Adipose to serum ratio and mixtures of persistent organic pollutants in relation to endometriosis: Findings from the ENDO Study

The study population has been described previously in detail ( Buck Louis et al., 2011 ). Briefly, women were eligible for inclusion if they were: aged 18–44 years, had no previous endometriosis, cancer or injectable hormone usage within two years, and were not breastfeeding within 6 months to minimize recent changes in POP concentrations. Fourteen clinical centers in the Salt Lake City, Utah and San Francisco, California geographic areas enrolled participants who were undergoing laparoscopy or laparotomy between 2007 and 2009. All participating study sites obtained institutional review board approval for the study. The women (n=339) participating in the study provided full consent prior to data collection and were compensated for their participation. Baseline in-person interviews were completed approximately two months before surgery to ascertain information on demographics, health history, and behaviors. Operative findings were collected using standardized processes. Non-fasting blood specimens (24 mL) were collected using containers free from POP contamination under study after the baseline interview. Samples were refrigerated until shipment to the laboratory for analysis. During surgery, omental fat (1–5 grams) samples were obtained subject to clinical judgement ( Buck Louis et al., 2012 ), given the observational design of the study that was not designed to change clinical practice. Omental fat specimens were stored in Wheaton brown glass bottles. For four women, epiploica appendiceal fat was obtained rather than omental fat, two from each study site. Incident endometriosis was defined by the clinical gold standard of surgical visualization ( Kennedy et al., 2005 ). Body mass index was ascertained as kg/m 2 , measured by trained study staff using a standardized protocol ( Backonja et al., 2017 ). Breast-feeding history was derived as a conditional variable based upon parity (nulliparous/parous) and categorized as: no prior birth and prior birth with or without breast-feeding. Recent weight loss or gain denoted a self-reported change of 10 or more pounds within 5 years and categorized as recent weight loss, weight gain or both weight loss and gain, or other. All chemicals were quantified using gas chromatography (GC) /mass spectrometry (MS) with GC/electron capture detector and GC/high-resolution MS (HRMS) ( Johnson-Restrepo et al., 2007 , 2005 ; Sjodin et al., 2004 ) for both serum and omental adipose tissue. Three chemical classes were measured: 1) 7 estrogenic polychlorinated biphenyl (PCB) congeners #44, 49, 52, 101, 177, 187, 201 and 6 anti-estrogenic PCB congeners #66, 74, 118, 128, 138, 170 ( Wolff et al., 1997 ); 2) 6 polybrominated diphenyl ether (PBDE) congeners #47, 99, 100, 153, 154, 209; and 3) 11 organochlorine pesticides [ p,p’ -dichlorodiphenyltrichloroethane ( p,p’ -DDT) and its metabolites o,p’ -DDT and p,p’ dichlorodiphenyldichloroethylene ( p,p’ -DDE), cis- and trans-nonachlor, hexachlorobenzene (HCB), beta-hexachlorcyclohexane (β-HCH) and gamma-hexachlorcyclohexane (ɣ-HCH), cis- and trans-chlordane, oxychlordane ( Buck Louis et al., 2012 ). Quality control procedures included external calibration standards that were analyzed with each set of samples and three procedural blanks in each batch ( Buck Louis et al., 2012 ). Instrument derived concentrations were reported without substituting for values below the limit of detection (LOD) to minimize bias when assessing human health outcomes ( Guo et al., 2010 ; Richardson and Ciampi, 2003 ; Schisterman et al., 2006 ). Total lipids (TL) were reported in mg/dL and were approximated as TL= (2.27 x total cholesterol) + triglycerides + 62.3 mg/dL ( Phillips et al., 1989 ). The ASR was calculated by dividing wet weight adipose concentrations (ng/g) by wet weight serum concentrations (ng/g). In cases where the use of instrument-derived values with background subtraction that resulted in negative serum values, a small chemical-specific constant was added to the serum value to ensure the ratio was positive with a smallest value of 10 −5 . The distributions of each chemical and measure were examined using histograms. The ASR measures for all chemicals were summarized using the median and interquartile range (IQR) along with chemical octanol-water partition coefficients (logK ow ). Correlations between chemicals across biological matrices and measures (serum, adipose, and ASR) were explored. For statistical modeling, natural log transformation was used to account for right-skewness in each chemical measure, and all measures were scaled by their