Widely Used Pesticides with Previously Unknown Endocrine Activity Revealed as in Vitro Antiandrogens

1Evidence suggests that prenatal and early-life exposure to pesticides may be causative factors in a variety of human disorders. For example, a meta-analysis by showed that maternally exposed offspring have increased risk of childhood leukemia [odds ratio = 2.64; 95% confidence interval (CI), 1.4–5].

2There are also indications that reproductive abnormalities, expressed as cryptorchidism, hypospadias, and decreased penile length, may be linked to pesticide exposure, most strikingly in maternally exposed boys. This is significant because male fertility is thought to be declining in many countries, and perinatal hypospadias/cryptorchidism are risk factors for reduced sperm quality and testicular cancer in adulthood. Banned persistent organochlorines [p,p′-DDT (1,1,1-trichloro-2-[o-chlorophenyl]-2,2-[p-chlorophenyl]ethane), p,p′-DDE (p,p′-1,1-bis-(4-chlorophenyl)-2,2-dichloroethene), β-hexachlorocyclohexane, hexachlorobenzene, α-endosulfan, cis-heptachloroepoxide, oxychlordane, dieldrin] were detected in all samples of breast milk in a case–control study of mothers in Denmark and Finland. Also, levels were significantly higher in samples from mothers of sons with cryptorchidism than in samples from matched controls (1997–2001; ). Female Danish greenhouse workers exposed to current-use pesticides were more likely to give birth to a son with cryptorchidism than were a random sample of mothers from the Copenhagen area (6.2% and 1.9%). Furthermore, sons of mothers who directly handled treated plants or were engaged in spraying pesticides had significantly smaller penises than did sons of mothers who had noncontact roles in the greenhouse industry. Last, in a recent meta-analysis of studies from the United States and Europe, reported that maternal occupational exposure to pesticides was associated with a 36% increased risk of hypospadias relative to the risk in mothers without exposure (risk ratio = 1.36; 95% CI, 1.04–1.77). The risk of developing cryptorchidism and hypospadias was also associated with paternal exposures to pesticides, mainly in greenhouses for the production of vegetables and flowers.

3The term “testicular dysgenesis syndrome” (TDS) has been proposed to explain the interrelated nature of these abnormalities. It is conceivable that estrogenic and/or antiandrogenic contaminants play a role in TDS. Experimental studies with rats have shown that maternal exposure to flutamide (a pharmaceutical antiandrogen) affects androgen-dependent developmental outcomes such as anogenital distance and nipple retention. However, ethinylestradiol has not been shown to affect these end points. Furthermore, hormone receptor screening in vitro suggests a preponderance of antiandrogenic activity compared with estrogenic activity in nonorganochlorine (current-use) pesticides. For example, screened 161 pesticides and reported that 52 were antiandrogenic, whereas only 29 were estrogenic, and reported that 6 of 12 pesticides screened were antiandrogenic and none were estrogenic. There is a good correlation between androgen receptor (AR) antagonist properties and in vivo antiandrogenic effects, and there is also good evidence that androgen-sensitive end points are demasculinized in male rats when exposed in utero to a wide range of pesticides. Antiandrogenic effects both in vitro and via maternal exposure in vivo have been reported in response to the herbicide linuron; the fungicides prochloraz, procymidone, tebuconazole, and vinclozolin; the organochlorine insecticides DDE and endosulfan; the organophosphate dimethoate; and the pyrethroid insecticide deltamethrin. However, with the exception of linuron, dimethoate, deltamethrin, and tebuconazole, the pesticides listed above have not been authorized for use in Europe during the past 5 years, which should result in lower occupational, residential, and dietary exposures. Endocrine-relevant data on current use pesticides is minimal—and in some cases completely absent—with most of the published literature focused on pesticides that are no longer registered for use.

4Therefore, the aim of this study was to test the antiandrogenic activity of currently used pesticides, with a view to informing future studies to determine their likely role in causing TDS. We selected compounds for testing based on evidence of human exposure (dietary intake data for Europe) and predicted AR antagonism according to the quantitative structure–activity relationship (QSAR) model developed by. Compounds predicted to be AR antagonists and compounds with high exposure scores were analyzed for AR antagonist properties using the MDA-kb2 assay. In addition, we used the yeast antiandrogen screen (YAS) to further test a subset of pesticides that were newly identified as AR antagonists or that had MDA-kb2 assay results that were discordant with QSAR predictions.

