Skip to main content

Advertisement

Public health and chronic low chlordecone exposure in Guadeloupe, Part 1: hazards, exposure-response functions, and exposures

Article metrics

Abstract

Background

Inhabitants of Guadeloupe are chronically exposed to low dose of chlordecone via local food. The corresponding health impacts have not been quantified. Nevertheless the public authority implemented an exposure reduction program in 2003. We develop methods for quantifying the health impacts of chlordecone and present the results in 2 articles: 1. hazard identification, exposure-response functions (ERF) and exposure in Guadeloupe, 2. Health impacts and benefits of exposure reduction. Here is the first article.

Methods

Relevant data are extracted from publications searched in Medline and Toxline. Available knowledges on mode of action and key-event hazards of chlordecone are used to identify effects of chlordecone that could occur at low dose. Then a linear ERF is derived for each possible effect. From epidemiological data, ERF is the delta relative risk (RR-1) divided by the corresponding delta exposure. From animal studies, ERF is the benchmark response (10 %) divided by the best benchmark dose modeled with BMDS2.4.0. Our goal is to obtain central values for the ERF slopes, applicable to typical human populations, rather than lower or upper bounds in the most sensitive species or sex.

Results

We derive ERFs for 3 possible effects at chronic low chlordecone dose: cancers, developmental impairment, and hepatotoxicity. Neurotoxicity in adults is also a possible effect at low dose but we lack quantitative data for the ERF derivation. A renal toxicity ERF is derived for comparison purpose. Two ERFs are based on epidemiological studies: prostate cancer in men aged >44y (0.0019 per μg/Lblood) and altered neurodevelopment in boys (−0.32 QIpoint per μg/Lcord-blood). Two are based on animal studies: liver cancer (2.69 per mg/kg/d), and renal dysfunction in women (0.0022 per mg/kg/d).

Conclusion

The methodological framework developed here yields ERFs for central risk estimates for non-genotoxic effects of chemicals; it is robust with regard to models used. This framework can be used generally to derive ERFs suitable for risk assessment and for cost-benefit analysis of public health decisions.

Background

The inhabitants of the French West Indies are chronically exposed to low doses of chlordecone via local food. The corresponding public health issues have not been satisfactorily quantified because of the lack of an appropriate risk assessment method. Nonetheless, several million euros have been invested every year since 2003 for the reduction of exposure in Martinique and Guadeloupe [13]. Official investigations had revealed that 1.3 % of adults in Guadeloupe were exposed above the “no effect threshold dose” (the reference dose RfD) of 0.5 μg/kg/d [4]. Exposure to chlordecone seems more common in children 3 to 5 years old than in adults, with up to 18.5 % above RfD in Martinique [5]. These estimates of ingestion exposures are based on food consumption survey data matched with the results of chlordecone measurements in food made for the screening and control program. The main activities of the control program are to withdraw food from the market and water from the distribution system when chlordecone is above the limit values [6]. However, some polluted food from individual gardens could be eaten by the owners or the relatives and friends because they escape the market control system.

The RfD is based on a chronic animal study that observed an increase in kidney damage (glomerulosclerosis) in female Wistar rats [7]. However, monitoring of poisoned workers manufacturing chlordecone in the USA (Hopewell and Baltimore) revealed no renal function impairment [8, 9]. In addition, a recent study showed that only mouse strains predisposed to auto immune diseases developed kidney lesions after ingesting chlordecone [10]. Accordingly, the hazards of chlordecone in people with exposure above this RfD are uncertain, and the results of exposure studies are difficult to translate into public health decisions. In particular, they do not provide answers to the following questions:

  • What effect will occur in persons exposed above RfD: that observed at the lowest dose tested in animals, other effects known at some higher doses, or even the effect of another ubiquitous pollutant that is enhanced by chlordecone?

  • If exposure exceeds the RfD what is the likelihood that an effect occurs?

  • Are we sure that there will be no effect from exposures below RfD?

Several epidemiological studies have recently been carried out in Guadeloupe. The results show that chronic low exposure to chlordecone, mainly below the value of the RfD, is correlated with an increased risk of prostate cancer in men over 45 years [11] and with impaired neurobehavioral development in the young child [12, 13]. Other effects were investigated but were not significantly associated with low exposure to chlordecone: change in sperm quality [14, 15], and the rate of circulating hormones in men [16]. No studies looked at possible kidney damage, the critical effect found in laboratory animals [17, 18] nor at liver cancer which is significantly increased in a chronic animal study by the NCI [19]. The latter is used by the US-EPA to develop a cancer slope factor. In epidemiological studies exposure was estimated via blood chlordecone concentration (Cl-b).

Another concern for chlordecone health risk assessment is that we do not know sufficiently well the correspondence between Cl-b and the dose by ingestion expressed as mg/kgBW. First, an ad hoc study in Guadeloupe published in 2010 found that exposure data based on food consumption combined with food concentration of chlordecone are poorly correlated (R2 = 0.20) with blood chlordecone [20]. Second, there was no PBPK (physiologically based pharmacokinetic) model available at the time of the literature search that could help to convert a dose by ingestion into a blood concentration of chlordecone.

Health risk assessment cannot be quantitative without a quantitative relationship between exposure and response. They are commonly available for carcinogenic effects but not for other effects that have a non-genotoxic mode of action. To quantify the public health impacts of chlordecone in Guadeloupe exposure-response functions (ERFs) are necessary. Following the recommendations of the Silver Book [21], we assume that non-genotoxic effects can occur at low doses if the mode of action has been found to be active at low doses and is shared by a frequent human disease or a ubiquitous chemical. In that case exposure-response functions, similar to those for genotoxic carcinogens, can be derived for non-genetoxic effects. Our objective is to develop a methodological framework based on: 1. the selection of potential hazards from low dose exposures; 2. deriving ERFs from human or animal data; 3. evaluation of health and economic impacts; 4. compare costs of the exposure control program with public health gains resulting from the decrease in population exposure. Because of their length, methods and results are described in 2 separate articles. In this first article, the objectives are: to identify and select the hazard of low dose chlordecone to derive the corresponding ERF and to find population exposure data that could fit requirements for risk assessment. The second article focuses on the quantitative assessment of the risks, impacts and costs. The overall objective is to carry out a cost-benefit analysis of the exposure control program that has been implemented in Guadeloupe since 2003, i.e. compare its cost with the benefit of reduced health impacts. Since the IOM report in 1981 [22] it is widely accepted that the cost of environment-related health effects should be consistently evaluated for decision making in public health.

Methods

Document search

In April 2013, we searched publications on pharmacokinetics, mechanisms of action and toxicity of chlordecone in MEDLINE and TOXLINE with the following key words: “Chlordecone” OR “Kepone” OR “143-50-0” without any other limitation. Some French scientific revues that are not indexed in MEDLINE were also screened. We excluded articles that are:

  1. 1.

    Abstracts only or conference proceedings

  2. 2.

    Editorials

  3. 3.

    Review articles on general topics (pesticides, organochlorides, reproductive toxins, etc.)

  4. 4.

    Anonymous references (no author name)

  5. 5.

    Reference duplicates

  6. 6.

    Eco-toxicity studies

  7. 7.

    The development of chemical or biochemical tests studies

  8. 8.

    Studies only of mirex, photomirex or kelevan

Based on information in the title and abstract, studies were classified into categories of interest for health risk assessment: Pharmacokinetics (absorption, distribution, metabolism, excretion); Mechanisms of action; Toxic Effects; Epidemiological Studies; Reviews and various.

Chronic low dose hazard identification

We use 2 definitions of “low dose”. When analyzing mode of action (MOA) for key-event identification in mechanistic studies, the “low dose” hazard is: an adverse biological change occurring in the range of typical human exposures or at doses lower than what has been used in standard toxicology testing protocols [23]. Mechanistic studies that explore the key event for a MOA are not chronic and mostly not low dose, because one needs dose levels high enough to produce the effect. Our objective is to assess risk from chronic exposure to low dose of chlordecone, because this is the situation in Guadeloupe. The gap between the available knowledge and the information needs for decision making is very frequent in environmental health. Here, to decide if an adverse effect can result from low dose exposure the available knowledge comes from acute exposure studies and there is no alternative. First we identify the key-events of a MOA in mechanistic studies, and then if the data meet our criteria (see below) we assume the effect resulting from the key-events to be possible at chronic low-dose.

The first human data on adverse effects of chlordecone come from a cohort of workers poisoned in a chlordecone factory because of poor industrial hygiene. Exposures were very high (above 1 mg/lblood) and caused poisoning cases ranging in severity from a mild illness to a severe, totally disabling infirmity [24, 25]. Then toxicologists tried to understand the adverse effects observed among the workers. Hence, chlordecone has undergone many toxicological studies in vitro or in vivo using mostly relatively high doses (>10 mg/kg/d). These studies are old and almost entirely devoted to acute and subchronic effects. However, many mechanistic studies have identified the role of chlordecone in endocrine disruption, liver enzyme induction and unbalanced cellular energy. The analysis of available knowledge on the mode of action (MOA) aims to identify the effects of chlordecone that can occur at low doses, effects for which we will derive ERFs. After verifying consistency between the human and animal data or consistency between the animal species tested, the effects whose MOA meets at least 2 of the following 3 criteria will be selected as potentially effective at low dose:

  1. 1.

    The key-event for a MOA is observed at concentration ≤ 1 μM or dose < 1 mg/kg/d

  2. 2.

    The MOA is identical to that of a frequent human disease

  3. 3.

    A ubiquitous chemical follows the same MOA

For chlordecone we have chosen the low dose values based on an inventory of the range of doses used in toxicological studies. Most of them are between 10 and 40 mg/kg/d, for acute effect testing. As mechanistic studies are often based on one dosage, we consider that less than 1 mg/kg/d is a low dose for acute chlordecone exposure. The corresponding concentration is approximatively 0.1 μM based on information reported by End et al. 1979 [26]. They wrote that a brain concentration of 2 μM corresponded to an external dose of 40 mg/kg. This is a 0.05 μM per mg/kg that we rounded to 0.1 μM taking into account the fact that the brain is not the most impacted organ after a 40 mg/kg dose (see Figure I in Additional file 1).

Deriving ERFs

Having identified via MOA analysis the possible effects at low dose, we derive the ERFs based on quantitative data from either human or animal studies. Studies used to derive an ERF must meet all the following criteria:

  1. 1.

    Epidemiological studies (cohort or case–control only); or animal studies following the OECD or US-EPA guidelines, with at least 2 dose groups in addition to the control group;

  2. 2.

    Study outcomes corresponding to one of the effects identified as occurring at low dose;

  3. 3.

    Quantitative measurement of chlordecone exposure (biomarkers or external doses);

  4. 4.

    Exposure duration ≥ 1 year or exposure in utero + post natal;

  5. 5.

    Medical or histopathological diagnosis of the disease;

  6. 6.

    Statistical tests of risk indicators;

  7. 7.

    Cofactor control in epidemiological studies (smoking, weight, age, gender, income level, other exposures, etc.).

Our study limited the ERF derivation to chronic exposure studies. In animal studies, this criterion corresponds to a length of at least 1 year of day to day consecutive exposure. Developmental studies need other criteria because in rodents developmental period is less than 1 year. Exposure only in utero or only post-natal will not adequately reflect the exposure of a new born and infant. It is why we selected only studies where exposures happen during gestation and early stage of life (postnal).

