, 0.97) in females and PCB 138 (FOR = 0.71; 95% CI: 0.52, 0.98) in males. conclusions: In this couple-based prospective cohort study with preconception enrollment and quantification of exposures in both female and male partners, we observed that a subset of persistent environmental chemicals were associated with reduced fecundity. states. Inclusion criteria were as follows: a) females were 18-40 years of age and males were ≥ 18 years of age; partners were in a committed relationship; neither partner was medically/surgically sterile; the female's menstrual cycle was between 21 and 42 days; the female received no injectable contraceptives within 12 months; the couple had not used contraception for < 2 months; and both partners were able to communicate in English or Spanish. Introductory letters were mailed to the targeted cohort (n = 424,423); after telephone screening to identify eligible couples (n = 1,184), 501 couples (42%) were enrolled (Buck Louis et al. 2011c).
Data and biospecimen collection. All data collection occurred in the couples' home. At the start of the study visit, the female partner provided a urine sample that was tested with a home pregnancy test capable of detecting 25 mIU/mL human chorionic gonado tropin (hCG) to ensure she was not pregnant. This important step permitted us to differentiate between couples achieving pregnancy in the first few weeks after enrollment (i.e., with no menstrual cycle occurring between enroll ment and pregnancy) versus during the first fully observed cycle, which we denote in the analysis as cycles 0 and 1, respectively. Each partner of the couple was interviewed separately by one of two research assistants. Interviews were followed by a standardized physical anthropometric assessment (Lohman et al. 1988) to determine body mass index (BMI), and blood and urine samples were collected. Specifically, ≈ 20 cc of nonfasting blood and ≈ 120 cc of urine were collected from each partner of the couple. Blood collec tion equipment was free of the contaminants under study. Couples were instructed how to complete daily journals regarding sexual intercourse and lifestyle factors (e.g., cigarette smoking); journals of females also recorded menstruation and results of home pregnancy tests. Couples had the option of completing journals either in hardcopy or online.
Female partners were instructed in the use of the Clearblue® Easy home fertility moni tor (Swiss Precision Diagnostics formerly Unipath). This urinebased monitor tracks the rise in estrone3glucuronide (E 3 G) and luteinizing hormone (LH) across the follicu lar phase of the ovarian cycle, and displays a prompt that ranges from low to peak fertil ity to help the couple time intercourse rela tive to ovulation. The monitor was used to enhance couples' ability to conceive, given that it is 99% accurate in detecting the LH surge, compared with the gold standard of vaginal ultrasonology (Behre et al. 2000). The moni tor was intended to minimize each couple's chances of missing ovulation, which might erroneously lengthen TTP. Women were also trained in the use and interpretation of the Clearblue® Easy home pregnancy test, a digital device that indicates the test result as either pregnant or not pregnant. All partici pants were remunerated $75 for full participa tion in the study. Approval for use of human subjects was obtained from all collaborating institutions, and all participants gave informed consent before participation.
Serum concentrations were measured using isotope dilution highresolution mass spec trometry for OCPs, PBBs, PBDEs, and PCBs, and isotope dilution tandem mass spectrometry for PFCs, following standard published operat ing procedures as described previously (Kato et al. 2011;Kuklenyik et al. 2005;Sjödin et al 2004). We did not perform automatic sub stitution of concentrations below the limit of detection or lipid adjustment in order to minimize bias associated with such practices when estimating health effects (Richardson and Ciampi 2003;Schisterman et al. 2005Schisterman et al. , 2006. Serum concentrations of cotinine [quantified using liquid chromatographyisotope dilu tion tandem mass spectrometry (Bernert et al. 1997)] and PFCs are reported in nanograms per milliliter; all other chemical concentra tions are reported in nanograms per gram of serum. Serum lipids, quantified using com mercially available enzymatic methods (Akins et al. 1989), were reported as total serum lipids (nanograms per gram of serum) using estab lished methods based on individual compo nents including phospholipids, tri glycerides, total cholesterol, and free cholesterol (Phillips et al. 1989).
Operational definitions. Couple fecun dity was measured by TTP, which denotes the number of menstrual cycles required by couples to achieve an hCG pregnancy. Couples achieving pregnancy within the first few weeks of enrollment or before a fully observed menstrual cycle were defined as hav ing a TTP of 0; couples not achieving preg nancy after 12 months of trying were censored (TTP > 12). Definitions of relevant covariates included BMI (measured weight in kilograms divided by height in meters squared), gravidity (number of pregnancies), parity (number of live births), and smoking status based on serum cotinine concentration (continous).