5Pesticides were selected using a combination of exposure scores and data about receptor-mediated antiandrogenic activity [see Supplemental Material, Figure 1 (doi:10.1289/ehp.1002895)]. First, we identified 134 pesticides with data suggesting relevant human exposures, including 58 pesticides identified at the highest concentrations and most frequently in European foods; 30 additional pesticides with relatively high daily dietary intakes (> 0.0004 μg/kg/day) identified by the FAO/WHO (Food and Agriculture Organization of the United Nations/World Health Organization) Joint Meeting on Pesticide Residues (JMPR); 44 additional pesticides identified in > 0.4% of fruits and vegetables during routine testing; and o,p′- and p,p′-DDE, which we included because of known adipose tissue levels. Each pesticide was assigned four scores, with each ranging from 1 to 10: a) maximum food residue level; b) estimated daily dietary intake; c) frequency of detection in fruits and vegetables, with a score of 5 assigned when data were not available; and d) a score according to the number of times pesticides were listed as one of the top 10 pesticides identified in fruits and cereals in Europe (a frequency score), with a score of 0 assigned if they were never listed. The four scores were summed to generate a “total exposure score,” with a maximum possible score of 40 (see Supplemental Material, Table 1).

6The second stage of compound selection for testing was an assessment of in vitro evidence of AR interaction in the available literature. Compounds previously shown not to be AR antagonists in vitro (n = 43) were removed from the list, which reduced the number of candidate pesticides from 134 to 91. Compounds previously reported to be AR antagonists (n = 27) were removed if the ratio of their total exposure score to their published IC20 [concentration that inhibits the androgenicity of DHT by 20%; total exposure score/published IC20 = “environmental relevance ratio” (ERR)] was < 3 (ERR was recalculated using our experimental data after the selection process). This left 14 previously reported AR antagonists for testing by the MDA-kb2 assay. For pesticides without published data (n = 64), AR antagonist activity was predicted using the QSAR developed by. These pesticides were tested using the MDA-kb2 assays if they were predicted to have AR antagonist activity (n = 4) or if they had high exposure scores (> 8) regardless of their QSAR status, including 6 pesticides that were predicted not to have AR antagonist activity and 13 pesticides that could not be predicted because they were out of the domain of the QSAR model. In total, 37 compounds were selected for testing in the MDA-kb2 assay. Finally, 8 pesticides that were newly described as highly active antiandrogens in the MDA-kb2 assay and 4 pesticides for which the QSAR prediction differed from the experimental result (including 1 out of the model domain) were subjected to further testing using the YAS (n = 14). For a summary of the selection process, see Supplemental Material, Figure 1 (doi:10.1289/ehp.1002895).

7Dihydrotestosterone (DHT; > 97% purity) was purchased from Steraloids Ltd. (Croydon, Surrey, UK); novaluron, dimethomorph, p,p′-DDE, methiocarb, and indoxacarb were purchased from Greyhound Chromatography and Allied Chemicals (all > 98.7% pure; Birkenhead, Merseyside, UK); and all other pesticides (all > 97% pure) were purchased from Sigma Aldrich (Poole, Dorset, UK). Ethanol (> 99.7% purity) was obtained from VWR International Ltd. (Leicestershire, UK). All test compounds were dissolved in ethanol to make stock solutions to be used in the assays.