ERF Derivation methods

In regulatory risk assessment, the objective is to protect the population from exposures that can lead to unacceptable risks. Here, we seek to know what are the actual risks in the French Indies population exposed to chlordecone. Accordingly, we are interested not only in the critical effect (generally defined as the first one to appear in the dose range tested) but all effects at low dose. There will be as many ERFs as effects of chlordecone at low dose for which there exist quantitative human or animal data. Then we are looking for representative average data and not just the data about the most sensitive specie or sex. If an effect is significant in both sexes or in both species, the ERF will be calculated as an average risk of both sexes or both species. However, if the effect is significant in only one of the 2 sexes, the ERF will be derived only from data of that sex and will apply in the risk assessment to persons of the corresponding sex. From animal studies as well as from epidemiological studies the derivation of ERFs involves 2 steps: choose a point of departure (POD) then draw a straight line from the POD to the origin [27].

ERFs based on epidemiological studies

When deriving ERFs from epidemiological results, attention must be paid to causality [28]. Greenland indicated that, even with a systematic analysis of the well-known Hill’s criteria [29], causality will have little support or plausibility if the mode of action is unknown. As our framework starts with a systematic analysis of MOA, ERFs based on epidemiological will only be established for effects with known MOA.

Some epidemiological studies give an ERF directly. Otherwise, the ERF is estimated with the data presented in the study. The POD for the ERF is the risk indicator: relative risk (RR) or odds ratios. Extrapolation of this POD towards the origin is to divide the difference in risk (RR-1) by the exposure difference between the exposed group and the reference group. If an epidemiological study gives risk indicators for more than 2 groups (exposed/not exposed), use the risk indicator of the first group statistically different from the reference group. This rule is made to prevent obvious uncertainty propagation and is discussed with results on prostate cancer. For cancer effects, we assume that the probability of response is proportional to the duration of the exposure. In this case, the average (or median if average is not available) length of exposure in the study is used to standardize the ERF for a standard human lifetime (i.e. 70 years). The ERF is calculated with the following equation:

$$ \mathbf{E}\mathbf{R}\mathbf{F} = \left(\mathbf{R}{\mathbf{R}}_{\mathbf{a}} - \mathbf{1}\right)\ /\ \mathbf{a} \times T W $$
(1)

ERF = Exposure-response function (lifetime) expressed as invers of unit exposure

a = exposure difference between the reference group and the RR exposed group (in μg/L)

RRa = relative risk related to the exposure difference “a”

TW = time weighted factor = standard human lifetime (70y)/mean duration exposure in the study (used only for cancer effects).

ERF from animal studies

Experimental animal data are modeled as benchmark dose (BMD) [30]. A BMD is a dose corresponding to a specified probability of occurrence (the risk) of the adverse effect (the hazard), also called benchmark response (BMR). Many models (or functions) can be fit to toxicological data. The US-EPA has developed software that simultaneously estimates several models from study data. We use the BMDS v2.4.0 software for BMD determination from animal data. This software is freely available on the website of the US-EPA [31].

From animal data, the POD will be the best BMD. The best BMD is the one from the model with the best fit to the data, or, when statistical tests cannot decide between different models, the geometric mean of the respective BMDs. US EPA, in order to protect the population, uses the lower limit of the confidence interval at 95 % around the BMD (Noted BMDL) as POD. In our work the objective is to obtain the best estimate of the impacts and costs, so we have to think in terms of expected value and not in terms of safety margin. Therefore, we will use the central value of the BMD.

Statistical tests before modeling the data

Original data from animal studies are organized into a data set which consists of 3 items: administered doses (e.g.: 0, 5, 10, 25, 50 mg/kg/d); number of animals per dose group; number of sick animals in each group. A data set is defined by gender, species and effect. Each data set is subjected to statistical tests before modeling. The first test, the Cochran Armitage trend test, verifies that the data set shows a significant trend of increasing incidence with increasing doses. It is performed with BMDS2.4.0 software. Second, the incidence rates of each dose group are tested against the incidence in the control group. The evaluation of the χ 2 test is performed with the R software [32]. If the unilateral trend test (incidence should increase with dose) does not reject the null hypothesis (p > 0.025 or Z <1.96), there is no significant increased trend and the data set is excluded. A dose group not significantly different from the control group is excluded from modeling. After exclusion of one or more dose groups, the data set is modeled if there remain at least 2 dose groups other than the reference group, otherwise it is excluded.

Benchmark dose modeling (BMD)

Briefly, in the BMDS2.4.0 software the models available for a dichotomous endpoint (quantal endpoint) are: Gamma*, Logistic, Log-Logistic*, Log-Probit*, Multistage*, Multistage-cancer, Probit, Weibull* Quantal-linear. The models marked with a star can be constrained by either power ≥ 1 (Gamma and Weibull) or slope ≥ 1 (Log-Logistic and log-probit) or number of polynomial coefficient ≥ 0 (Multistage). The Multistage-cancer model is available with 1, 2 or 3 stages. The polynomial coefficients of the Multistage-cancer model are always constrained to be ≥ 0. Thus, they are 16 models available. We do not give more details on the equations and parameters of these models because they are fully described in the user guide of the BMDS2.4.0 software [31].

BMDs can be computed relative to the background incidence as a Relative Risk (noted “extra risk” in BMDS software) or independent of the background incidence as an Absolute Risk (noted “added risk” in BMDS software). But since the background incidence in animals cannot be considered representative for humans, the BMDs will be computed as Absolute Risk, so that ERF can be used in any human population without knowing the background incidence. Each data set will be computed with the 16 models, and the best is selected according to the statistical tests described above.

Choice of risk level (BMR)

As a general rule, the BMR (the probability of an adverse effect) should be selected close to the incidence in the first dose group so that modeling does not exceed too much the range of observations. A default 10 % BMR is recommended by US EPA for conventional animal data, because usually that corresponds to the power of studies with 20–50 animals per group. Here, we will use in all cases a 10 % BMR because this will facilitate comparaison with other BMDs derived by Public Agencies.

Dosimetric adjustments

After modeling, animal BMDs are converted to human BMD (BMDHED) with a dosimetric adjustment (DAF), following the recommended use of “Body Weight3/4” as the default method [33]:

$$ \mathbf{B}\mathbf{M}{\mathbf{D}}_{\mathbf{HED}} = \mathbf{B}\mathbf{M}\mathbf{D} \times \mathbf{D}\mathbf{A}\mathbf{F} $$
(2)

DAF : Dosimetric Adjustment Factor for oral exposure (−) = (BWA/BWH)1/4

BWA = Animal body weight (kg)

PCH = human body weight (kg)

Point of departure (POD)

The point of departure is the best BMD10-HED (BMDHED for a 10 % BMR). Out of the 16 model results, we exclude results that are significantly different from the data set: Goodness of Fit test with p <0.1; residual value near the BMR is higher than │2│; those that give aberrant BMD (usually infinitesimal); those for which the calculation of statistical tests failed; and those which have more parameters than the number of dose groups modeled. In quantal models, often a background parameter quantifies the probability that the outcome being modeled can occur in the absence of exposure. This information comes from the zero dose groups, so these groups are included in the number of groups taken into account. If several models remain acceptable after this procedure, a further test is applied: if the ratio between the lowest and the highest BMD is less than or equal to 3, then the POD is equal to the geometric mean of acceptable BMDs; if the ratio is greater than 3, then the POD is equal to the BMD having the smallest value of AIC (Akaike's Information Criterion). The AIC test refers to the principle of parsimony: the smallest value of AIC means, in a group of models, the one that best fits with the least possible parameters.

Exposure-response function derivation

The ERF is the risk (here the BMR) divided by the point of departure (here the BMD10-HED):

$$ \mathbf{ERF} = \mathbf{BM}{\mathbf{R}}_{10}/\mathbf{POD} $$
(3)

ERF = Exposure-response function expressed as (mg/kg/d)−1.

BMR10 = Benchmark response of 10 % (−)

POD = BMD10-HED = Benchmark dose human equivalent for a BMR of 10 % (mg/kg/d)

Assuming that the probability of cancer at a constant dose increases linearly with exposure duration, a time weighting is applied to ERFs for carcinogenic effects if the exposure duration in the original study is not equal to the standard default lifespan of the animal species tested. Standard lifespans for humans, rats and mice, are respectively 70 years, 2 years and 2 years [34]. The weighting is to multiply the ERF by the ratio of the standard lifespan and the duration of exposure in the study.

Grouping data sets

Because grouping data sets can strengthen the statistical power and robustness of the derived BMD it could be interesting to group data sets before modeling them if they meet some statistical and biological criteria. Unfortunately there was no possibility for grouping data sets with the available data on chlordecone. So this point will not be detailed here. Interested readers can find description of the methods and discussions on advantage and weakness elsewhere [35].

Exposure data from the Guadeloupe population

Two approaches have been used in the past to assess exposures to chlordecone in the Guadeloupe population: biomarker measurement (called here “blood chlordecone”) and exposure estimated by food consumption combined with concentrations of chlordecone in food (called here “dose by ingestion”). An ad hoc study published in 2010 found that dose by ingestion in Guadeloupe is poorly correlated (R2 = 0.20) with blood chlordecone [20]. Moreover, for the comparative analysis of health costs and expenditures for preventive actions (see article 2) we need data spaced over time to estimate the exposure difference attributable to the exposure reduction program. The chronology of the public actions for reducing exposure was reported in a recent study [36]. Accordingly, one can consider that the exposures measured until 2003 describe the situation before the reduction program. Those measured since 2004 are influenced by this program [13].

Due to the specific persistence of chlordecone in blood, the best indicator of chronic exposure appears to be the blood chlordecone concentration (expressed in μg/Lblood). Furthermore there are no ingestion exposure data before 2005, which prevents estimating the ingestion exposure difference due to the program. Therefore we sought all studies that have measured blood chlordecone concentrations in Guadeloupe.

Substitution method for censored data

Where measurements are below the limit of detection (LD), the maximum likelihood estimation method (MLE) is often considered the gold standard for substituting values for the censored data. Ganser and Hewett published a simplified method called “β-substitution” giving very similar results, sometimes better than the MLE method and always better than the simplified DL/2 or DL/√2 methods [37]. Here, the mean exposure of groups that are below the LD will be estimated by using the β-substitution method from Ganser and Hewett.

Data preparation

Distributions that are not normal can be described by the value of quartiles or other percentiles deemed appropriate. With quartiles, the population is divided into 4 groups: those whose exposure is between zero and the 25th percentile (P25), then those between P25 and P50, those between P50 and P75 and finally those from P75 to the maximum value. When a sample includes censored data due to the LD, the first group is divided in 2 parts: those whose exposures are somewhere between 0 and LD and those between LD and P25. For this particular group average value will be calculated according to β-substitution method called “βM”. Finally, the average exposure in each group shall be calculated as follows:

1st group (results < LD), exposure equals the mean β-substitution (βM)

2nd group (from βM to P25), exposure equals the geometric mean of βM and P25

3rd group (P25-P50), exposure equals the mean of P25 and P50

4th group (P50-P75), exposure equals the mean of P50 and P75

5th group (P75-max), exposure equals the geometric mean of P75 and max.

If the proportion of results < LD exceeds 25 %, then the second group takes the average value of βM and P50 and there are only 4 exposure groups. If the proportion of results < LD exceeds 50 %, then the second group takes the average value of βM and P75 and there are only 3 exposure groups.

Results

Available studies on the toxicity of chlordecone for low dose hazard identification (mode of action and key-events) are numerous but generally old. These studies were analyzed in depth, and the details are reported elsewhere [35]. Here we provide a summary of the main results for MOA and key events. The studies were mainly focused on the acute toxic effects observed in workers poisoned in the 70s because of poor industrial hygiene in a chlordecone factory (Hopewell, USA) [24, 25, 3841].

Chlordecone is a known endocrine disruptor in humans with low agonist affinity for estrogen receptor ER-α [4249]. There are numerous studies on estrogen affinity in other species. They are less relevant and not different from those in humans. However, its adverse effects involve mechanisms partially comparable to estradiol. Chlordecone follows other toxicological pathways, mainly:

  • induction of hepatic cytochrome [5069].