Statistical analysis. In the descriptive phase of analysis, we assessed the distributions of all exposures and relevant covariates. We analyzed data under the missingatrandom assumption. Specifically, we implemented Markov Chain Monte Carlo methods to impute missing chemical, cotinine, and lipid (≤ 4%) data arising from insufficient blood for analysis (Schafer 1997). We used other chemical exposures for the imputation process. Geometric means (GMs) and 95% confidence intervals (CIs) were calculated for all chemicals, cotinine, and serum lipids.
We used daily journals supplemented with fertility monitors as needed to define men strual cycles distinct from episodic bleeding. Specifically, a menstrual cycle denoted the interval (in days) from the onset of bleeding that increased in intensity and lasted ≥ 2 days to the onset of the next similar bleeding epi sode. Pregnancy was defined as a positive (hCGconfirmed) test on the day of expected menstruation.
The analytic phase was conducted in two parts. First, we estimated unadjusted fecund ability odds ratios (FORs) and accompanying 95% CIs for all 63 chemi cals by class (i.e., OCPs, PBB, PBDEs, PCBs, PFCs) using Cox models (Cox 1972) for discrete survival time (SAS version 9.2; SAS Institute Inc., Cary, NC) to estimate the odds of becoming preg nant during each cycle given exposure and conditional on not being pregnant in the pre vious cycle. This model allows the odds for pregnancy to vary from cycle to cycle through a cyclevarying intercept. Each chemical con centration (nanograms per gram of serum or wet weight) was logtransformed and divided by its standard deviation to rescale concentra tions for biologic interpretation of the FORs, given the small unit size of chemical concen trations. Next, for chemicals that were sig nificantly associated with TTP on the basis of unadjusted estimates, we ran additional mod els adjusted for a priori potential confounders [i.e., continuous age, BMI, serum cotinine, and serum lipids (except for PFC models); research site (Michigan or Texas); and the sum of the logtransformed serum concentrations of all other measured chemicals in the same class as the chemical being evaluated (American Society for Reproductive Medicine 2008a; Augood et al. 1998;Hassan and Killick 2004;RamlauHansen et al. 2007)]. In addition, we accounted for left truncation reflecting time (≤ 2 months) couples did not use con traception before enrollment into the study. In this approach we assumed that months cor respond to menstrual cycles and that all unob served time is at risk for pregnancy despite the absence of data on sexual intercourse in rela tion to the fertile window. Underlying linear ity for all continuous covariates were assessed using the Kolmogorovtype supremum test based on martingale residuals, and the propor tional hazards assumptions were verified for all discretetime models (Grambsch and Therneau 1994;Therneau and Grambsch 2000).
We also evaluated interactions between each chemical and age categorized as ≤ 27 versus > 27 years for females and ≤ 28 versus > 28 years for males based on previous evi dence for this categorization (Dunson et al. 2002) and as corroborated in our cohort. However, we did not include interactions in our final models because none were observed.
Finally, we ran separate models adjusted for parity in sensitivity analyses, given the uncertain causal relationship between parity, POPs, and TTP. Specifically, we modeled par ity conditional on gravidity by categorizing it as no prior pregnancy, prior pregnancy with out live birth(s), or prior pregnancy with live birth(s) (Buck Louis et al. 2006).
Separate models were run for each chemi cal and partner. The concentrations of the chemicals evaluated were highly correlated with each other [for correlations in samples from females and males, see Supplemental Material, Figures S1 and S2, respectively (http://dx.doi.org/10.1289/ehp.1205301)] and between partners (see Supplemental Material, Figure S3; r = 0.71-0.97), which precluded joint modeling. Couples who withdrew from the study before pregnancy or before com pleting 12 months of followup (n = 100) or who were not pregnant after 12 months of followup (n = 54) were censored in all analy ses. Statistical significance (p < 0.05) was determined using the chisquare statistic for categorical data, Student's ttest or Wilcoxon non parametric test for continuous data, or 95% CIs that excluded one. We did not adjust for multiple comparisons consistent with the exploratory nature of this work.

