This study was conducted in Guadeloupe (French West Indies), a Caribbean archipelago. The TIMOUN Mother-Child Cohort was established to investigate the consequences of prenatal (maternal or in utero) exposure to chlordecone on pregnancy and child development. From November 2004 to December 2007, 1,068 pregnant women from the general population were enrolled during their third-trimester prenatal visit at public health centers (University Hospital of Pointe-à-Pitre, General Hospital of Basse-Terre, and antenatal care units) (6). Questionnaires were administered at inclusion to assess their social, demographic, occupational, medical, and family characteristics, as well as lifestyle habits. At delivery, data concerning maternal diseases during pregnancy, delivery, and the health status and anthropometric characteristics of the newborn at birth were collected by the medical staff and maternal and cord blood samples were obtained. Follow-up visits of the children were conducted afterwards at various ages (3, 7, and 18 months) to evaluate the development of the children (9, 10, 33). When the children were seven years of age (May 2011 to October 2015), 1,033 mothers initially included in the cohort were contacted for a follow-up interview and a medical examination of their child. Among them, 589 (57% of the initial cohort) were interviewed and had their child examined at the University Hospital of Guadeloupe. During the medical examination of the children, face-to-face interviews were conducted with their mothers to collect information about the socio-economic context in which the child was growing and the health of the child, including medication intake. A blood sample was also collected from the children at the end of the examination. We excluded five children from the present study because of major congenital anomalies or severe diseases that could affect hormone levels. We then retained those for whom at least one blood hormone concentration was successfully measured at seven years of age, leaving 438 children (210 boys and 228 girls). Among them, a cord-blood chlordecone measurement was available for 285 (124 boys and 161 girls). No children were treated with hormonal or anti-hormonal medications.

The study was approved by the relevant ethical committee for studies involving human subjects (Comité de Protection des Personnes Sud-Ouest et Outremer III (n° 2011-AOOSSI-40). Each parent provided written informed consent.

Blood samples at birth (umbilical cord blood) and at seven years of age were collected in EDTA tubes and, after centrifugation, plasma samples were conserved in polypropylene Nunc^® tubes at -30°C. They were transferred on dry ice by airmail to the Center for Analytical Research and Technology (CART) at Liège University in Belgium for organochlorine analysis and to the steroid mass spectrometry platform of the Mondor Institute for Biomedical Research, Créteil, France, for hormone analysis.

Blood samples were analyzed for chlordecone, p,p´−dichlorodiphenyldichloroethylene (DDE, the major and most persistent metabolite of dichlorodiphenyltrichloroethane, DDT), and the non-dioxin-like polychlorinated biphenyl congener 153 (PCB-153) by high-resolution gas chromatography with Ni63 electron capture detection. Detailed information about the sampling, analysis, and quality assurance and control have been provided elsewhere (11, 34). Among PCBs, we selected PCB-153 because it correlates very well with the total PCB concentration in plasma (35). The analytical limit of detection (LD) was 0.06 μg/L for chlordecone in cord blood, 0.05 µg/L for DDE and PCB-153 in cord blood, and 0.02 μg/L for chlordecone in the children’s blood.

TSH, FT3, and FT4 were measured by immuno-radiometric assay (IM 3712 TSH Irma Kit, IM 1579 FT3, IM 1363 FT4, Beckman Coulter). Results were expressed as concentrations (milli-international units per liter for TSH and picomoles per milliliter for FT3 and FT4). The intra- and inter-assay coefficients of variation (CVs) were ≤3.7 and ≤8.6 for TSH, ≤6.4 and ≤5.5 for FT3, and ≤10.29 and ≤7.58% for FT4, respectively.

Insulin growth-factor 1 (IGF-1) was measured by radio-immunoassay assay (IGF1-RIACT, Cisbio Bioassays, Codolet, France) and concentrations are expressed in ng per milliliter. The intra-assay CV was ≤3.8 and the inter-assay CV ≤8.2%. Leptin was measured by enzyme-linked immuno-sorbent assay (BioVendor Human Leptin ELISA, Brno, Czech Republic) and concentrations are expressed in ng per milliliter. The intra-assay CV was ≤7.6 and the inter-assay CV ≤6.7%. Total adiponectin was measured by enzyme-linked immunosorbent assay (ALPCO, Salem, NH, USA) and concentrations are expressed in µg per milliliter. The intra-assay CV was ≤5.7 and the inter-assay CV ≤6.4%.