  • inhibition of ATPase in the liver [5255, 69].

  • inhibition of ATPase in the brain [7082].

  • perturbation of neurotransmitters such as dopamine, epinephrine, norepinephrine, serotonin, beta-endorphin, enkephalin and gamma-aminobutyric acid [50, 83100].

  • leakage of intracellular calcium [26, 101106].

The US EPA and the IARC considered chlordecone non-genotoxic. Some epigenetic mechanisms of chlordecone carcinogenicity have been tested on human cell culture. It has proved capability to: inhibit arginine methylation [107], inhibit aromatase activity [108], cause DNA single strand breaks [109], limit intercellular communication [110, 111] and increase the activity of certain protein kinases (MAPK, PI3-K) [112]. Most of these effects are also observed in laboratory animals [113116]. Finally, a recent study showed the ability of chlordecone to promote angiogenesis and neovascularization via activation of ER-α at low doses [117].

Our in depth analysis of MOA and key-events led us to consider 4 non-genotoxic effects of chlordecone potentially effective at low dose:

  • increased mostly hormone-dependent cancers (prostate and breast),

  • hepatotoxicity,

  • neurotoxicity,

  • and developmental impairment (brain and probably heart),

For cancer promotion, the MOA effective at low doses are aromatase inhibition [108], disruption of intercellular communication [116, 118], and the increase of angiogenesis [117]. The ubiquitous substances that share the same MOA are bisphenol A, DDT, DDE and lindane. Regarding hepatotoxicity, the MOA is the induction of cytochrome P450 [60, 62], which can also led to cancer development via reactive oxygen species production. Induction of CYP450 is a reversible effect but it is shared by numerous chemicals ubiquitous in the human environment: 2,3,7,8-TCDD, PCBs and lead. Regarding neurotoxicity in adulthood, the key-event observed at low doses is a decrease of ATPase activity [73, 76, 80, 81]. This can be related to human diseases, for example lower brain ATPase activity in the elderly may contribute to the onset or worsening of neurodegenerative diseases such as Parkinson’s disease or Alzheimer’s. This MOA is shared by acrylamide and cadmium and other ubiquitous metals. Other key-events were studied for neurotoxicity in adulthood, like cellular calcium leakage or perturbation of neurotransmitters (see references above) but unfortunately not studied at low doses. Moreover, we made the hypothesis that if those key-events (decreased ATPase cellular, calcium leakage and some specific endocrine disruption) occurred in utero and during early months of life, the central nervous system development could be impacted leading to cognitive impairment. These key-events have not been studied in “developmental studies”. In vivo developmental studies involved in low dose neurotoxicity looked mostly at altered stress response [119122] that is less relevant in human. Our hypothesis is reinforced with the fact that other developmental neurotoxic substances (i.e. heavy metals) that impair cognitive ability of children, share some of the same cellular key-events. At least, authors of the TIMOUN cohort study (description of this study appears in next section) have postulated the same hypothesis, mostly based on endocrine disruption, and found inverse correlation between chlordecone blood concentration of mothers during pregnancy and decrease in neurological performance in boys at 18 months of age but not in girls (this discrepancy could confirm that an endocrine mechanism is involved).

Nephrotoxicity is not among possible low dose effects according to our analysis of the MOA, mainly because the underlying mechanisms remain unclear (Is the chlordecone responsible for triggering autoimmune diseases that will subsequently impact the kidneys? Or it has a direct toxicity to the kidney glomeruli?), there are inconsistencies between humans and animals and inconsistencies between different rodent species or strains. However, a doubt remains as to the role of chlordecone in promoting autoimmune diseases like systemic lupus erythematosus. Kidney damage is the critical effect chosen by several health safety agencies [17, 18, 123]. All have derived a reference dose from data of the Larson et al. study [7]. In order to compare the risk assessment of this effect with others, we will derive an ERF for kidney damages following the same methods.

For deriving ERFs we found quantitative studies (epidemiological and toxicological) for only 3 out of the 4 possible low-dose effects of chlordecone: cancer promotion, liver toxicity, and developmental impairment. Neurotoxicity in adults is a possible low dose effect of chlordecone but for which no ERF can be derived up to now (December 2015) for lack of a quantitative study.

In humans, studies of workers poisoned in the Hopewell factory are not suitable for deriving chronic ERFs because the exposure time could have been less than one year; moreover, all neurological, hepatic, or fertility symptoms disappeared some weeks after exposure cessation [24, 25, 3841]. There are 2 recent epidemiological studies in the French Antilles with exposure time longer than 1 year and eligible effects. The first is a case control study that found a statistically significant dose-dependent association between risk of prostate cancer and blood levels of chlordecone [11]. The second is a cohort of mothers and children followed from birth (cohort TIMOUN). Published in 2012, the first results have identified signs of neurotoxicity in children 7 months after birth correlated with chlordecone levels in cord blood and breast milk [12]. New results published in 2013 confirmed the previous results 18 months after birth and with more detail. Alteration of neurobehavioral development at 18 months is characterized by a decrease in fine motor skills statistically significant only in boys [13].

In animals 3 chronic (>1 year) and multi dose studies meet our criteria [7, 19, 124]. The NCI study shows a significant increase in liver cancers (carcinomas) in rats (Osborne-Mendel) and mice (B6C3F1) [19]. It is of intermediate quality because the initial doses, due to their excessive toxicity, were reduced during the course of the experiment. In addition, the control groups consisted of 10–20 animals against 50 in the dose groups. This weakness was offset by the addition of control animals from other experiments in the same laboratory during the same period. The doses were subject to time weighted averaging, which is acceptable from a mathematical point of view but maybe not from a biological point of view. The Reuber study [124] looked at liver cancers in albino rats (unspecified strain). Published data are incomplete and far from scientific quality standards. Nevertheless, the results could be modeled for comparison. The Larson et al. study [7] reports several effects including hepatotoxicity. Six dose groups were tested (1, 5, 10, 25, 50, 80 ppm), but in the last 2 groups all animals died during the first 6 months (unspecified cause of death). The dose group at 1 ppm is part of a separate experiment. There are also high rates of missing animals at the end of the study, ranging from 30 % in the controls to over 80 % in the group of 25 ppm. The causes of these “disappearances” are not reported. Despite these weaknesses, the study data could be modeled because its scientific qualities were considered sufficient to support a reference dose derivation by many health safety agencies [17, 18, 125].

Available data for exposure characterization come directly from epidemiological studies conducted in Guadeloupe since 1999. They provide some distributions of the blood chlordecone concentration in the sampled population. The first one, called INSERM, is a cross-sectional study held between 1999 and 2001 with 2 different sample populations: agricultural workers (n = 42) and non-agricultural workers (n = 45). It aimed to measure the exposure of men to chlordecone and to study its impact on male fertility [15]. We will use only non-agricultural worker data because the exposures of the agriculture workers were much too high to be representative of the general population. The second study, called HIBISCUS and conducted in 2003, aimed to investigate the cross sectional prevalence of exposure to chlordecone in pregnant women (n = 112) and neonates (n = 109) [126]. The third, called cohort TIMOUN and carried out between 2004 and 2007, aimed to assess the impact of the exposure on the course of pregnancy (n = 371) and on the development of the child (n = 265) [20]. The fourth, called KARUPROSTATE and carried out between 2004 and 2007, is a case–control study (n = 623/671). It aimed to investigate the relationship between the risk of prostate cancer and exposure to chlordecone in men aged over 44 years [11]. The quartiles of measured Cl-b levels in these studies were reported in a recent review article [127].

The ERFs derived from the available epidemiological or experimental studies are presented in Table 1. The benchmark dose model results are shown in Tables 3, 4 and 5.

Table 1 Exposure response function (ERF) and main information taken into account for derivation

For prostate cancer, the KARUPROSTATE study includes 623 cases (new cases of prostate cancer diagnosed in Guadeloupe between 2004 and 2007) and 671 controls. Exposures are measured by Cl-b (LD = 0.25 μg/Lblood). The authors found an increased risk of cancer significantly correlated with the increase in current Cl-b (p trend = 0.002). The odds ratio is only significant in the most highly exposed group. The Table 2 presents the ERFs of the 3 exposure groups, calculated using Equation 1. The ERF retained is that of the most exposed group: 0.114 (μg/Lblood)−1. Linear regression of the 3 ERFs gives a fairly similar slope factor of: 0.096 (μg/Lblood)−1 (cf. Fig. 1). The mean exposure duration was 33y, thus the ERF weighted for a standard lifespan exposure is 0.242 (μg/Lblood)−1. According to the cancer registry, the background incidence of prostate cancer in Guadeloupe is 259 per 100 000 (gross rates, all ages), based on an average 510 new cases a year [128]. Assuming there is no risk of prostate cancer before the age of 44 years and taking the number of men aged over 44 years (65000 in 2007, according to INSEE), the incidence rate is 770 per 100000 (510 cases/65000 men). Hence, the ERF based on relative risk can be converted to an absolute risk equivalent of 0.0019 (μg/L)−1 (=0.242 (μg/L)−1 × 0.0077). The latter applies to any population of men aged over 44 years.

Table 2 Prostate cancer risk increase and ERF calculation (data from Multigner, 2010 [11])
Fig. 1
figure1

Linear regression for the ERF from three exposure groups in the KARUPROSTE study. ERFs calculated with equation 1 for each exposure group in the KARUPROSTATE study are plotted as squares. Dashed line is the “best” regression (best R2) between delta exposure and delta risk (Power function: y = 0.2685x0.5664, R2 = 0.9919). The thick line is the linear regression (equation and R2 values shown to the left). The thin line is the graphical representation of the final ERF (equation and R2 values shown to the right) derived from the first exposure group that significantly differs from the reference group, i.e. the highest exposed group

For developmental impairment, in the prospective cohort TIMOUN cognitive ability scores were measured in 141 young children exposed to chlordecone during pregnancy and lactation. The exposures were measured at birth by chlordecone concentration in cord blood (μg/Lcord-blood) and 3 months later by the chlordecone concentration in breast milk. The latter is not used here. There are 30 % of censored data in cord blood measurements (LD = 0.06 μg/Lcord-blood). Cognitive abilities were measured at age 7 months [12] then at 18 months [13]. At 18 months, the authors used a standardized procedure to evaluate cognitive development named “Age and Stage Questionnaire” (ASQ). It includes 5 items: “personal-social”, “Communication”, “Problem solving”, “Fine motor skills”, and “Gross motor skills”. Only fine motor skills scores of boys were significantly and inversely correlated with the concentration of chlordecone in the cord blood. According to the authors, this is consistent with the available knowledge including the endocrine disruption mechanism. This article gives a slope coefficient of the multivariate linear regression of −0.32 points of fine motor skills per μgchlordecone/Lcord-blood. The relationship between the ASQ scores and IQ scores is not well known. For the purposes of our study we considered that an ASQ fine motor point equals an IQ point. According to our hypothesis the ERF would be −0.32 IQ point per μg/Lcord-blood and applies only to newborn males.

The ERF for liver cancer is based on animal studies. With the NCI data on hepatic carcinomas, we obtained 4 sets of data: female and male rats, female and male mice. There is no significant difference between the control group and the first dose group in male and female rats (see: Additional file 2: Table B). After excluding the first group in both datasets there remains only one dose group. This is insufficient for modeling, so those datasets are excluded. In B6C3F1 mice, all dose groups were significantly different from the control group and trend tests are significant. The grouping of males and females before modeling is not recommended because the incidence of hepatocellular carcinomas in the female control group is zero while it is 21 % in males (p <0.0006). Only one out of the 16 tested models fits the female data (unconstrained Multistage). Ten models differ significantly from the data (p <0.1). The results of Probit and Logistic models exhibit residuals greater than 2. Computation failed with 2 models: log-Probit, log-logistic. One model gives an outlier result: gamma (see Table 3). The BMD10 of the Multistage model is 0.274 mg/kg/day, and a BMD10-HED of 0.0412 mg/kg/d is calculated with a DAF of 0.15 (see DAF calculation in Additional file 2: Table A).