Results
Socio demographic and lifestyle charac teris tics differed between couples who became preg nant or completed 12 months of followup and those who withdrew before pregnancy or the end of followup (Table 1). Participants who withdrew were more likely to selfidentify as non white and to be without health insurance. Compared with women who completed the study, those who withdrew had signifi cantly higher mean BMIs (27.2 vs. 29.4 kg/m 2 ) and logtransformed serum cotinine concentra tions (0.51 ng/mL vs. 1.08 ng/mL). Men who withdrew from the study had higher log transformed serum cotinine concentrations (1.98 ng/mL) than those who completed the study (1.04 ng/mL). The probability of preg nancy at cycles 1, 3, 6, and 12 were 0.27 (95% CI: 0.23, 0.31), 0.52 (95% CI: 0.48, 0.57), 0.68 (95% CI: 0.64, 0.73), and 0.81 (95% CI: 0.76, 0.85), respectively. Table 2 presents the GMs and 95% CIs for chemicals that were significantly associ ated with fecundity based on the unadjusted FORs; for corresponding estimates for all other chemicals see Supplemental Material, Tables S1 and S2 (http://dx.doi.org/10.1289/ ehp.1205301). Among the chemicals shown in Table 2, GMs were comparable or higher in couples who withdrew before 12 months of followup or did not become pregnant dur ing 12 months of followup compared with couples who became pregnant during fol low up, although only values for PBDE 183 and PCB 138 in men were significantly dif ferent (0.003 vs. 0.002; p < 0.01 and 0.044 vs. 0.038; p < 0.05, respectively).
Of all the chemicals tested in models adjusting for left truncation or any time off contraception before enrollment, only PCB 101 in men had a significant positive association with TTP (unadjusted or adjusted), indicating a shorter time to   (Table 3). The specific chemicals that were significantly associated with reduced FORs differed between females and males, with considerably more significant negative associations noted for exposures in men. Chemicals with significant negative associations based on unadjusted models were p,p´DDE, PBDE 183,and PCB congeners 101,138,153,156,157,167,170,172,180,and 209 in males,and HCB,PFOSA,and PCB congeners 118,167,and 209

Discussion
Findings from the LIFE Study, a prospective cohort of couples enrolled prior to conception and followed for up to a year while attempt ing to become pregnant, provide empirical evidence that selected persistent environmen tal chemicals from various chemical classes (i.e., OCPs, PCBs, and PFCs) may adversely affect couple fecundability. Serum concen trations among the LIFE Study participants were largely below those reported for U.S. populations during a comparable time period (Centers for Disease Control and Prevention 2009), possibly reflecting the younger age structure of the LIFE Study cohort relative to the United States as a whole.
A novel finding is that the chemicals asso ciated with reduced couple fecundability dif fered between males and females, but with a larger number of associations observed for males. We previously observed a similar find ing for heavy metals and TTP in the LIFE Study, in which increased concentration in male partners was associated with a longer TTP (Buck Louis et al. 2012). These find ings underscore the importance of males when assessing coupledependent reproductive outcomes such as TTP. However, serum concentrations of monoortho PCB 167 and non-dioxinlike PCB 209 were associated with approximately a 20% reduction in the probability of an hCGdetected pregnancy per standard deviation increase in the log transformed chemical concentration in both men and women, although it is important to note that serum concentrations of these PCB congeners were below the limit of detection in 71% and 77% of women and men, respec tively, for PCB 167 and in 77% and 51% of women and men for PCB 209. Still, we know of no a priori reason or empirical evidence that supports a systematic difference in laboratory detection capability by couple fecundability, particularly given the blinding of laboratory personnel to fecundity status in the LIFE Study. The findings warrant further inquiry, given the study's exploratory nature.
Interpreting our findings for partner specific associations in the context of the previous literature is limited by the absence of simi lar preconception couplebased cohorts with exposure characteriza tion for a mixture of persistent chemicals and 12 cycles of followup consistent with the clinical diagnosis of infertility (American Society for Reproductive Medicine 2008b). Previous studies have assessed selected chemicals and TTP, but these studies were primarily among pregnant women with blood collection at varying times during gestation and retrospectively reported TTP.