DHEA, TT, DHT, and E2 were assayed simultaneously by gas chromatography-mass spectrometry, as previously described (36). Briefly, deuterated steroid internal standards (CDN Isotopes, Inc., Point- Claire, Quebec, Canada) were added to all plasma samples, which were then extracted with 1-chlorobutane. The organic extracts were purified on conditioned high-purity silica LC-Si SPE columns (Varian, Les Ulis, France). All steroids were derivatized with pentafluorobenzoyl chloride (77253-1ml, Sigma-Aldrich, Steinheim, Germany). The final extracts were reconstituted in isooctane and then transferred to conical vials for injection into the GC system (GC-2010 Plus, Shimadzu, Japan), equipped with a 50% phenylmethylpolysiloxane VF-17MS capillary column (20 m x0.15 mm, internal diameter, 0.15 mm film thickness; Agilent Technologies, Les Ulis, France). A TQ8050 (Shimadzu, Japan) triple quadrupole mass spectrometer equipped with a chemical ionization source and operating in Q3 single-ion monitoring mode was used for detection. Concentrations were reported for DHEA in nmol per liter and for DHT, TT, and E2 in pmol per liter. The intra- and inter-assay CVs were ≤3.5 and ≤4.7 for DHEA, ≤2.2 and ≤2.1 for TT, ≤3.0 and ≤3.1 for DHT, and ≤3.5 and ≤4.1% for E2, respectively.

Total cord plasma cholesterol and triglyceride concentrations were determined enzymatically (DiaSys Diagnostic Systems GmbH; Holzheim, Germany) and the total lipid concentration calculated as previously described (37).

All analyses were stratified by sex because of gender differences in hormone production and potential sexual dimorphism related to the effect of chlordecone. Continuous variables are described as means, medians, inter-quartile ranges, and percentiles. Mean ranks between unpaired groups were compared using the Mann Whitney test in descriptive bivariate analyses.

Cord-blood chlordecone concentrations were considered to be categorical (quartiles, based on their distribution in the population study) or continuous variables after log₁₀ transformation. Chlordecone values below the LOD were imputed by a maximum likelihood estimation method (38).

Associations between in utero chlordecone exposure and hormones with ≥90% detectable values (TSH, FT3, FT4, DHEA, IGF-1, and adiponectin for both sexes, and leptin for girls) were analyzed by multiple linear regression, allowing calculation of the β regression coefficient and its 95% confidence interval (95%CI). Hormones were included in the model after the imputation of values < LOD (38) and a log₁₀ transformation, as they were log-normally distributed. Hormones with 14.3 to 72.6% detectable values (DHT, TT, and E2 for both sexes, and leptin for boys; see Table 3 ) were dichotomized according to their LOD (< LOD vs ≥ LOD) and potential associations with in utero chlordecone exposure analyzed using multiple logistic regression models, allowing estimation of the odds ratio (OR) and its 95%CI.

The following maternal covariates were considered to be potential confounding factors: age at delivery (years), geographic origin (Caribbean vs European), body mass index (BMI, kg/m²), weight gain during pregnancy (insufficient, normal, excessive, or very excessive, according to the guidelines of the Institute of Medicine (39), education (< 12 years schooling vs ≥ 12 years schooling), smoking during pregnancy (never vs ever), and alcohol consumption during pregnancy (never vs ever). We also considered the following child covariates: exact age at examination (years), z-score height, z-score BMI, preterm birth (yes, no), small for gestational age (tenth percentile (40); yes, no), breastfeeding (yes, no), time of day when blood was drawn, and total cord-plasma lipid concentration (g/L). The characteristics of the samples were taken into account, as hemolysis was present in some (ranked by visual inspection as none or light, moderate, or strong) and could have affected the hormonal analysis. These covariates, including hemolysis, were included in all statistical models and then selected by applying a backward stepwise elimination procedure at P < 0.2. Missing values for each covariate were coded as an indicator variable to indicate missing. The Hosmer-Lemeshow goodness-of-fit test was used for the final logistic models.

We also implemented general additive models (GAM), including restricted cubic splines, to fit a potential non-linear association between in utero chlordecone exposure and hormone levels (41). Analyses were carried out using Stata 15.1 (College Station, Texas, USA) for linear and logistic regression modeling and SAS (SAS Institute Inc., North Carolina, USA) for GAM modeling. All tests were two-tailed, and P values <0.05 were considered statistically significant.