Table 3 BMD10 for liver carcinomas in female mice (from NCI, 1976)

In male mice, 4 models fit the data (Logistic, Multistage-cancer1, Probit, quantal-linear). Goodness of fit test was not computable with the other models (see Table 4). The ratio between the smallest and the largest BMD is 2.1. The geometric mean BMD10 is equal to 0.372 mg/kg/d and the BMD10-HED equal to 0.057 mg/kg/d (DAF = 0.152). The geometric mean of the 2 BMD10-HED (male mice and female mice) is 0.0483 mg/kg/d, and the ERF is 2.071 (mg/kg/d)−1 for an absolute risk of 10 %. In this study, mice were exposed for 80 consecutive weeks, shorter than the standard lifespan. A temporal weighting of 1.30 (=104/80) must therefore be applied. The ERF for liver cancer with a lifetime exposure in men and women is 2.692 (mg/kg/d)−1.

Table 4 BMD10 for liver carcinomas in male mice (data from NCI, 1976)

The Reuber study provides 2 sets of data for liver cancer (male and female rats). None of the 3 dose groups were significantly different from the control group (see: Additional file 2: Table C). According to the criteria exposed in the section “statistical test before modeling”, the results of this study cannot be modeled (no significant difference from the control group).

In the Larson study, examination of rats showed some hyperplasia lesions (3 females at 10 ppm and 1 female and 2 males at 25 ppm) that could be precancerous (without histopathological confirmation). The doses are expressed as ppm chlordecone in the diet. From the average food consumption (in gfood/kg/d) by group and by sex measured at different time intervals (5, 13, 26, 52 and 104 weeks) we calculated the weighted average doses of males and females with this data (see: Additional file 2: Table D). Only one dose group is significantly different from the control group in males (25 ppm) or in females (10 ppm) (see: Additional file 2: Table E). Hence, these datasets cannot be modeled (not enough of dose groups for modelisation).

For other endpoints of hepatotoxicity there is no data meeting our criteria. The Larson study shows hepatotoxicity characterized by some hyperplasia in the high dose group and changes of liver fat in all dose groups. Two data sets (male and female) were created by adding these 2 effects (see: Additional file 2: Table F). In males as in females only a single dose group is statistically different from that of the control group. These data sets are therefore excluded. There is no other quantitative data for chronic hepatotoxicity of chlordecone.

For renal dysfunction, ERF data are from animal studies. The Larson study has found a significant adverse effect of chlordecone on kidneys. In females the trend test is significant. The incidence of kidney damage (15 %) in the first dose group is not different from the control group (12 %) contrary to the other dose groups (see: Additional file 2: Table G). This dose group is excluded from modeling. In males, the trend test was not significant (p = 0.094). No dose group is different from the control group whose background incidence (55 %) is already very high. Combining the 2 sexes is not acceptable because of a significant difference in the incidence in the control groups of males and females (p <0.0005). Only the female data set, without the first dose group, can be modeled to estimate a BMD. All models, except Logistic and Probit, fit the data set (see Table 5). The ratio between the smallest and largest BMD was 29, so the BMD10 having the smallest AIC score (LogLogistic restricted) is retained: 0.0495 mg/kg/d. From this a BMD10-HED of 0.0126 mg/kg/d (DAF = 0.255) is calculated, yielding an ERF of 7.923 (mg/kg/d)−1 for increased kidney damage among women with SLE (Systemic Lupus Erythematosus). The ERF is limited to women with SLE because glomerulosclerosis in mice and rats are found only in strain prone to SLE [10]. The incidence of SLE in women remains poorly quantified between 15 and 50 per 100 000. We assume a geometric mean incidence rate of 27 per 100 000 to convert the ERF for women with SLE into an ERF of 0.0022 (mg/kg/d)−1 (=7.923 (mg/kg/d)−1 × 0.00027) for all women.

Table 5 BMD10 for glomerulosclerosis in female rat (data from Larson, 1979, dose group 1 excluded)

Exposure data are from epidemiological studies and presented in Table H of Additional file 2. Parameter values of the β-substitution model for each study are also in Additional file 2 (Table I). The average exposure values spread into 5 groups are shown in Table 6. To compare these exposure data with the RfD (0.5 μg/kg/d) we converted the RfD to a blood chlordecone concentration equivalent with a conversion factor CFe/i of 0.0064 (μg/kg/d)/(μg/l) (see details in Additional file 1) giving a value of 7.81 μg/l. This level is reached only in the last group of exposure (group 5) of 2 populations: adult men and women only before 2003. The weighted mean (last column in Table 6) of all groups are below this value. Therefore, blood chlordecone concentrations measured in epidemiological studies are mainly under the RfD value. If chlordecone would not cause some effects at this low exposure levels, epidemiological studies that take cofactors and confounders into account would probably not be able to detect significant associations especially in newborns. The analytical limits of detection (LD) were lowered between the first studies (1.5 and 0.5 μg/L) and those realized after 2003 (0.25 μg/L). Despite this decrease, the rates of results below the LD increase significantly after 2003. This indicates that the exposure decrease is associated with a strong shift in the distribution to the left. However, the maximum values do not seem to follow the same decreasing trend and are even higher after 2004 in pregnant women and newborns. Because of the known kinetics of chlordecone in human blood (half-life of about 6 months), these observations could be interpreted as a decrease in exposures of continuous and homogeneous nature, for example via tap water, and the persistence of higher intermittent exposures, for example via food consumed occasionally. These exposure data will subsequently be used to assess the risks and impacts and the health and economic benefits attributable to the exposure control program.

Table 6 Blood chlordecone concentrations: averages by exposure group for purpose of risk assessment

Discussion

Our literature search has been updated from January 2014 up to March 2016. Only 5 new toxicological mechanistic studies were found. Two of them confirm estrogenicity of chlordecone and were more designed to develop new estrogenic tests [129, 130]. One study shows how chlordecone among others regulates the liver drug transporter activity and expression [131]; that may contribute to the liver toxicity that we have already taken into account. Another study [132] does not add anything beyond what is already known about chlordecone neurotoxicity. The last study, using quantitative PCR, revealed an induction of genes involved in defense mechanism against oxidative stress (catalase and selenium-dependent glutathione peroxidase) in prawns exposed to low environmental concentrations of chlordecone after 12 and 24 h of exposure. In prawns reared in a contaminated farm, transcription of genes involved in the biotransformation process (cytochrome P450 and glutathione-S-transferase (GST)) was induced after 8 days of exposure [133]. None of these studies change our conclusions on key-events and MOA.

Several new epidemiological studies have been published, mostly based on data from the TIMOUN cohort [134138]. No significant associations were observed between chlordecone exposure and the risk of preeclampsia or gestational diabetes mellitus. This study suggests an inverse association between chlordecone exposure during pregnancy and gestational hypertension. The authors conclude that further studies are required to determine the underlying mechanism [134]. A 1-log10 increase in chlordecone concentration was associated with a decreased length of gestation (−0.27 weeks) and an increased risk of preterm birth (60 %). These associations may result from the estrogen-like and progestin-like properties of chlordecone [136]. Perinatal exposure to chlordecone may affect thyroid-stimulating hormone (TSH) and thyroid hormone levels at 3 months, differently according to the sex of the infant. This disruption however did not appear to intervene in the pathway between prenatal chlordecone exposure and fine motor development of children [137]. Chlordecone in cord blood was associated with a higher BMI in boys at 3 months and in girls at 8 and 18 months. Postnatal exposure was associated with lower height, weight and BMI at 3, 8 and 18 months, particularly in girls. Chlordecone exposure may affect growth trajectories in children aged 0 to 18 months [138]. Based on incident prostate cancer cases from 1981 to 2005 in Martinique Cancer Registry, incidence was found to increase by 5.07 % annually during this period of time. But an inverse association between exposure area and prostate cancer risk was found, with the highest prostate cancer incidence observed in urban zones showing the lowest soil contamination levels by chlordecone [135]. The studies confirm and amplify previous knowledge on developmental effects of chlordecone. The last study could raise doubts about chlordecone and prostate cancer. Nevertheless, the strength of an ecological design is not enough to negate the results of a case control designed study (here the KARUPROSTATE study). Moreover, because exposure to chlordecone is mainly due to food contamination (vegetable and seafood product) exposure to chlordecone is not limited to people living on contaminated soil. No ERF can be derived from these new epidemiological studies.

Following NRC 2009 recommendations for health risk assessment we assume a low dose effect if the MOA is active at low dose and/or is shared by a human disease and/or by a ubiquitous chemical [21]. Based on analysis of MOA studies, we have identified 4 potentially effective low dose effects of chlordecone: cancer promotion, developmental impairment, hepatotoxicity, and neurotoxicity. Only hepatotoxicity is not driven by an endocrine disruption mechanism. ERFs with a linear shape at low dose have been derived for these effects from epidemiological studies (prostate cancer and cognitive development) and animal studies (liver cancer). There are no quantitative data for neurotoxic effects in adulthood. Since the early 80s when damage to DNA appeared to be a major cause of most cancers, non-mutagenic chemicals (such as chlordecone) have been thought unable to generate cancers in human. In a very recent collaborative research, involving almost 200 cancer biologists, examination of MOA found that among 85 non-mutagenic chemicals (not IARC group 1, carcinogens) only 15 % show evidence of a dose–response threshold, whereas 59 % exerted low-dose effects (no dose–response information was found for the remaining 26 %). The authors suggests that the cumulative effects of individual non-carcinogenic chemicals acting on different pathways, and a variety of related systems, organs, tissues and cells could plausibly conspire to produce carcinogenic synergies [139]. Moreover, some endocrine disrupting chemicals are known to have adverse effects at low-dose. Most of them could act on development, fertility, immunity response, or central nervous system. Nonmonotonic dose response seems to be a common hallmark endocrine disrupting chemicals, with stronger response at the low-dose than at higher doses [140]. Thus our findings on chlordecone appear consistent with others involving new approaches based on MOA and weight of evidence analysis.

An additional ERF was derived for kidney damage to allow comparison with the 3 low dose effects. Kidney damage is considered by health and safety agencies as the most sensitive effect. Unfortunately its mechanism of action is not well known. Chlordecone seems to favor the occurrence of SLE (Systemic Lupus Erythematosus) in genetically susceptible animals [10], but this effect is not observed in strains that are not lupus-prone [10]. The underlying mechanism favoring autoimmunity could be an increase of B cells in ways that are not fully comparable with those of a pure estrogen such as estradiol [141]. On the other hand, direct toxicity of chlordecone on renal tissues cannot be excluded [18]. Unfortunately, data from Larson et al. relate to the onset of kidney damage and not to the onset of SLE. Given these shortcomings one can hypothesize that if chlordecone causes kidney damage it is only in individuals prone to SLE.

We have derived 4 ERFs to quantify those effects at low doses of chlordecone, each with a different MOA. Cancers of the liver may be due to the production of reactive oxygen species via induction of CYP450. The other 3 effects are potentially linked to endocrine disruption mechanisms but they affect different organs and different population categories: brain development of boys, prostate cancer in men over 44 years, and renal disease in women. Using these 4 ERFs in the same population therefore cannot produce double counting of damages. An adult can suffer from 2 effects of chlordecone via 2 distinct mechanisms occurring in 2 distinct organs, for men: liver cancer and prostate cancer; for women: liver cancer and kidney damage.