8MDA-kb2 cells are human breast cancer cells stably transfected with a firefly luciferase reporter gene that is driven by an androgen-response element–containing promoter. Details of the modified assay were published previously. Briefly, cells were seeded at a concentration of 1 × 105 cells/mL in phenol red–free Leibowitz-15 medium (Invitrogen Ltd., Paisley, UK) containing 10% (charcoal-stripped) fetal calf serum (Invitrogen Ltd.) in white luminometer plates and allowed to attach for 24 hr. Cells were then exposed to eight serial dilutions of selected pesticides with or without DHT (0.25 nM). After 24 hr, luciferase activity was determined with SteadyGlo assay reagent (Promega UK Ltd., Southampton, Hampshire, UK) and measured in a plate reader (FLUOstar Optima, BMG Labtech GmbH, Offenburg, Germany). The following controls were run on each plate: media, ethanol, DHT coexposure (0.25 nM), DHT serial dilutions (0.002–10 nM), and flutamide (0.013–8 μM) or procymidone (0.005–3.2 μM) serial dilutions. All concentrations were tested in duplicate over two plates, and each pesticide was measured at least twice in separate experiments. For comparative purposes, luminescence was normalized to DHT alone at coexposure concentration (maximum response, 100%) and solvent-only (ethanol) controls (minimum response, 0%). Initially, flutamide was used as the internal quality control for antiandrogenicity; however, because of overlap of toxic effects on the cells with antiandrogenic activity, it was replaced by procymidone, which is more potent [IC50 (50% concentration that inhibits): flutamide, 1.56 μM; procymidone, 0.53 μM] but nontoxic to MDA-kb2 cells in the concentration range associated with receptor antagonism. Pesticides were initially tested over a concentration range of 0.64 nM–50 μM (5× dilutions) as a range-finding exercise. Subsequently, the concentration ranges were modified to reflect the potency and toxicity of each individual compound. Because cytotoxic effects could not be distinguished from antiandrogenic effects in the coexposed treatments, any readings of the pesticide statistically significantly below the mean ethanol control level (0%) were considered toxic to MDA-kb2 cells, and the corresponding coexposure data were not classified as antiandrogenic. Sixty percent of the pesticides were repeat tested using the same product but with new stock solutions and by a different experimenter.

9The methods for the YAS have been described previously. Briefly, stimulation of the transfected AR causes a color change in the media, which is measured by absorbance at 540 nm (Labsystems Multiskan Multisoft, Vienna, VA, USA). Plates were also measured at 620 nm to measure cell growth (turbidity) to check for any cytotoxic effects that may have occurred. Pesticides were coincubated with DHT (6.4 nM). Controls run in each experiment were ethanol, DHT serial dilutions (0.0026–100 nM), and flutamide serial dilutions (0.19–100 μM). The pesticide concentration range varied according to potency observed in MDA-kb2 assay but was between 0.016 and 750 μM for all test compounds. Incubation time was 53 hr at 28°C. Where turbidity readings were significantly depressed, toxicity was indicated and the effect could not be considered antiandrogenic; therefore, these dilutions were removed from analysis. Pesticide serial dilutions were tested in duplicate over two plates and were tested in two separate experiments.

10To analyze antiandrogenic action, raw luminescence readings were normalized on a plate-by-plate basis to the means of the positive DHT controls (n = 8) and the solvent controls (n = 8). We pooled all data from the same test compound and conducted statistical concentration–response regression analyses using the best-fit approach. Specifically, a variety of nonlinear regression models were fitted independently to the same data set, and the best-fitting model was selected using a statistical goodness-of-fit criterion. Concentration–response data from different researchers were first analyzed one by one using regression models, and differences in regression analyses due to data from different researchers were judged as statistically significant when the 95% CIs of the regression curves did not overlap. Such statistical differences between researchers were not observed, and thus data were pooled for final analysis. Luminescence readings from pesticides tested in the absence of DHT were divided by the mean of the solvent controls from the same plate and analyzed for negative and positive trends (suggestive of cytotoxic or androgenic action, respectively) by nonparametric contrast tests. Data considered to be statistically significant at p < 0.05 were analyzed using the best-fit approach as described above. All statistical analysis was performed using SAS statistical software (SAS Institute Inc., Cary, NC, USA). From the best-fitting model, we derived inhibitory concentrations for antiandrogenicity and effect concentrations for cytotoxicity.

11We derived the within-plate variation from readings of the positive DHT controls as a coefficient of variation (CV), with 95% of all CVs falling between 2.1% and 12.9% (mean, 6.5%). Of the 37 tested compounds, 24 pesticides were antiandrogenic in the MDA-kb2 assay, 9 of which are newly described (Table 1, Figure 1). The most potent in vitro AR antagonist was fenitrothion (IC20 = 0.098 μM), and the least potent was pyrimethanil (IC20 = 27.2 μM). All 14 compounds previously reported in the literature as antiandrogenic were confirmed using our test system. Two of 4 previously untested pesticides that were predicted to be AR antagonists in the QSAR were positive in the MDA-kb2 assay, and 3 of 13 pesticides that could not be predicted using the QSAR (i.e., they were out of the model domain) were also antiandrogenic. Five of 6 pesticides predicted to be inactive based on the QSAR were AR antagonists in the MDA-kb2 assay, but 3 were out of the QSAR prediction range because they were antiandrogenic at a concentration higher than the exclusion criterion of the QSAR (limit of detection, IC25 ≤ 10 μM; IC20: cyprodinil, 15.1 μM; pyrimethanil, 27.2 μM; cyhalothrin, 23.1 μM).