standard deviation. The distributions of each chemical and measure were examined using histograms. Correlations between chemicals across biological matrices and measures (serum, adipose, and ASR) were explored. Additionally, Spearman correlation coefficients were calculated within-chemicals for different measures (e.g., the correlation between serum PCB 44 vs. adipose PCB 44). A previous study characterized associations between individual chemicals measured in serum and omentum adipose tissue and odds of an incident endometriosis diagnosis ( Buck Louis et al., 2012 ). Separate logistic regression models estimated associations between endometriosis and each chemical using three measures: serum, adipose, and ASR. Based upon a priori knowledge, models were adjusted for age in years, body mass index, serum cotinine, and breastfeeding conditional on parity (nulliparous, parous and never breastfed, parous and breastfed). Results were reported as odds ratios (OR) and 95% confidence intervals (CI), specifically the odds of an incident endometriosis diagnosis for a one standard deviation increase in the chemical measure. A goal of this study was to determine whether estimated associations with endometriosis varied by serum, adipose, and ASR chemical measures to assess potential value of ASR. The number of estimated associations from single chemical models is large, presenting challenges for ad hoc comparisons of ORs between chemical measures. Therefore, to quantitatively compare results from logistic regression between serum, adipose, and ASR measures by chemical class, we applied Bayesian hierarchical models (BHM). BHM are widely used in the air pollution epidemiology literature to pool associations across multiple regression models while incorporating standard errors for each estimated coefficient ( Dominici et al., 2000 ; Krall et al., 2018 ). The BHM approach incorporates the estimated uncertainty from each single chemical regression model and directly estimates differences in associations with endometriosis between, for example, using adipose vs. serum chemical levels. The full BHM is described in detail in the supplementary material . The BHM provides the average percent change in the OR and 95% posterior interval (PI) when using adipose and ASR chemical measures compared to using less invasive serum chemical measures. In addition to estimating associations with endometriosis for single chemicals individually, we also estimated associations for the cumulative mixture effects of multiple chemicals within a class simultaneously to more closely approximate real-world exposures ( Braun et al., 2016 ; Carlin et al., 2013 ). We applied Bayesian kernel machine regression (BKMR) ( Bobb et al., 2015 ), which flexibly models correlated exposures together without a specific exposure-response function pre-specified, allowing for potential nonlinearity and interaction between chemicals. Separate BMKR models were applied to each measure (adipose, serum, ASR) and each chemical class: estrogenic and anti-estrogenic PCBs, PBDEs, and OCPs. Chemical classes were used to account for common exposure sources. Across measures, there were some pairs of chemicals with high within-class correlations ( Figures S1 – S3 ), which we accounted for within BKMR models using component-wise variable selection. Chemical classes were used to account for common exposure sources and correlations within class. Models were run for 50,000 iterations, with a burn-in of 30,000. Convergence was assessed using trace plots and by comparing results from two chains for each model. Results represent the difference in risk of endometriosis and 95% PI when the mixture is at exposure percentile p (e.g., 75 th exposure percentile) compared to the 25 th exposure percentile ( Bobb et al., 2018 , 2015 ). The 25 th exposure percentile was chosen as the referent given our a priori assumption of no negative (reduced risk) associations ( Ouidir et al., 2020 ). Analyses were completed using R (Vienna, Austria) using the bkmr package ( Bobb, 2017 ). We compared our logistic regression results to models that adjust for lipids by dividing serum and adipose levels by total lipids. Sensitivity analyses were also conducted to evaluate the robustness of our logistic regression results to effect modification by BMI category (underweight, normal weight and overweight; BMI <30.0 kg/m 2 ) in comparison with obese (BMI ≥30.0 kg/m 2 ) and Bonferroni correction will be applied to account for multiple comparisons. Last we compared our results to models with adjustment for recent substantial weight loss or gain, defined as a ≥10 pound change in weight (increase, decrease, both, or unknown) in the last five years.