Animal data were modeled as absolute risk (“added risk” in BMDS). In addition all data sets have been modeled with the option of relative risk (“extra risk” in BMDS) for comparison purposes (results not shown). When the incidence in the control group is zero (for example liver cancer in rats of both sexes and in female mice) relative risk and absolute risk BMD10 are identical. When the incidence of hepatocellular carcinoma in the control group is not zero, like in male mice, the relative risk BMD10 (logistic model: 0.44 mg/kg/d) is less than the absolute risk BMD10 (logistic model: 0.54 mg/kg/d). This is an increase of 22 % of the dose giving a 10 % risk. This proportion corresponds to the background incidence in male mice: 21 %. This is also true for kidney damage where the incidence in the control group of female rats is 12 %. The difference between the relative risk BMD10 (0.109 mg/kg/d) and the absolute risk BMD10 (0.126 mg/kg/d) is 15 %. But since the background incidence of diseases in laboratory animals cannot be considered representative of the background incidence in human population, we recommend using the option of absolute risk when estimating BMD for ERF derivation. Thereby our ERF from animal data is used for humans without the need to know the background incidence.

The ERFs from human studies are based on relative risk and weighted by the background incidence, whereas those from animal studies are based on absolute risk. We wanted to know if the use of relative risk with animal data would change the numerical value of the ERF. This is possible only when the background incidence of the effect is not zero. In the Larson study, the incidence of kidney damage in the female control group was not zero (12 % of kidney damage). So we calculated the relative risks using the R software as in an epidemiological study. ERFs are then derived according to equation 1: ERF = (RR-1)/a. The geometric mean of the 3 ERFs for the 3 dose groups significantly different from the control group was 67 (mg/kg/d)−1. If we multiply this ERF by background incidence (I0), an absolute risk ERF is obtained (RR-1/a × I0 = (I0 - Ie)/a). Here the absolute risk ERF would be 8.04 (mg/kg/d)−1 = 67 (mg/kg/d)−1 × 12 %. This value is close to the absolute risk ERF obtained through BMD modeling: 7.92 (mg/kg/d)−1. If one derives the ERF from the first significant dose group, then the absolute ERF is 10.26 (mg/kg/d)−1, with the second group it is 10.74 (mg/kg/d)−1 and with the third group 4.72 (mg/kg/d)−1. The 2 approaches appear to give comparable results. Conversely, one could derive an ERF from epidemiological data with BMD models, but in the case of chlordecone there is no data for testing this approach. It would require cohort studies because in the case–control studies background incidence is not known. One would also need at least 2 exposure groups in the epidemiological study significantly different from the reference exposure group. The corresponding epidemiological data for these 2 conditions are unusual. In summary the 2 approaches, epidemiological RR or BMD modeling, can be considered as equivalent for ERF derivation.

The study used to derive the ERF for prostate cancer was conducted in Guadeloupe. With over 600 incident cases, it was the largest study on the association between exposure to organochloride pesticides and risk of prostate cancer known at the time of its publication. The multivariate adjustment increases the odds ratios, which is unusual and may indicate the presence of protective cofactors among confounders. The measured increased risk does not include the exposure during the periods of development in utero or of adolescence, critical periods for carcinogenic effects and endocrine disruption. The authors discuss the role of chlordecone as promoter via its affinity for cell estrogenic receptors (ERα), confirmed by our analysis of the MOA. Affinity for this receptor ERα could also promote angiogenesis [117]. The ERF based on this study derives from the highest exposure group because it is the only one that is statistically significant. The ERF from the lower exposure group would be 4 times higher and would estimate 4 times more risk and impacts. However, that would propagate high uncertainty because the lower bound of this RR is less than 1 (meaning less risk in this exposed group than in reference group). Thus, our ERF appears to be of good quality and there are more arguments in favor of an underestimation of risk than in favor of an overestimation.

A strength of the study supporting the ERF for altered cognitive development is its prospective cohort design. However the number of children involved is quite low. Some results are inconsistent, for example for communication skills the slope is negative (deterioration) in the low exposure group and positive (improvement) in the high exposure group; however, those results are not statistically significant. Only the decrease in fine motor skills in boys is statistically significant. The greatest uncertainty related to this effect is not in the results of the study but in the interpretation we have made for the monetary valuation. The measurement of cognitive development levels uses the ASQ scores, but we do not know the monetary value of an ASQ point while we know quite well that of an IQ point. We assume that an ASQ score point at 18 months is equal to an IQ point at age 6–7 years. It is very difficult to know if this assumption overestimates or underestimates the true strength of the effect because there are no relevant studies. To estimate the uncertainty we consider as lower bound that an ASQ point of fine motor skills at 18 months equals 1/5 IQ point at 7 years (only one of 5 ASQ scores is reduced by chlordecone), and as the upper bound we consider that an ASQ point is equal to 2 IQ points (the deterioration of cognitive development suffered in utero continues in the following years since exposure is continuous).

The ERF for liver cancer due to chlordecone derives from an animal study [19]. Another study found increased liver hyperplasia [7] but did not use histological examinations to know if the lesions are carcinogenic or not. The NCI study involves 2 species (mouse and rat), both sexes in each species and 50 animals per dose group. Many other organs were subjected to histological examinations but only the hepatic carcinomas were correlated with chlordecone doses. However, the NCI study has weaknesses already mentioned. The US EPA has used a time to tumor model to estimate BMD10 for female rats and mice of both sexes [18]. The geometric mean of 3 BMD10 obtained with the time to tumor model is 0.075 mg/kg/d, that of mice only is 0.038 mg/kg/d. By comparison, our results with the time adjustment are: 0.076 mg/kg/d and 0.037 mg/kg/d. In theory the time to tumor model is better but it needs the date of onset of tumors, information that is not always available in publications. Here the 2 approaches give very similar results; this indicates a good robustness of our approach.

The ERF for prostate cancer derives from a recent human study that took into account the cofactors and its variability seems to us properly represented by the 95 % confidence interval of the odds ratio. In the study on impaired cognitive development, the main uncertainty comes from our assumption about the relationship between ASQ scores and IQ points. We used a simple assumption, easily modifiable when better knowledge becomes available. For the ERF derived from animal studies, the main variability comes from either the BMD value or the conversion factor of ingestion dose to blood chlordecone concentration (CFe/i). This factor is based on chlordecone pharmacokinetic knowledge that is described in the Additional file 1. The available human data indicate a half-life of blood chlordecone ranging from 63 to 192 days with an average of 127.5 j. The CFe/i obtained from these half-lives are: mean CFe/i = 0.064 (mg/kg/d)/(μg/L); low CFe/i = 0.043 (mg/kg/d)/(μg/L); high CFe/i = 0.131 (mg/kg/d)/(μg/L) (see Additional file 1), numbers we use for framing the variability of the ERF if the range (range = 3.1 = 0.131/0.043) is larger than the range between the smallest and the largest BMD value.

There are only 2 exposure studies before 2003 (INSERM and HIBISCUS) and their enrollment is lower than in the studies after 2003 (KARUPROSTATE and TIMOUN). The representativeness of the INSERM study data for men ages 20 to 45 is uncertain because the participants are workers followed by the occupational medical service. This excluded those who are unable to work, unemployed or independent workers. However these are the only data available before 2004. In addition, one of the objectives of this cross-sectional study was to describe the exposure to chlordecone in adult men. Moreover the age group in the INSERM study (20–45 years) does not match the age group in the KARUPROSTATE study (>44 years). In KARUPROSTATE the control group is a random sample of men aged over 44 years summoned by the national health insurance for free and systematic health screening; a recruitment bias is unlikely. The representativeness of the HIBISCUS study of women of childbearing age also involves uncertainty, but more limited than for INSERM. Recruitment for HIBISCUS was similar to the TIMOUN study: women in the sixth month of pregnancy planning to give birth in a public hospital in Guadeloupe (about 70 % of births). In summary, the exposure data available before 2004 aimed to assess the exposure of men and women of reproductive age. The representativeness of HIBISCUS and especially of INSERM for the general population involves uncertainties that are difficult to estimate. Anyway, these are the only exposure data available in 2013, and they will be used to assess the risks and impacts of exposure to chlordecone in Guadeloupe.

Conclusions

This is the first study, to our knowledge, to yield explicit exposure-response functions (ERFs) for non-genotoxic effects of chronic low exposure to chlordecone. It is also one of the first studies explicitly designed to implement the recommendations of the North American National Academy of Sciences, described in the Silver Book in 2009 [21].

The methodological framework developed here allows estimating ERFs that produce a central risk estimate. These ERFs are robust with regard to choice of models, thus our methodological framework could be used generally. Furthermore, the ERFs obtained could be used in any population, without need to know the background incidence of a disease. Some ERFs are expressed in blood chlordecone concentration (μg/l), others are expressed in dose by ingestion (chlordecone from food intake). For use in a risk assessment study, it may be necessary to have a conversion factor between ingestion dose and blood levels of chlordecone. In the evaluation of health and economic impacts, the ERF uncertainties can be framed by the lower and upper bounds of each ERF. In any case, these ERF pave the way for a quantitative assessment of risks and impacts for non-mutagens chemicals, a great added value for public health decisions.

Abbreviations

AIC, Akaïke’s Information Criterion; ASQ, age and stage questionnaire; BMD, benchmark dose; BMD10, benchmark dose for an extra risk of 10 %; BMDHED, benchmark dose expressed as human equivalent dose; BMR, benchmark response; CF, conversion factor; CYP450, cytochrome P450; DAF, dosimetric adjustment factor; ERF, exposure-response function; ERα, estrogen receptor alpha; IQ, intellectual quotient; LD, Limit of Detection; LOAEL, lowest observed adverse effect level; MLE, maximum likelihood estimation; MOA, mode of action; NCI, National Cancer Institute; NOAEL, no observed adverse effect level; OR, odds ratio; P25, P50, Px, percentile 25, percentile 50, percentile x.; POD, point of departure; RfD, reference dose; RR, relative risk; SLE, systemic lupus erythematosus; US EPA, United State Environmental Protection Agency; WHO, World health Organization; βM, mean of the censured data estimated with β-substitution method

References

  1. 1.

    DGS. Plan d’action chlordécone en Martinique et en Guadeloupe 2008–2010. Paris: Direction Générale de la Santé; 2008.

  2. 2.

    DGS. Plan d’action chlordécone en Guadeloupe et en Martinique 2008–2010. Bilan par action 2010. Paris: Direction Générale de la Santé; 2011.

  3. 3.

    DGS. Plan d’action contre la pollution par la chlordécone en Guadeloupe et en Martinique 2011–2013. Paris: Direction Générale de la Santé; 2011.

  4. 4.

    InVS. Impact sanitaire de l’utilisation du chlordécone aux Antilles françaises Recommandations pour les recherches et les actions de santé publique. Saint Maurice: InVS, INSERM; 2009.

  5. 5.

    Godard E, Guldner L. Évaluation et gestion du risque alimentaire associé au chlordécone pour les populations de Guadeloupe et de Martinique. Numéro thématique – Chlordécone aux Antilles : bilan actualisé des risques sanitaires BEH. 2011;3-4-5:34–6.

  6. 6.

    AFSSA, AFSSA. Actualisation de l’exposition alimentaire au chlordécone de la population antillaise. Evaluation de l’impact de mesures de maîtrise des risques. final. Maisons-Alfort: Agence Française de Sécurité Sanitaire des Aliments; 2007.

  7. 7.

    Larson PS, Egle Jr JL, Hennigar GR, Lane RW, Borzelleca JF. Acute, subchronic, and chronic toxicity of chlordecone. Toxicol Appl Pharmacol. 1979;48(1):29–41.

  8. 8.

    Guzelian PS. Comparative toxicology of chlordecone (Kepone) in humans and experimental animals. Annu Rev Pharmacol Toxicol. 1982;22:89–113. doi:10.1146/annurev.pa.22.040182.000513.

  9. 9.