12All 14 pesticides tested using the YAS were antiandrogenic, including two that lacked activity in the MDA-kb2 assay [tolylfluanid (out of domain of QSAR) and bifenthrin (predicted active in QSAR)] (Table 1).

13Twenty-two of the 37 pesticides analyzed in the MDA-kb2 assay were cytotoxic. The concentrations required to elicit cytotoxicity were between 2.1 times (quinoxyfen) and 50 times (bromopropylate) higher than the concentrations associated with antiandrogenicity [based on the ratio of EC20 (concentration that produces a 20% effect) for cytotoxicity and IC20 for antiandrogenicity]. Seven of the chemicals analyzed in the MDA-kb2 assay showed AR agonist activity when tested in the absence of DHT coexposure, including two (cyprodinil and chlorpropham) with androgenic activity occurring at lower concentrations than antiandrogenic activity (Table 1, Figure 1). Four of 14 pesticides were cytotoxic in the YAS assay (cyprodinil, pyrimethanil, tolylfluanid, and difenoconazole), whereas we observed no AR agonism in this assay (Table 1).

14Our results indicate that systematic testing for antiandrogenic activity of currently used pesticides is urgently required. For example, 20 of the 50 pesticides with the highest exposure scores were antiandrogenic in at least one assay, including 8 that have not been identified as antiandrogens previously [see Supplemental Material, Figure 2 (doi:10.1289/ehp.1002895)]. In previous in vitro screenings of current-use pesticides, proportions of antiandrogenic pesticides were broadly similar [32% (52 of 161),; 50% (6 of 12),; 62% (38 of 61), ], further supporting the possibility that a large fraction of untested pesticides may be antiandrogenic. In contrast, estrogenic activity appears to be less common in current-use pesticides [18% (29 of 161),; 0% (0 of 100),; 0% (0 of 12), ]. Some discrepancy between our data and published data exists; for example, pirimiphos-methyl was previously reported to have no antiandrogenic activity, and chlorpropham has been reported to have no activity and to be antiandrogenic. These differences are most likely due to differences among the assay systems used. We also observed differences between findings based on the MDA-kb2 assay and the YAS assay. However, IC20 values based on the two assays never deviated by more than one order of magnitude, with the exception of two pesticides (tolylfluanid, bifenthrin) that were cytotoxic in the MDA-kb2 assay, and dimethomorph, for which we observed a large divergence in AR antagonist activity (IC20: MDA-kb2, 0.263 μM; YAS, 38.5 μM).

15We did not design our study to evaluate the QSAR by, and the number of chemicals falling within the applicability domain of the model was low; however, we note that several pesticides with antiandrogenic activity in vitro were not predicted by the QSAR, in part because some of the compounds were less potent than the prediction domain of the QSAR, which classifies chemicals with an IC25 > 10 μM as devoid of antiandrogenicity. The large percentage of pesticides for which the QSAR was not able to provide predictions (45 of 64) suggests that extending the applicability domain would increase the usefulness of the model.

16The ranking according to our exposure scoring system was similar to the listed “adjusted theoretical maximum dietary intake” of pesticides (58% concordance among the top 40 compounds) previously reported by, which is based on actual French consumption data and maximum residue levels. Consequently, the ERR was similar, using either our exposure scores or the adjusted theoretical dietary intake published by [see Supplemental Material, Table 2 (doi:10.1289/ehp.1002895)]. Both our exposure data and those used by were sourced from before 2008 (except JMPR reports from 2008 and 2009) and therefore may not be fully representative of current exposures. Indeed, from 2005 through 2010, the authorizations for use granted by European Union authorities expired for 12 of the tested pesticides, including several in vitro AR antagonists (procymidone, prochloraz, vinclozolin, ethoxyquin, endosulfan, azinphos-methyl, bromopropylate, dicofol, and fenitrothion) and 3 without evidence of antiandrogenic activity (bifenthrin, propargite, and profenofos). Thus, exposure to some of the tested compounds should decrease, whereas exposure to replacement products may increase. For example, a pesticide formulation called Switch, which contains cyprodinil and fludioxonil (both of which were antiandrogenic in our test system), was recommended as a replacement for the vinclozolin formulation Ronilan.