Chemical ASRs were available for 339 women who had adipose and serum chemical measures. The incidence of endometriosis and participant characteristics have been previously reported. Briefly, participants had a median age of 33 years (interquartile range IQR: 11 years) and median body mass index 26.4 (IQR: 9.95 kg/m 2 ) ( Buck Louis et al., 2011 ). Table 1 presents the medians and IQR for the ASRs of POPs. The highest median (IQR) ASRs were p,p’ -DDE, 754 (IQR 1064) and oxychlordane, 738 (IQR 1311). The lowest median ASRs were for PCBs 44, 49, 52 and 66, which were all <1. The logK ow varied from a low of 3.72 for β -HCH and γ -HCH and a high of 10.10 for BDE209. Generally, PCBs with lower logK ow had lower ASRs but this was not consistent for PBDEs (as BDE 209 accumulated less in adipose) or OCPs. Median ASRs were higher among women with than without endometriosis for all estrogenic PCB congeners except PCB177. Median PBDE ASRs were mostly higher among women without than with endometriosis, except for BDE154 and BDE209. Women with endometriosis had higher median ASRs for hexachlorobenzene, β-HCH, ɣ-HCH, oxychlordane and trans-chlordane than unaffected women. Correlation coefficients were generally weak between serum and adipose levels for all POPs, whereas correlations were generally high for adipose and ASR ( Figure 1 ). Correlations of chemicals by serum, adipose, and ASR are presented in Supplemental Figure S1 – S4 . The POP levels in serum and adipose tissue were presented previously ( Buck Louis et al., 2012 ). Medians and IQRs for adipose and serum levels are presented in Supplemental Table S1 . Associations for ASR of estrogenic PCBs with endometriosis were generally positive ( Figure 2 ). Two ASR for OCPs, β-HCH OR=1.5 (95% CI: 1.2, 1.9) and ɣ-HCH OR=1.6 (95% CI: 1.2, 2.0), were positively associated with endometriosis. There was less indication of associations for other OCPs, anti-estrogenic PCBs, and PBDEs. Bayesian hierarchical models (BHM) applied to estimated associations demonstrated that OR were 14% (95% PI: 6%, 22%) and 20% (95% PI: 12%, 29%) higher for adipose and ASR measures, respectively, compared to serum levels ( Figure 3 ). Associations varied by chemical class. For estrogenic PCBs, ORs were 29% (95% PI: 12%, 49%) higher based upon adipose levels and 36% (95% PI: 18%, 56%) higher for ASR measures compared to serum levels. For anti-estrogenic PCBs and OCPs, the ORs for ASR measures were more than 20% higher compared to serum levels. In contrast, the ORs for ASR of PBDEs were comparable regardless of biologic media. Results from the BKMR models, which compared quantiles of chemical class mixtures to the 25 th percentile, were largely consistent with logistic regression results ( Figure 4 ). We utilized BKMR models with component-wise variable selection. Across serum, adipose, ASR measures and chemical classes, there were no suggestions of non-linear associations for the cumulative mixture. For simplicity, we focus on results comparing the 75 th percentile to the 25 th percentile. For adipose and ASR, associations were increased at high quantiles compared to 25 th percentiles for estrogenic PCBs and OCPs, whereas there was less indication of associations for PBDEs and anti-estrogenic PCBs. For the ASR, the estimated change in estimate and 95% PIs comparing 75 th percentile to 25 th percentiles for estrogenic PCBs was 0.37 (95% PI: −0.06, 0.80), for anti-estrogenic PCBs was −0.02 (95% PI: −0.32, 0.27), for PBDEs was −0.08 (95% PI: −0.38, 0.21) and for OCPs was 0.26 (95% PI: −0.05, 0.57). For adipose, comparing 75 th percentile to 25 th percentile for estrogenic PCBs the change in estimate was 0.27 (95% PI: −0.18, 0.72), for anti-estrogenic PCBs, −0.08 (95% PI: −0.46, 0.31), for PBDEs, −0.26 (95% PI: −0.58, 0.06) and for OCPs, 0.17 (95% PI: −0.21, 0.56). Across chemical classes, there was little indication that serum levels were associated with endometriosis. For serum, comparing 75 th percentile to 25 th percentile for estrogenic PCBs the change in estimate was −0.04 (95% PI: −0.2, 0.12), for anti-estrogenic PCBs it was −0.05 (95% PI: −0.19, 0.10), for PBDEs it was −0.09 (95% PI: −0.38, 0.20) and for OCPs the change in estimate was −0.08 (95% PI: −0.36, 0.21). The logistic regression results for the ASR were not sensitive to lipid adjustment ( Supplemental Figure S5 ). There were 113 obese women (33%) in our study, and our results differed somewhat by BMI category ( Supplemental Figure S6 ). The specific estrogenic PCB congeners and OCP in adipose tissue and through ASR associated with endometriosis varied between obese and non-obese women. There was some indication that anti-estrogenic PCBs, primarily PCB congeners 114, 138 and 170 were associated with endometriosis among obese women. After applying Bonferroni correction, no interactions were statistically significant suggest the findings were robust to recent weight change adjustment ( Supplemental Figure S7 ).

Endometriosis is an estrogen-dependent disease associated with significant morbidity and infertility, with annual costs in the U.S. on par with those associated with diabetes ( Simoens et al., 2012 ). Endometriosis affects 6–11% of premenopausal women ( Balasch et al., 1996 ; Buck Louis et al., 2011 ; Rawson, 1991 ), and its etiology is unclear. Some endocrine disrupting chemicals are associated with an endometriosis diagnosis ( Smarr et al., 2016 ) and, more broadly, are estimated to contribute to the burden of gynecologic disease with annual costs in excess of $1.5 billion annually, when including uterine leiomyoma and adenomyosis ( Hunt et al., 2016 ). Some persistent organic pollutants (POPs), including polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs), and organochlorine pesticides (OCPs) have been linked with endometriosis risk ( Cano-Sancho et al., 2018 ; Heilier et al., 2008 ; Louis, 2012 ; Porpora et al., 2009 ; Reddy et al., 2006 ; Simsa et al., 2010 ; Smarr et al., 2016 ; Upson et al., 2013 ), although some studies found no such relationship ( Fierens et al., 2003 ; Hoffman et al., 2007 ; Niskar et al., 2009 ; Pauwels et al., 2001 ). PCBs, OCPs, and PBDEs are stable and resist degradation in the environment. Their lipophilicity leads to biomagnification in food as demonstrated by marine and Arctic wildlife and trophic studies ( Borga et al., 2004 ; de Wit et al., 2010 ; Nakata et al., 2005 ). Diet is the primary source of human exposure to PCBs ( Domingo and Bocio, 2007 ) and OCPs ( Tsukino et al., 2006 ) , while ingestion and absorption of house dusts is a major source of PBDEs ( Johnson-Restrepo and Kannan, 2009 ). PCBs are industrial chemicals used until they were banned in the US in 1979. In serum, PCB half-lives vary by congener from 1.6 years for PCB44 to infinite for PCB153 ( Wolff et al., 1992 ). More recently, estimates of intrinsic half-lives identified 2.6 years for PCB44 to 15.5 years for PCB170 ( Ritter et al., 2011 ). PBDEs have been used as flame retardants in textiles and plastics ( Darnerud et al., 2001 ). Higher-brominated PBDEs had shorter half-lives in blood, such as 15 days for decabromodiphenyl ether (BDE209) ( Thuresson et al., 2006 ) while for BDE153 the half-life was 4 years ( Sjodin et al., 2020 ). Organochlorine pesticides were used as insecticides globally during 1940s-1970s. Due to their persistence in the environment, exposure continues, although current fat and dairy intake may not be strong exposure determinants ( Bradman et al., 2007 ). The OCP beta-hexachlorocyclohexane ( β -HCH) half-life in blood was 7.2 years and 7.6 years in lipids ( Jung et al., 1997 ). The biological half-life for the insecticide 1,1,1-trichloro-2,2-bis( p- chlorophenyl)ethane (DDT) is 7 years ( Woodruff et al., 1994 ). Due to their lipophilic nature, these chemicals are considered persistent. POPs are stored in adipose tissue, though they are typically measured in serum as a more accessible proxy of adipose concentration ( Johnson-Restrepo et al., 2005 ). The partition of POPs between serum and adipose tissue depends on the chemical’s structural properties ( Jackson et al., 2017 ) and biological factors that contribute to mobilization of lipids. Even within a chemical class, the extent of uptake into adipose tissue is not uniform. For example, uptake into adipocytes varies across