    Guzelian PS. The clinical toxicology of chlordecone as an example of toxicological risk assessment for man. Toxicol Lett. 1992;64–65:589–96.

  10. 10.

    Sobel ES, Wang F, Butfiloski E, Croker B, Roberts SM. Comparison of chlordecone effects on autoimmunity in (NZBxNZW) F (1) and BALB/c mice. Toxicology. 2006;218(2–3):81–9.

  11. 11.

    Multigner L, Ndong JR, Giusti A, Romana M, Delacroix-Maillard H, Cordier S, et al. Chlordecone exposure and risk of prostate cancer. J Clin Oncol. 2010;28(21):3457–62.

  12. 12.

    Dallaire R, Muckle G, Rouget F, Kadhel P, Bataille H, Guldner L, et al. Cognitive, visual, and motor development of 7-month-old Guadeloupean infants exposed to chlordecone. Environ Res. 2012;118:79–85. doi:10.1016/j.envres.2012.07.006.

  13. 13.

    Boucher O, Simard MN, Muckle G, Rouget F, Kadhel P, Bataille H, et al. Exposure to an organochlorine pesticide (chlordecone) and development of 18-month-old infants. Neurotoxicology. 2013;35C:162–8. doi:10.1016/j.neuro.2013.01.007.

  14. 14.

    Multigner L, Kadhel P, Huc-Terki F, Thome JP, Janky E, Auger J. Exposure to Chlordecone and Male Fertility in Guadeloupe (French West Indies). Epidemiology. 2006;17(6):S372.

  15. 15.

    Multigner L, Kadhel P, Pascal M, Huc-Terki F, Kercret H, Massart C, et al. Parallel assessment of male reproductive function in workers and wild rats exposed to pesticides in banana plantations in Guadeloupe. Environmental health: a global access science source. 2008;7:40. doi:10.1186/1476-069x-7-40.

  16. 16.

    Emeville E, Giton F, Giusti A, Oliva A, Fiet J, Thome JP, et al. Persistent organochlorine pollutants with endocrine activity and blood steroid hormone levels in middle-aged men. PLoS ONE. 2013;8(6), e66460. doi:10.1371/journal.pone.0066460.

  17. 17.

    ATSDR. Toxicological profile for mirex and chlordecone. Washington: Agency for Toxic Substances and Disease Registry. US Department of Health and Human Services; 1995.

  18. 18.

    US-EPA. Toxicological review of chlordecone (Kepone). Washington: United States Environmental Agency; 2009.

  19. 19.

    NCI. Report on carcinogenesis bioassay of technical grade chlordecone (Kepone) CAS No. 143-50-0: U.S. Department of health, education and welfare, Public Health Service, National Institut of Health. 1976.

  20. 20.

    Guldner L, Multigner L, Heraud F, Monfort C, Thome JP, Giusti A, et al. Pesticide exposure of pregnant women in Guadeloupe: ability of a food frequency questionnaire to estimate blood concentration of chlordecone. Environ Res. 2010;110(2):146–51.

  21. 21.

    NRC. Science and decisions. Advancing Risk Assessment. Washington: The National Academies Press; 2009.

  22. 22.

    Institute of Medicine. Cost of Environmental-Related Health Effects: A Plan for Continuing Study. Washington: The National Academies Press; 1981.

  23. 23.

    Melnick R, Lucier G, Wolfe M, Hall R, Stancel G, Prins G, et al. Summary of the National Toxicology Program’s report of the endocrine disruptors low-dose peer review. Environ Health Perspect. 2002;110(4):427–31.

  24. 24.

    Cannon S, Veazey JJ, Jackson R, Burse V, Hayes C, Straub W, et al. Epidemic kepone poisoning in chemical workers. Am J Epidemiol. 1978;107(6):529–37.

  25. 25.

    Guzelian PS, Vranian G, Boylan JJ, Cohn WJ, Blanke RV. Liver structure and function in patients poisoned with chlordecone (Kepone). Gastroenterology. 1980;78(2):206–13.

  26. 26.

    End DW, Carchman RA, Ameen R, Dewey WL. Inhibition of rat brain mitochondrial calcium transport by chlordecone. Toxicol Appl Pharmacol. 1979;51(1):189–96.

  27. 27.

    US-EPA. Guidelines for Carcinogen Risk Assessment. final. Washington: U.S. Environmental Protection Agency; 2005.

  28. 28.

    Greenland S. Concepts and pitfalls in measuring and interpreting attributable fractions, prevented fractions, and causation probabilities. Ann Epidemiol. 2015;25(3):155–61. doi:10.1016/j.annepidem.2014.11.005.

  29. 29.

    Hill AB. The Environment and Disease: Association or Causation? Proceedings of the Royal Society of Medicine. 1965;58(5):295–300.

  30. 30.

    US-EPA. Benchmark Dose Technical Guidance Document. Final. Washington: U.S. Environmental Protection Agency; 2012.

  31. 31.

    US-EPA. Benchmark Dose Software (BMDS) v2.4.0. Washington: U.S. Environmental Protection Agency; 2013.

  32. 32.

    The R Foundation. The R Project for Statistical Computing. https://www.r-project.org/

  33. 33.

    US-EPA. Recommended Use of Body Weight3/4 as the Default Method in Derivation of the Oral Reference Dose. Final. Washington: U.S. Environmental Protection Agency; 2011.

  34. 34.

    US-EPA. Recommendations for and Documentation of Biological Values for Use in Risk Assessment. Washington: United States Environmental Agency; 1988.

  35. 35.

    Nedellec V. Méthodes quantitatives pour évaluer les risques non mutagènes des substances chimiques. Application au cas du chlordécone. Paris: CNAM; 2015.

  36. 36.

    Joly P-B. La saga du chlordécone aux Antilles françaises. Reconstruction chronologique 1968–2008. Document réalisé dans le cadre de l’action 39 du plan chlordécone Convention de collaboration AFSSET – INRA. final: INRA. 2010.

  37. 37.

    Ganser GH, Hewett P. An accurate substitution method for analyzing censored data. J Occup Environ Hyg. 2010;7(4):233–44. doi:10.1080/15459621003609713.

  38. 38.

    Martinez AJ, Taylor JR, Dyck PJ, Houff SA, Isaacs E. Chlordecone intoxication in man. II. Ultrastructure of peripheral nerves and skeletal muscle. Neurology. 1978;28(7):631–5.

  39. 39.

    Taylor JR. Neurological manifestations in humans exposed to chlordecone and follow-up results. Neurotoxicology. 1982;3(2):9–16.

  40. 40.

    Taylor JR. Neurological manifestations in humans exposed to chlordecone: follow-up results. Neurotoxicology. 1985;6(1):231–6.

  41. 41.

    Taylor JR, Selhorst JB, Houff SA, Martinez AJ. Chlordecone intoxication in man. I. Clinical observations. Neurology. 1978;28(7):626–30.

  42. 42.

    Bolger R, Wiese TE, Ervin K, Nestich S, Checovich W. Rapid screening of environmental chemicals for estrogen receptor binding capacity. Environ Health Perspect. 1998;106:9.

  43. 43.

    Kuiper GG, Lemmen JG, Carlsson B, Corton JC, Safe SH, van der Saag PT, et al. Interaction of estrogenic chemicals and phytoestrogens with estrogen receptor beta. Endocrinology. 1998;139(10):4252–63.

  44. 44.

    Lemaire G, Mnif W, Mauvais P, Balaguer P, Rahmani R. Activation of alpha- and beta-estrogen receptors by persistent pesticides in reporter cell lines. Life Sci. 2006;79(12):1160–9. doi:10.1016/j.lfs.2006.03.023.

  45. 45.

    Li Y, Luh CJ, Burns KA, Arao Y, Jiang Z, Teng CT, et al. Endocrine-Disrupting Chemicals (EDCs): In Vitro Mechanism of Estrogenic Activation and Differential Effects on ER Target Genes. Environ Health Perspect. 2013. doi:10.1289/ehp.1205951.

  46. 46.

    Thomas P, Dong J. Binding and activation of the seven-transmembrane estrogen receptor GPR30 by environmental estrogens: a potential novel mechanism of endocrine disruption. J Steroid Biochem Mol Biol. 2006;102(1–5):175–9. doi:10.1016/j.jsbmb.2006.09.017.

  47. 47.

    Wu F, Khan S, Wu Q, Barhoumi R, Burghardt R, Safe S. Ligand structure-dependent activation of estrogen receptor alpha/Sp by estrogens and xenoestrogens. J Steroid Biochem Mol Biol. 2008;110(1–2):104–15. doi:10.1016/j.jsbmb.2008.02.008.

  48. 48.

    Yoon K, Pallaroni L, Stoner M, Gaido K, Safe S. Differential activation of wild-type and variant forms of estrogen receptor alpha by synthetic and natural estrogenic compounds using a promoter containing three estrogen-responsive elements. J Steroid Biochem Mol Biol. 2001;78(1):25–32.

  49. 49.

    Yoon K, Pellaroni L, Ramamoorthy K, Gaido K, Safe S. Ligand structure-dependent differences in activation of estrogen receptor alpha in human HepG2 liver and U2 osteogenic cancer cell lines. Mol Cell Endocrinol. 2000;162(1–2):211–20.

  50. 50.

    Baggett J, Klein RL, Mehendale HM, Thureson-Klein AK. Acute kepone treatment of rats: a biochemical and ultrastructural study. Pharmacologist. 1977;19:2.

  51. 51.

    Chambers JE, Trevathan CA. Effect of mirex, dechlorinated mirex derivatives and chlordecone on microsomal mixed-function oxidase activity and other hepatic parameters. Toxicol Lett. 1983;16(1–2):109–15.

  52. 52.

    Curtis LR. Chlordecone is a potent in vitro inhibitor of oligomycin-insensitive Mg2 + − ATPase of rat bile canaliculi-enriched fraction. J Biochem Toxicol. 1988;3:321–8.

  53. 53.

    Curtis LR, Mehendale HM. Hepatobiliary dysfunction and inhibition of ATPase activity of bile canaliculi-enriched fractions following in vivo mirex, photomirex and chlordecone exposures. Toxicol Appl Pharmacol. 1981;61(3):429–40.

  54. 54.

    Desaiah D, Ho IK, Mehendale HM. Effects of kepone and mirex on mitochondrial Mg2 + −ATPase activity in rat liver. Toxicol Appl Pharmacol. 1977;39(2):219–28.

  55. 55.

    Desaiah D, Ho IK, Mehendale HM. Inhibition of mitochondrial Mg2+ ATPase activity in isolated perfused rat liver by kepone. Biochem Pharmacol. 1977;26(12):1155–9.

  56. 56.

    Eaton DL, Klaassen CD. Effects of 2,3,7,8- tetrachloro dibenzo-p- dioxin, Kepone, and poly brominated biphenyls on transport systems in isolated rat hepatocytes. Toxicol Appl Pharmacol. 1979;51(1):137–44.

  57. 57.

    Ebel RE. In Vitro Effects Of Chlordecone (Kepone) On Hepatic Microsomal Cytochrome P-450. Pestic Biochem Physiol. 1980;14(3):221–6.

  58. 58.

    Ebel RE. Alterations in microsomal cytochrome P-450-catalyzed reactions as a function of chlordecone (Kepone) induction. Pestic Biochem Physiol. 1982;18(1):113–21.

  59. 59.

    Fabacher DL, Hodgson E. Induction of hepatic mixed-function oxidase enzymes in adult and neonatal mice by kepone and mirex. Toxicol Appl Pharmacol. 1976;38(1):71–7.

  60. 60.

    Kaminsky LS, Piper LJ, McMartin DN, Fasco MJ. Induction of hepatic microsomal cytochrome P-450 by mirex and kepone. Toxicol Appl Pharmacol. 1978;43(2):327–38.

  61. 61.