In the TIMOUN Mother-Child Cohort Study, we previously reported that cord-blood chlordecone concentrations were monotonously associated with increased TSH levels at three months of age in boys only, without modification of FT4 or FT3 levels (33). At seven years, we no longer observed such a profile in boys. However, we observed this profile in girls but with a non-monotonic dose-response pattern. In our population study, 7 (4.4%) of the 159 girls showed a slight increase in TSH levels to between 4.5 and 6.6 mIU/L, with strictly normal free thyroid hormone levels. The clinical importance of such a slight increase in plasma TSH levels (below 10 mIU/L) and the precise upper limit of the normal range for plasma TSH levels is still debated (45, 46). Although we cannot predict the clinical significance of our observations, the natural history of slight TSH elevations in healthy children who were not being drug treated show spontaneous normalization of TSH values (47). The biological mechanisms by which chlordecone could affect the thyroid axis are still unknown. A series of in vivo studies reported thyroid disruption in embryo and adult rare minnows exposed to chlordecone (48). However, complementary in vitro and in silico experiments showed only weak potency for the interaction of chlordecone with thyroid-related proteins (including thyroid receptors α and β) and suggested that thyroid alterations could be attributed to its interactions with ERs (48). Thus, the observed thyroid alterations in fish may have resulted from the well-recognized estrogenic activity of chlordecone. Whether a similar biological mechanism by which chlordecone could affect the thyroid axis in humans, and possibly differentially according to sex, is yet to be established.

We did not observe any association between in utero chlordecone exposure and metabolic hormone levels for either sex in our study population. To date, no experimental or toxicological studies have addressed the question of possible relationships between chlordecone exposure or its effects and metabolic hormones. During the follow-up of the TIMOUN Mother-Child Cohort at the age of seven years, we found no clear evidence supporting an adipogenic effect of in utero chlordecone exposure. Despite significantly higher adiposity in the third quartile of in utero chlordecone exposure (particularly in boys), we were unable to formally establish a significant non-linear trend (Costet N, Lafontaine A, Rouget F, Michineau L, Monfort C, Thomé JP, Kadhel P, Multigner L, Cordier S; unpublished data). Given the present results concerning metabolic hormones, it appears unlikely that these hormones mediate changes in adiposity, if any, in response to chlordecone exposure.

We observed increased levels of DHEA, TT, and DHT in children at seven years of age, but not E2, for the same third quartile of cord-blood chlordecone exposure relative to the lowest quartile. These hormones are located in the successive classical pathways of sex steroid production that include androstenedione and androstenediol, two alternative intermediate steps between DHEA and TT that we did not measure in the present study (29). Such relationships suggest that the increase in TT and DHT levels may result from an initial increase in the level of the substrate DHEA, consistent with the law of mass action, although increased enzyme activity in these pathways (3-β-hydroxy steroid dehydrogenase, 17-β- hydroxy steroid dehydrogenase, 5-α reductase) cannot be excluded. However, E2 levels were not modified, regardless of the level of in utero chlordecone exposure. Chlordecone is a recognized inhibitor of aromatase, the enzyme that converts TT to E2 (26). Thus, we cannot exclude the possibility that chlordecone-mediated inhibition of aromatase prevents increased E2 levels, despite an excess of TT as substrate. Finally, the origin of the increased levels of DHEA may be from any step upstream of cholesterol involving liver cytochrome P450 enzymes. Interestingly, experimental studies in rodents have shown that chlordecone induces cytochromes P450 enzymes (49) and may impair cholesterol homeostasis and tissue distribution (27, 50).

The non-monotonic dose-response (NMDR) we observed between in utero chlordecone exposure and certain hormones (inverted U-shaped curve) is characterized by associations at intermediate exposure concentrations and no association at lower and higher exposure concentrations. Such a relationship is not entirely unexpected, as it is recognized that certain EDCs may exhibit such patterns in experimental studies (51, 52). NMDRs can arise from numerous molecular mechanisms, such as opposing effects induced by multiple receptors differing in their affinity, receptor desensitization, negative feedback with increasing dose, or dose-dependent modulation of metabolism (53). In the TIMOUN Mother-Child Cohort Study, we previously observed a similar NMDR between in utero chlordecone exposure and birth weight in overweight and obese mothers (13).

The main strengths of this study lie in its prospective design, its being a population-based cohort study, and the exposure and outcome measurements. Prenatal (in utero) chlordecone exposure was determined using cord-blood samples, providing a representative measure of fetal exposure during the entire pregnancy, because the half-life of chlordecone in blood is approximately six months (54) and mothers were continuously exposed via the dietary intake of contaminated foods (6). Co-exposure to other EDCs was also considered, such as that to DDE and PCB-153, as well as childhood chlordecone exposure. We simultaneously measured a large number of hormones using standardized methods and gas chromatography–mass spectrometry for sex-steroid hormones, a method considered to be the gold standard for steroid hormone assays (55). Nevertheless, associations we observed in the present study, including the U-shaped relationships, should be interpreted with caution and we cannot exclude that it may result from a chance finding resulting from residual confounding or multiple comparisons.