PCB congeners and depends on lipophilicity and the size of the molecule ( Bourez et al., 2013 ; Louis et al., 2016 ). An additional key feature is that the composition of adipose tissue is not uniform, as it contains various cell types including fibroblasts, preadipocytes and macrophages ( Sbarbati et al., 2010 ). POPs are stored predominantly in adipocytes ( Bourez et al., 2013 ). The octanol-water partitioning coefficient (K ow ) measures lipophilicity with higher values indicating greater affinity for lipids ( Schwarzenbach et al., 2005 ). The log K ow of PCB congeners ranges from 4.30 to 8.26 ( Ritter et al., 1995 ) depending on the degree of chlorination and molecular size ( Levitt, 2010 ). A related measure is the adipose-serum partition coefficient, which is the ratio of the concentration in adipose to serum at equilibrium that may depend on individual-level factors ( Petreas et al., 2004 ). Storage of POPs in adipose tissue may diminish acute exposure to target tissues. However, chronic POP concentrations in adipose tissue may affect lipid dynamics by interfering with metabolism or contributing to inflammation in adipose tissue ( La Merrill et al., 2013 ). Adipose tissue is a storage depot for POPs and a source for continual or chronic internal exposure ( La Merrill et al., 2013 ). There is evidence of higher POP levels in visceral than subcutaneous adipose tissue ( Pestana et al., 2014 ), though others found no differences ( Malarvannan et al., 2013 ). These functional and compositional differences support the use of adipose tissue for quantifying lipophilic chemicals to the extent it is available for study populations. The accumulation of specific POPs in a biologic compartment may be driven by properties of the chemicals and/or biospecimen. Given these aspects of POP storage and due to the convention that adipose measurement on a population level is not feasible, most studies investigating health effects of lipophilic chemicals use serum or plasma for exposure assessments. Nevertheless, adipose tissue is thought of as a marker of cumulative internal dose for lipophilic chemicals ( Aronson et al., 2000 ; Whitcomb et al., 2005 ; Yu et al., 2011 ). Yet, correlations between serum and adipose levels of lipophilic chemicals vary, indicating that the relationship between circulating levels (blood) and cumulative internal dose (adipose) varies. For example, correlations between serum and adipose levels of DDE ranged from 0.30–0.76 ( Arrebola et al., 2012 ; Artacho-Cordon et al., 2017 ; Lopez-Carrillo et al., 2001 ; Stellman et al., 1998 ; Whitcomb et al., 2005 ). Furthermore, associations in relation to endometriosis varied in direction and magnitude for the same chemicals measured across serum, plasma, and adipose tissue ( Buck Louis et al., 2012 ; Cai et al., 2011 ; Ploteau et al., 2017 ). The dynamic equilibrium in POPs levels in human tissues can be affected by various physiological conditions (e.g., illness, starvation). The equilibrium between POPs stored in adipose tissue and circulating in serum may underlie the true relationship between these lipophilic chemicals with endometriosis and other health outcomes. One previous study reported differences in the novel POP adipose versus serum ratio (ASR) between endometriosis cases compared to controls ( Ploteau et al., 2016 ), indicating that the ASR may be an important marker of POP exposure. The ASR can incorporate individual differences in metabolism and biotransformation between adipose and serum compartments that may be more relevant to the target tissue than merely considering levels circulating in serum. Therefore, it may be considered as a new and underutilized approach to consider how lipophilic chemicals affect human health ( Ploteau et al., 2016 ). Prior studies have not quantitatively compared associations between serum, adipose, and ASR POPs or their cumulative chemical mixtures in relation to endometriosis. To address this gap, we evaluated the ASR levels for individual chemicals and within-class mixtures of 13 PCB congeners, 6 PBDE congeners, and 11 OCPs in relation to the odds of an incident endometriosis diagnosis. We compared