    Kocarek TA, Schuetz EG, Guzelian PS. Selective induction of cytochrome P450e by kepone (chlordecone) in primary cultures of adult rat hepatocytes. Mol Pharmacol. 1991;40(2):203–10.

  62. 62.

    Kocarek TA, Schuetz EG, Guzelian PS. Regulation of cytochrome P450 2B1/2 mRNAs by Kepone (chlordecone) and potent estrogens in primary cultures of adult rat hepatocytes on Matrigel. Toxicol Lett. 1994;71(2):183–96.

  63. 63.

    Lamartiniere CA, Nicholas JM. Neonatal chlordecone alteration of the ontogeny of sex-differentiated hepatic drug and xenobiotic metabolizing enzymes. Biochem Pharmacol. 1984;33(24):4092–5.

  64. 64.

    Lee J, Scheri RC, Zhang Y, Curtis LR. Chlordecone, a mixed pregnane X receptor (PXR) and estrogen receptor alpha (ERalpha) agonist, alters cholesterol homeostasis and lipoprotein metabolism in C57BL/6 mice. Toxicol Appl Pharmacol. 2008;233(2):193–202.

  65. 65.

    Lewandowski M, Levi P, Hodgson E. Induction of cytochrome P-450 isozymes by mirex and chlordecone. J Biochem Toxicol. 1989;4(3):195–9.

  66. 66.

    Madhukar BV, Matsumura F. Comparison Of Induction Patterns Of Rat Hepatic Microsomal Mixed-Function Oxidases By Pesticides And Related Chemicals. Pestic Biochem Physiol. 1979;11:1–3.

  67. 67.

    Mehendale HM, Takanaka A, Desaiah D, Ho IK. Kepone induction of hepatic mixed function oxidases in the male rat. Life Sci. 1977;20(6):991–8.

  68. 68.

    Mehendale HM, Takanaka A, Desaiah D, Ho IK. Effect of preexposure to Kepone on hepatic mixed-function oxidases in the female rat. Toxicol Appl Pharmacol. 1978;44(1):171–89.

  69. 69.

    Simmons JE, Berman E, Jackson M, Lewtas J. In Vitro and In Vivo Toxicity: A Comparison of Acrylamide, Cyclophosphamide, Chlordecone, and Diethylstilbestrol. J Environ Sci Health Part A Environ Sci Eng. 1987;7:639–64.

  70. 70.

    Desaiah D, Koch RB. Inhibition of ATPases activity in channel catfish brain by Kepone and its reduction product. Bull Environ Contam Toxicol. 1975;13(2):153–8.

  71. 71.

    Folmar LC. In Vitro Inhibition Of Rat Brain ATPase, pNPPase, And ATP-32Pi Exchange By Chlorinated-Diphenyl Ethanes And Cyclodiene Insecticide. Bull Environ Contam Toxicol. 1978;19(4):481–8.

  72. 72.

    Doherty JD, Salem N, Lauter CJ, Trams EG. Mn2+ and Ca2+ ATPases in lobster axon plasma membranes and their inhibition by pesticides. Comp Biochem Physiol C. 1980;69(2):185–90.

  73. 73.

    Desaiah D, Gilliland T, Ho IK, Mehendale HM. Inhibition of mouse brain synaptosomal ATPases and ouabain binding by chlordecone. Toxicol Lett. 1980;6(4–5):275–85.

  74. 74.

    Desaiah D, Trottman CH. Sensitivity of rat brain synaptosomal ATPases to several structurally related organochlorine compounds. Toxicol Lett. 1980;5:1.

  75. 75.

    Bracken WM, Sharma RP, Kleinschuster SJ. The effects of select neurotoxic chemicals on synaptosomal monoamine uptake and K + −dependent phosphatase. Fundamental and applied toxicology : official journal of the Society of Toxicology. 1981;1(6):432–6.

  76. 76.

    Bansal SK, Desaiah D. Effects of chlordecone and its structural analogs on p-nitrophenyl phosphatase. Toxicol Lett. 1982;12(2–3):83–90.

  77. 77.

    Chetty SC, Aldous CN, Rashatwar SS, Desaiah D. Effect of chlordecone on pH and temperature dependent substrate activation kinetics of rat brain synaptosomal ATPases. Biochem Pharmacol. 1983;32(21):3205–11.

  78. 78.

    Bansal SK, Desaiah D. Chlordecone toxicity: effect of withdrawal of treatment on ATPase inhibition. Neurotoxicology. 1985;6(3):103–7.

  79. 79.

    Desaiah D, Chetty CS, Rao KS. Chlordecone inhibition of calmodulin activated calcium ATPase in rat brain synaptosomes. J Toxicol Environ Health. 1985;16(2):189–95. doi:10.1080/15287398509530732.

  80. 80.

    Jinna RR, Uzodinma JE, Desaiah D. Age-related changes in rat brain ATPases during treatment with chlordecone. J Toxicol Environ Health. 1989;27(2):199–208. doi:10.1080/15287398909531291.

  81. 81.

    Uzodinma JE, Desaiah D. Age-Related Changes in Rat Brain ATPases during Treatment with Chlordecone. J Toxicol Environ Health. 1989;27(2):199–208.

  82. 82.

    Maier WE, Costa LG. Sodium, potassium-ATPase in rat brain and erythrocytes as a possible target and marker, respectively, for neurotoxicity: Studies with chlordecone, organotins and mercury compounds. Toxicol Lett. 1990;51(2):175–88.

  83. 83.

    Chang-Tsui YYH, Ho IK. Effects of Kepone (chlordecone) on synaptosomal g-aminobutyric acid uptake in the mouse. Neurotoxicology. 1979;1(2):357–67.

  84. 84.

    Hwang EC, Van Woert MH. Serotonin-norepinephrine interactions in the tremorolytic actions of phenoxybenzamine and trazodone. Pharmacol Biochem Behav. 1979;10(1):27–9.

  85. 85.

    Baggett JM, Thureson-Klein A, Klein RL. Effects of chlordecone on the adrenal medulla of the rat. Toxicol Appl Pharmacol. 1980;52(2):313–22.

  86. 86.

    Chang-Tsui YYH, Ho IK. Effect of Kepone (chlordecone) on synaptosomal catecholamine uptake in the mouse. Neurotoxicology. 1980;1(3):643–51.

  87. 87.

    Seth PK, Agrawal AK, Bondy SC. Biochemical changes in the brain consequent to dietary exposure of developing and mature rats to chlordecone (kepone). Toxicol Appl Pharmacol. 1981;59(2):262–7.

  88. 88.

    Ali SF, Hong JS, Wilson WE, Lamb JC, Moore JA, Mason GA, et al. Subchronic dietary exposure of rats to chlordecone (Kepone) modifies levels of hypothalamic beta-endorphin. Neurotoxicology. 1982;3(2):119–24.

  89. 89.

    Fujimori K, Nabeshima T, Ho IK, Mehendale HM. Effects of oral administration of chlordecone and mirex on brain biogenic amines in mice. Neurotoxicology. 1982;3(2):143–8.

  90. 90.

    Gerhart JM, Hong JS, Tilson HA. The possible sites of chlordecone-induced tremor in rats. Toxicol Appl Pharmacol. 1983;70(3):382–9.

  91. 91.

    Aldous CN, Chetty CS, Mehendale HM, Desaiah D. Lack of effects of chlordecone on synthesis rates, steady state levels and metabolites of catecholamines in rat brain. Neurotoxicology. 1984;5(2):59–65.

  92. 92.

    Gandolfi O, Cheney DL, Hong JS, Costa E. On the neurotoxicity of chlordecone: a role for gamma-aminobutyric acid and serotonin. Brain Res. 1984;303(1):117–23.

  93. 93.

    Hong JS, Tilson HA, Uphouse LL, Gerhart J, Wilson WE. Effects of chlordecone exposure on brain neurotransmitters: possible involvement of the serotonin system in chlordecone-elicited tremor. Toxicol Appl Pharmacol. 1984;73(2):336–44.

  94. 94.

    Hudson PM, Yoshikawa K, Ali SF, Lamb JC, Reel JR, Hong JS. Estrogen-like activity of chlordecone (kepone) on the hypothalamo-pituitary axis: effects on the pituitary enkephalin system. Toxicol Appl Pharmacol. 1984;74(3):383–9.

  95. 95.

    Desaiah D. Chlordecone interaction with catecholamine binding and uptake in rat brain synaptosomes. Neurotoxicology. 1985;6(1):159–65.

  96. 96.

    Bloomquist JR, Adams PM, Soderlund DM. Inhibition of gamma-aminobutyric acid-stimulated chloride flux in mouse brain vesicles by polychlorocycloalkane and pyrethroid insecticides. Neurotoxicology. 1986;7(3):11–20.

  97. 97.

    Fujimori K, Benet H, Mehendale HM, Ho IK. In vivo and in vitro synthesis, release, and uptake of [3-H]-dopamine in mouse striatal slices after in vivo exposure to chlordecone. J Biochem Toxicol. 1986;1(4):1–12.

  98. 98.

    Tilson HA, Hudson PM, Hong JS. 5,5-Diphenylhydantoin antagonizes neurochemical and behavioral effects of p, p’-DDT but not of chlordecone. J Neurochem. 1986;47(6):1870–8.

  99. 99.

    Uphouse L, Eckols K. Serotonin receptors in striatum after chlordecone treatment of adult female rats. Neurotoxicology. 1986;7(1):25–32.

  100. 100.

    Carr RL, Couch TA, Liu J, Coats JR, Chambers JE. The interaction of chlorinated alicyclic insecticides with brain GABAA receptors in channel catfish (Ictalurus punctatus). J Toxic Environ Health A. 1999;56(8):543–53.

  101. 101.

    Hoskins B, Ho IK. Chlordecone-induced alterations in content and subcellular distribution of calcium in mouse brain. J Toxicol Environ Health. 1982;9(4):535–44. doi:10.1080/15287398209530186.

  102. 102.

    Komulainen H, Bondy SC. Modulation of levels of free calcium within synaptosomes by organochlorine insecticides. J Pharmacol Exp Ther. 1987;241(2):575–81.

  103. 103.

    Bondy SC, Halsall LC. GM1 ganglioside enhances synaptosomal resistance to chemically induced damage. Neurosci Lett. 1988;84(2):229–33.

  104. 104.

    Bondy SC, Komulainen H. Intracellular Calcium as an Index of Neurotoxic Damage. Toxicology. 1988;49(1):35–41.

  105. 105.

    Bondy SC, McKee M. Prevention of chemically induced synaptosomal changes. J Neurosci Res. 1990;25(2):229–35. doi:10.1002/jnr.490250211.

  106. 106.

    Inoue K, Nakazawa K, Obama T, Fujimori K, Takanaka A. Chlordecone Inhibits Three Types of Ion Channels in a Neural Cell Line. Pharmacol Toxicol. 1991;67(5):444–6.

  107. 107.

    Cheng D, Bedford MT. Xenoestrogens regulate the activity of arginine methyltransferases. Chembiochem : a European journal of chemical biology. 2011;12(2):323–9. doi:10.1002/cbic.201000522.

  108. 108.

    Benachour N, Moslemi S, Sipahutar H, Seralini GE. Cytotoxic effects and aromatase inhibition by xenobiotic endocrine disrupters alone and in combination. Toxicol Appl Pharmacol. 2007;222(2):129–40. doi:10.1016/j.taap.2007.03.033.

  109. 109.

    Bjárge C, Brunborg G, Wiger R, Holme JA, Scholz T, Dybing E, et al. A comparative study of chemically induced DNA damage in isolated human and rat testicular cells. Reprod Toxicol. 1996;10(6):509–19.

  110. 110.

    Caldwell V, Loch-Caruso R. Chlordecone rapidly and reversibly inhibits gap junctional communication in human embryonic palatal mesenchyme cells. In Vitro Toxicol. 1992;5(2):113–22.