Several findings from the LIFE Study are globally consistent with earlier studies that reported reduced FORs for various PCBs (Axmon et al. 2005;Gesink Law et al. 2005), although only the results reported by Axmon et al. (2005) were statistically significant. Our findings also corroborate the lack of an association between female concentrations of p,p´-DDT, o,p´-DDT, and p,p´-DDE and fecundabil ity (Harley et al. 2008). Unlike the findings of Harley et al. (2010), we did not observe a relation ship between PBDEs and FORs, possibly reflecting differences in the adjusted models or the earlier studies' use of retrospec tively measured TTP. Recently, three papers have addressed the relationship between select PFCs and fecundity (Fei et al. 2009;Vestergaard et al. 2012;Whitworth et al. 2012), although only one used a prospective couplebased cohort design (Vestergaard et al. 2012). Specifically, Vestergaard et al. (2012) observed no consistent pattern between eight PFCs, including PFOSA, measured in serum from 222 (52%) participating female partners who were followed for up to six cycles of attempting to become pregnant in 1992-1995. Important differences exist between this Danish cohort and the LIFE Study, including a shorter duration of fol lowup (6 vs. 12 cycles) in the earlier study,  Olsen et al. (2009). The inclusion of parity conditional on gravidity in our sensitivity models did not sub stantially alter associations, which we believe is inconsistent with reverse causation. However, the association between serum PFOSA con centration in women and longer TTP, as indi cated by the FOR < 1, must be interpreted with caution because concentrations were non detectable in 90% of samples, possibly because U.S. production of PFOSA ceased in 2002. The magnitude of negative associations with fecundability estimated for several of the POPs assessed in the LIFE Study are comparable to associations with other recog nized fecundity determinants such as male and female age, BMI, and cigarette smoking (Bolumar et al. 1996;Dunson et al. 2002;Menken et al. 1986;RamlauHansen et al. 2007) that we adjusted for in our statistical models. These findings underscore the impor tance of environ mental factors that may affect couple fecundity as measured by TTP. Our findings are relatively consistent with studies of women undergoing assisted reproductive technologies that can examine associations with early reproductive outcomes (e.g., fer tilization, cleavage, implantation) that are not observable in the general population. For example, evidence of negative associations between TTP and serum concentrations of HCB and PCB 118 in women is consistent with findings for implantation failures asso ciated with these exposures among women undergoing assisted reproductive technolo gies (Mahalingaiah et al. 2012;Meeker et al. 2011). Although speculative, these findings suggest that associations between these chemi cals and longer TTPs may reflect, in part, diminished implantation success. Still, our findings have important limitations, including the absence of information on the timing of exposures during sensitive windows for human reproduction (e.g., folliculo genesis, spermato genesis), the absence of exposure data on shortlived chemicals such as bis phenol A and phthalates (Crain et al. 2008;Jurewicz and Hanke 2011), potential selec tion biases arising from enrollment of couples planning pregnancies, and possible residual confounding associated with more educated women using the monitor more effectively than lessereducated women. We did not observe any differences in frequency or timing of intercourse as aided by the monitor and female education, nor did women experience difficulties complying with the monitor (data not shown).
The etiologic mechanisms by which endocrinedisrupting chemicals (EDCs), including those quantified in the LIFE Study, may affect human reproduction remain elusive, but globally these mechanisms are hypothe sized to affect hormonal milieu through alterations in the production, release, transport, metabolism, and/or elimination of hormones (Sonnenschein and Soto 1998). With regard to ovarian function, experimental and human evidence suggests that EDCs alter both the expression/activity of enzymes required for ovarian sex steroid synthesis/catabolism, and the expression/ability of hormone receptors to bind endogenous ligands, as recently reviewed by Craig et al. (2011). Such changes that occur within the ovary are not in isolation because other endocrine organs that are relevant to reproduction, such as the thyroid, also may be affected by EDCs (Boas et al. 2012). The potential for diverse mechanisms of action underscore the need to consider couples' exposures in relation to a broad spectrum of human reproductive outcomes, including altered hormonal profiles or sexual libido in either partner, semen quality in the male partner (Hauser et al. 2006), and changes in menstrual and ovarian cycles (Perry et al. 2006) and effects on ovulation or implantation in the female partner. Each of these end points, either alone or in combination, may manifest as a longer TPP.
Delineating an underlying causal model that might explain associations between EDCs and TTP remains a critical data gap. An exposome approach that captures the totality of non genetic exposures from concep tion onward (Wild 2005) would allow chemi cal exposures to be evaluated in the context of lifestyle, behavior, and macrolevel factors that also may be relevant to human fecun dity and fertility. The need for a comprehen sive approach to improve under standing of risk factors and underlying mechanisms has been proposed in relation to the testicular dys genesis syndrome (Skakkebaek et al. 2001) and, subsequently, the ovarian dysgenesis syndrome (Buck Louis et al. 2011a), both of which may result in part from early expo sures that may permanently reprogram fecundity and have implications across the life span. Such conceptual and methodologic approaches will facilitate understanding of the up and downstream effects that EDCs may pose for human reproduction and health.