the ASR generated with those measured in either serum or adipose tissue (viz., omentum fat). As such, the ASR is intended to be an equilibrium measure of the two biological compartments that may offer new insight into the relationship of endometriosis with PCBs by estrogenic grouping ( Wolff et al., 1997 ), PBDEs, and OCPs. To our knowledge, this is the first study to utilize the ASR for assessing specific POPs and an incident endometriosis diagnosis) to delineate environmental chemicals associated with this gynecologic disease and to quantitatively compare across measurement approaches (serum, adipose, and ASR). This study will inform evaluating and measuring POPs in relation to endometriosis.

This study found that mixtures of adipose and ASR OCPs and estrogenic PCBs were associated with incident endometriosis diagnosis. This was the first study to our knowledge to apply BKMR to evaluate POP mixtures and endometriosis. For single chemical models, the strongest associations with endometriosis were for the ASR. Notably, the ASR was not appreciably more informative than adipose. This may provide context for studies which found no associations between serum or plasma POPs and endometriosis. Several ASR POPs were associated with a greater odds of an endometriosis diagnosis, including for organochlorine pesticides β-HCH and ɣ-HCH, and for PCB congeners 49 and 201. Our use of the ASR as a measure of POP exposure is novel but our findings were largely consistent between adipose and ASR. This suggests that although associations with endometriosis were slightly stronger compared to POPs in adipose, the ASR does not provide greater insight into the influence of POPs on endometriosis. Our sample size of 339 women may not be sufficiently large for examining differences between adipose and ASR, particularly for flexible BKMR models. The ASRs for PCB congeners 49 and 201 were associated with endometriosis. Other estrogenic PCB congeners were similarly associated, though associations were smaller in magnitude. Previous studies of PCBs have been equivocal and there are no other studies of ASR chemicals in relation to endometriosis with which to compare our findings. A case-control study found no difference in sum fasting serum PCB and peritoneal endometriosis, but was limited by small sample size with 7 endometriosis cases ( Heilier et al., 2004 ). Anti-estrogenic PCBs in serum were associated with endometriosis but individual PCBs were not associated, nor were estrogenic PCBs ( Louis et al., 2005 ). However, ASR anti-estrogenic PCBs were not associated with endometriosis in our study. We did not find that the ASR for PBDEs was associated with endometriosis. Smaller ASRs for lower brominated PBDE congeners were observed among women with than without endometriosis. PBDEs are a group of flame-retardant chemicals which are added to plastic and foam consumer products and the primary route of human exposure is believed to be via indoor dust from product degradation ( Johnson-Restrepo and Kannan, 2009 ). Although PBDEs have higher K ow than PCBs, they do not bioaccumulate in comparison to PCBs ( Johnson-Restrepo et al., 2005 ), indicating that PBDEs are metabolized to some extent, which may help explain the lack of association we observed. Our findings that the ASR for β-HCH and ɣ-HCH were associated with an increased odds of an endometriosis diagnosis are supported by prior studies, albeit in adipose tissue or serum, rather than together as a ratio. In fat, ɣ-HCH was associated with an increased odds of endometriosis, while in serum, β-HCH was associated with an increased odds of endometriosis ( Buck Louis et al., 2012 ). Similarly, β-HCH was associated with endometriosis in a case control study ( Upson et al., 2013 ). Associations for oxychlordane were positive and greater for ASR compared to adipose and serum levels. Oxychlordane was not associated with endometriosis in adipose or serum in this cohort ( Buck Louis et al., 2012 ). Oxychlordane levels in serum were not different by endometriosis diagnosis in a case control study ( Lebel et al., 1998 ). Oxychlordane is an organochlorine pesticide metabolite of