  111. 111.

    Starcevic SL, Bortolin S, Woodcroft KJ, Novak RF. Kepone (chlordecone) disrupts adherens junctions in human breast epithelial cells cultured on matrigel. In vivo. 2001;15(4):289–94.

  112. 112.

    Li X, Zhang S, Safe S. Activation of kinase pathways in MCF-7 cells by 17beta-estradiol and structurally diverse estrogenic compounds. J Steroid Biochem Mol Biol. 2006;98(2–3):122–32. doi:10.1016/j.jsbmb.2005.08.018.

  113. 113.

    Ikegwuonu FI, Mehendale HM. Biochemical assessment of the genotoxicity of the in vitro interaction between chlordecone and carbon tetrachloride in rat hepatocytes. J Appl Toxicol. 1991;11(4):303–10.

  114. 114.

    Moser GJ, Smart RC. Hepatic tumor-promoting chlorinated hydrocarbons stimulate protein kinase C activity. Carcinogenesis. 1989;10(5):851–6.

  115. 115.

    Rotenberg SA, Weinstein IB. Two polychlorinated hydrocarbons cause phospholipid-dependent protein kinase C activation in vitro in the absence of calcium. Mol Carcinog. 1991;4(6):477–81.

  116. 116.

    Tsushimoto G, Trosko JE, Chang CC, Matsumura F. Inhibition of intercellular communication by chlordecone (kepone) and mirex in Chinese hamster V79 cells in vitro. Toxicol Appl Pharmacol. 1982;64(3):550–6.

  117. 117.

    Clere N, Lauret E, Malthiery Y, Andriantsitohaina R, Faure S. Estrogen receptor alpha as a key target of organochlorines to promote angiogenesis. Angiogenesis. 2012;15(4):745–60. doi:10.1007/s10456-012-9288-7.

  118. 118.

    Skalsky HL, Fariss MW, Blanke RV, Guzelian PS. The role of plasma proteins in the transport and distribution of chlordecone (Kepone) and other polyhalogenated hydrocarbons. Ann N Y Acad Sci. 1979;320:231–7.

  119. 119.

    Rosecrans JA, Johnson JH, Tilson HA, Hong JS. Hypothalamic-pituitary adrenal (HPAA) axis function in adult Fischer-344 rats exposed during development to neurotoxic chemicals perinatally. Neurobehav Toxicol Teratol. 1984;6(4):281–8.

  120. 120.

    Rosecrans JA, Squibb Jr RE, Johnson JH, Tilson HA, Hong JS. Effects Of Neonatal Chlordecone Exposure On Pituitary-Adrenal Function In Adult Fischer-344-Rats. Neurobehav Toxicol Teratol. 1985;7(1):33–7.

  121. 121.

    Rosenstein L, Brice A, Rogers N, Lawrence S. Neurotoxicity of kepone in perinatal rats following in utero exposure. Toxicol Appl Pharmacol. 1977;41(1):142–3.

  122. 122.

    Squibb RE, Tilson HA. Effects of gestational and perinatal exposure to chlordecone (Kepone) on the neurobehavioral development of Fischer-344 rats. Neurotoxicology. 1982;3(2):17–26.

  123. 123.

    AFSSA. Avis de l’Agence française de sécurité sanitaire des aliments relatif à l’actualisation des données scientifiques sur la toxicité du chlordécone en vue d’une éventuelle révision des limites tolérables d’exposition proposées par l’Afssa en 2003. final. Maisons-Alfort: Agence Française de Sécurité Sanitaire des Aliments; 2007.

  124. 124.

    Reuber MD. Carcinomas of the liver in rats ingesting kepone. Neoplasma. 1979;26(2):231–5.

  125. 125.

    AFSSA. Avis de l’Agence française de sécurité sanitaire des aliments relatif à l’évaluation des risques liés à la consommation de denrées alimentaires contaminées par la chlordécone en Martinique et en Guadeloupe. final. Maisons-Alfort: Agence Française de Sécurité Sanitaire des Aliments; 2003.

  126. 126.

    Multigner L, Kadhel P. Exposition à des polluants environnementaux chez la femme enceinte et son enfant en Guadeloupep Niveaux de chlordécone dans le sang maternel et étude des déterminants de l’imprégnation. Inserm U625 & Service Gynécologie-obstétrique CHU Pointe-à-Pitre, Pointe-à-Pitre. 2004.

  127. 127.

    Guldner L, Seurin S, Héraud F, Multignier L. Exposition de la population antillaise au chlordécone. Numéro thématique – Chlordécone aux Antilles : bilan actualisé des risques sanitaires BEH. 2011;3-4-5:25–8.

  128. 128.

    Joachim C, Macni J, Véronique-Baudin J, Piccotti C, Escarmant P. Epidémiologie du cancer de la prostate aux Antilles-Guyane : données des registres généraux des cancers. BVS. 2013;8–9:3–5.

  129. 129.

    Giusti A, Lagadic L, Barsi A, Thome JP, Joaquim-Justo C, Ducrot V. Investigating apical adverse effects of four endocrine active substances in the freshwater gastropod Lymnaea stagnalis. Sci Total Environ. 2014;493:147–55. doi:10.1016/j.scitotenv.2014.05.130.

  130. 130.

    Wang S, Aarts JM, De Haan LH, Argyriou D, Peijnenburg AA, Rietjens IM, et al. Towards an integrated in vitro strategy for estrogenicity testing. J Appl Toxicol. 2014;34(9):1031–40. doi:10.1002/jat.2928.

  131. 131.

    Bucher S, Le Vee M, Jouan E, Fardel O. Regulation of hepatic drug transporter activity and expression by organochlorine pesticides. J Biochem Mol Toxicol. 2014;28(3):119–28. doi:10.1002/jbt.21543.

  132. 132.

    Costa LG. The neurotoxicity of organochlorine and pyrethroid pesticides. Handb Clin Neurol. 2015;131:135–48. doi:10.1016/b978-0-444-62627-1.00009-3.

  133. 133.

    Gaume B, Dodet N, Thome JP, Lemoine S. Expression of biotransformation and oxidative stress genes in the giant freshwater prawn Macrobrachium rosenbergii exposed to chlordecone. Environ Sci Pollut Res Int. 2015;22(11):7991–8002. doi:10.1007/s11356-014-3134-y.

  134. 134.

    Saunders L, Kadhel P, Costet N, Rouget F, Monfort C, Thome JP, et al. Hypertensive disorders of pregnancy and gestational diabetes mellitus among French Caribbean women chronically exposed to chlordecone. Environ Int. 2014;68:171–6. doi:10.1016/j.envint.2014.03.024.

  135. 135.

    Dieye M, Banydeen R, Macni J, Michel S, Veronique-Baudin J, Sasco A, et al. Geographic variations and temporal trends in prostate cancer in Martinique over a 25-year period. BMC research notes. 2014;7:262. doi:10.1186/1756-0500-7-262.

  136. 136.

    Kadhel P, Monfort C, Costet N, Rouget F, Thome JP, Multigner L, et al. Chlordecone exposure, length of gestation, and risk of preterm birth. Am J Epidemiol. 2014;179(5):536–44. doi:10.1093/aje/kwt313.

  137. 137.

    Cordier S, Bouquet E, Warembourg C, Massart C, Rouget F, Kadhel P, et al. Perinatal exposure to chlordecone, thyroid hormone status and neurodevelopment in infants: the Timoun cohort study in Guadeloupe (French West Indies). Environ Res. 2015;138:271–8. doi:10.1016/j.envres.2015.02.021.

  138. 138.

    Costet N, Pele F, Comets E, Rouget F, Monfort C, Bodeau-Livinec F, et al. Perinatal exposure to chlordecone and infant growth. Environ Res. 2015;142:123–34. doi:10.1016/j.envres.2015.06.023.

  139. 139.

    Goodson WH, Lowe L, Carpenter DO, Gilbertson M, Manaf Ali A, De Cerain L, Salsamendi A, et al. Assessing the carcinogenic potential of low-dose exposures to chemical mixtures in the environment: the challenge ahead. Carcinogenesis. 2015;36(1):S254–S96. doi:10.1093/carcin/bgv039.

  140. 140.

    Vandenberg LN, Colborn T, Hayes TB, Heindel JJ, Jacobs DR, Lee D-H, et al. Hormones and Endocrine-Disrupting Chemicals: Low-Dose Effects and Nonmonotonic Dose Responses. Endocr Rev. 2012;33(3):378–455. doi:10.1210/er.2011-1050.

  141. 141.

    Wang F, Roberts SM, Butfiloski EJ, Morel L, Sobel ES. Acceleration of autoimmunity by organochlorine pesticides: a comparison of splenic B-cell effects of chlordecone and estradiol in (NZBxNZW)F1 mice. Toxicol Sci. 2007;99(1):141–52.

Download references

Acknowledgements

We thank the scientific committee and also the people involved in the administrative management of the PNRPE program to have selected our project and having supported it to the end.

Funding

This work was done during the BAREPE project funded by grant from the French Ministry in charge of the environmental protection (Grant number: 11-MRES-PNRPE-3CVS-029) through the “Programme National de Recherche sur les Perturbateurs Endocriniens” PNRPE (The National Research Programme on Endocrine Disruptors).

The funding body has had no role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript.

Availability of data and materials

Because our work is entirely based on published data and information, the data sets supporting the conclusions of this paper are available either in the text, the tables and the additional files or in the original studies from which we have extracted results. The references of originals studies are given in the text. Contact with Vincent Nedellec, the corresponding author, is encouraged for any inquiries. (vincent.nedellec3@gmail.com).

Authors’ contributions

All authors of this manuscript have made substantial intellectual contributions by directly participating in the planning, execution, or analysis of the study. VN was responsible for bibliographical research, data extraction and modeling, performed the statistical analysis and contributed to the interpretation of data. WD supervised methods of modeling and contributed to the interpretation of data. AR participated in the development of the methodological framework, oversaw the monetization of health impacts and contributed to the interpretation of data. VN wrote the draft article, WD and AR reviewed, corrected and amended the text. All authors read and approved the final manuscript.

Competing interests

The authors have no competing interests.

Author information

Correspondence to Vincent Nedellec.

Additional files

Additional file 1: Figure I.

Decrease of chlordecone concentrations in rat organs and internal fluids over a 182 days period of time. Figure II. Decrease of blood chlordecone, in humans and in rats, after exposure cessation. Figure III. Modeled increase of blood chlordecone in humans and in rats following a continuous daily exposure to 0.5 μg/kg/d. (RTF 11334 kb)

Additional file 2: Table A.

DAF Calculated with the data of the studies. Table B. Data set on liver cancer from NCI 1976 study, and statistical tests. Table C. Data set on liver cancer from Reuber 1979 study, and statistical tests. Table D. Food consumption per dose group and sex in the Larson 1979 study, and conversion of dose to metric units. Table E. Data set on liver hyperplasia from Larson 1979 study, and statistical tests. Table F. Data set on hepatotoxicity (hyperplasia and change of liver fat) from Larson 1979 study, and statistical tests. Table G. Data set on glomerulosclerosis from Larson 1979 study, and statistical tests. Table G. Distribution of blood chlordecone concentration in Guadeloupe by period of time, data extracted from Guldner 2011. Table I. Parameters of the β-substitution model for substituted mean value to results < LD in epidemiological studies on chlordecone effects in Guadeloupe. (DOCX 116 kb)

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Keywords

  • Chlordecone
  • Low-dose
  • Mode of action
  • Key-events
  • Exposure-response function
  • Risk assessment
  • Non-mutagenic agent
  • Endocrine disrupter

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate. Please note that comments may be removed without notice if they are flagged by another user or do not comply with our community guidelines.