cis- and trans-chlordane and cis- and trans-nonachlor. Oxychlordane is more toxic than its parent compounds ( “ATSDR - Toxicological Profile,” n.d. ). Several other studies reported adipose tissue to serum ratios which varied considerably, highlighting that there is not a uniform relationship between chemicals in serum and adipose tissue. An example of p,p’ -DDE illustrates the extent of variability between adipose tissue and serum. In our study, the median p,p’ -DDE ASR was 754.3 (IQR 1063.6). In contrast, a ratio of 1.1 for p,p’ -DDE was reported in a Mexican population ( López-Carrillo et al., 1999 ) while a ratio of 115.8 was identified among a US population ( Stellman et al., 1998 ) and higher wet weight ratios were found for healthy volunteers ( Mussalo-Rauhamaa, 1991 ). These studies collected biospecimens 20–30 years ago, and demonstrate that POP dynamics for a single chemical metabolite vary considerably. Taken together, this suggests that specific chemical partitioning depends on more than chronological time of collection in relation to the proximity of use of legacy POPs. Additional factors such as geographic location, health status, and age of study participants play a role. Inconsistent correlations between serum and adipose concentrations suggests that blood measurements of exposure do not adequately portray adipose concentrations ( Aronson et al., 2000 ), as these measurements were often made at non-equilibrium conditions. Concentrations of specific POPs in serum and adipose tissue may have different biological implications ( Artacho-Cordón et al., 2015 ). This underscores that conceptualizing the relationship between an adipose storage depot and a circulating serum proportion depends on many factors. The biological mechanism by which POPs may affect endometriosis development is unknown. The etiology of endometriosis is posited to reflect an interplay between endocrine, immune, and processes that promote inflammation and abnormal angiogenesis ( Zondervan et al., 2018 ). Endocrine system modification is plausible as POPs are endocrine disruptors, and human endometrium is a site of hormonal and angiogenic activity. Supporting this, total DDT, p,p’ -DDE, HCB, β-HCH, and PCB congeners 138, 153, and PCB180 were detected in human endometrium tissue ( Schaefer et al., 2000 ). Reduced endometrial progesterone receptivity has also been identified among women with endometriosis and induced experimentally following 2,3,7,8-tetrachlorodibenzo-p-dioxin exposure ( Igarashi et al., 2005 ). The halogenated diphenyl ether, 4-chlorodiphenyl ether, increased the survival of surgically induced endometriosis in mice ( Yang et al., 1997 ), likely be due to their estrogenic mechanism of action ( Meerts et al., 2001 ). Further studies are necessary to better understand PBDEs and endometriosis. Immune system modulation influences endometriosis development ( Herington et al., 2011 ). In particular, endometrial macrophages, immature dendritic cells and regulatory T cells exhibited different function between women with and without endometriosis ( Vallvé-Juanico et al., 2019 ). Several POPs can modify immune function ( Quaranta et al., 2006 ), indicating that may be a potential mechanism. Concentrations of POPs in serum and adipose tissue are dynamic and not constant, depending on their octanol-water partition coefficient (K ow )( Bourez et al., 2013 ) as well as person characteristics such as recent weight changes. We examined weight change in our study and our findings were robust to adjustment to such changes. We investigated weight change with effect modification by BMI because of the role they may play for health effects and POPs. It is accepted that adipose tissue reflects long-term POPs exposure, while serum levels are a mixture of current exposure combined with mobilization from stored adipose depots ( Archibeque-Engle et al., 1997 ). Adipose tissue serves as both storage and a source of POP exposure for other tissues ( La Merrill et al., 2013 ). However, weight loss can modify serum exposure ( De Roos et al., 2012 ) by increasing the release of POPs from adipose tissue ( Louis et al., 2014 ). Complicating the picture, POPs can lead to adipocyte dysfunction ( Howell and Mangum, 2011 ), as evident by findings that some POPs such as DDE may be obesogenic ( Cano-Sancho et al., 2017 ). Health status can further modify this balance by altering the lipid content of serum ( Guo et al., 1987 ). Therefore, prior work confirming consistent strong correlations between adipose tissue and serum concentrations of persistent chemicals ( Archibeque-Engle et al., 1997 , 1997 ; Kanja et al., 1992 ; López-Carrillo et al., 1999 ; Mussalo-Rauhamaa, 1991 ; Petreas et al., 2004 ; Rusiecki et al., 2005 ; Stellman et al., 1998 ), should be considered in context. Many of these studies collected biospecimens when legacy pollutants were in contemporary use, which could influence the relationship between serum and adipose tissue. Some studies comparing serum with adipose levels were restricted to women with breast cancer ( Artacho-Cordón et al., 2015 ; Petreas et al., 2004 ; Rusiecki et al., 2005 ) or were restricted to fecund women by design, collecting samples during Cesarean births ( Kanja et al., 1992 ). Depending on the underlying causal structure, if chemicals contributed to breast cancer or fecundity, then the relationship between serum and adipose tissue among these selected populations may not reflect such relationships, more broadly. Most importantly, as our data show, the relationship between ASR levels of POPs varies by chemical class and within chemical classes. Therefore, considering POPs as a group with a uniform relationship between biospecimens oversimplifies a dynamic system. Despite this study’s strength of incident endometriosis diagnosis and surgical confirmation, the exposures measured in serum and adipose tissue were obtained about two months prior to diagnosis, and may therefore not reflect exposures relevant for endometriosis development. Measuring exposure prior to endometriosis onset is challenging, particularly given the potential of an in utero origin for this condition ( Signorile et al., 2012 , 2010 ). Future studies should attempt to capture in utero exposure and follow individuals forward until incident endometriosis diagnosis. Despite the long half-life of POPs considered in this study, if the relevant time window was during fetal development rather than adulthood, this may lead to bias. Despite this limitation, we considered two biological compartments and the relationship of their ratio measurement with incident endometriosis, an innovation beyond considering either compartment separately. There is potential that our findings are subject to selection bias if exposure is associate with seeking care or the decision to undergo surgery. By design, all study participants underwent laparoscopic or laparotomic surgery in our study in part for the visualization of disease or its absence and to collect operative samples. Women without access to medical care are therefore not included in this study, which may limit generalizability of our findings. The original study addressed this limitation in that they also recruited a population cohort that was randomly sampled from the catchment areas of the hospitals from the operative cohort. A related issue is that of racial bias in the diagnosis of endometriosis, wherein Black women with pelvic pain are less often recommended for follow up care that would lead to an endometriosis diagnosis ( O. Bougie et al., 2019 ; Olga Bougie et al., 2019 ) than white women. This could potentially bias our findings, since race seems to be a predictor of exposure to POPs ( Ruiz et al., 2018 ). However, given the lack of a noninvasive diagnostic for endometriosis, it is challenging to study endometriosis without this potential selection bias.

The ASR for several POPs were associated with an increased odds of an incident endometriosis diagnosis including when using BKMR modelling techniques for ASR and adipose estrogenic PCBs and OCPs. Associations for ASR were generally similar to associations for adipose tissue, but differed from serum levels. Ultimately, measurements in eutopic and ectopic endometrium may provide even greater insight as they represent the target tissue.