17To our knowledge, except for two reports to date, there is a complete absence of published human biomonitoring data for pesticides in Europe, and therefore, it is impossible to predict how the levels eliciting an effect in vitro may correspond to human internal concentrations. Similarly, although the National Health and Nutrition Examination Survey (NHANES) in the United States incorporates human biomonitoring of pesticides, exposure concentrations in human target tissues are very poorly understood, because of the almost complete lack of toxicokinetic data, short half-lives of current use pesticides, unspecific urinary metabolites, and unknown metabolic pathways (see ). Pesticides with relatively large ERRs, including dimethomorph (expiration of European Union authorization, September 2017), fludioxonil (October 2018), fenhexamid (May 2011), imazalil (December 2011), linuron (December 2013), ortho-phenylphenol (December 2019), tebuconazole (August 2019), and pirimiphos-methyl (September 2017), may be important antiandrogenic pollutants at present and in the future (Table 1). Linuron and tebuconazole are known in vivo antiandrogens; however, data on the other pesticides are much more limited. This is especially true of dimethomorph, fludioxonil, and fenhexamid, for which we were unable to identify previous publications regarding endocrine disruption. These compounds are newly formulated fungicides (dimethomorph, 2007; fludioxonil, 2008; fenhexamid, 2001), which are stable on food commodities (> 70% of the parent compound) and remain unchanged on the commodity when reaching the consumer. Dimethomorph and fenhexamid belong to the fungicide group of sterol biosynthesis inhibitors, as do the in vivo antiandrogenic conazoles (e.g., ) and imidazoles. A study of the sterol biosynthesis inhibitors imazalil, propiconazole, triadimefon, triadimenol, and prochloraz indicated that all inhibited aromatase in human placental microsomes, but to our knowledge, effects of dimethomorph and fenhexamid on steroidogenesis in mammalian cells have not been assessed. Imazalil and the in vivo antiandrogen prochloraz are both classified as imidazole fungicides, and in vitro potency estimates for the two compounds were similar (IC20: imazalil, 3.23 μM; prochloraz, 2.39 μM), but the possible effects of imazalil in vivo have not been evaluated. Therefore, it is our view that dimethomorph, fludioxonil, fenhexamid, and imazalil should be tested in vivo as a matter of urgency. Another relevant pesticide may also be ortho-phenylphenol, which is used as a fungicide in agriculture and as a wood preservative, and also has a wide variety of industrial applications (e.g., preservation of glues, plastic additives in flame retardants, disinfectant in hospitals). In our exposure ranking system, it ranked 12th out of 37 test compounds (Table 1). Considering that ortho-phenylphenol was highly ranked by exposure and that nonagricultural sources were absent from our exposure scores, it is not surprising that it was detected in all human urine samples tested in two studies [mean concentration, 2.9 nM, n = 30 samples; 35.2 nM, n = 22 samples], 85% of breast milk samples [mean concentration, 10.6 nM, n = 20 samples], and 30% of amniotic fluid samples [mean concentration, 0.76 nM, n = 20 samples] in the United States. ortho-Phenylphenol was previously identified as a receptor-mediated antiandrogen, but no data are available on its possible effects in vivo. Pirimiphos-methyl is an organothiophosphate insecticide that is stable on stored grain (< 24 weeks, 70% unchanged parent compound; ). There are also indications that it may be antiandrogenic in vivo because maternal and postnatal exposure of rats to 12 mg/kg body weight/day caused testicular tubular atrophy. In addition, treatment of adult male rats for 90 days resulted in decreased sperm density and mobility (125 mg/kg body weight/day), testicular atrophy (lowest observed adverse effect level, 41.67 mg/kg body weight/day), and decreased fertility (125 mg/kg body weight/day). There is insufficient evidence to assess the risk of tested pesticides to human health because of a lack of data. However, to our knowledge, all of the pesticides (with the possible exception of fenitrothion;; ) identified as in vitro AR antagonists in our study have also been reported to have antiandrogenic effects in vivo in animal models. We also identified 7 compounds that appeared to be androgenic because they stimulated activity in the absence of DHT. The mechanism of action for this response is not well characterized; however, it has been previously detected in this assay and was proposed to be due to conformational change of the ligand-binding pocket in such a way that simultaneous androgenic and antiandrogenic activities were possible. We are unable to confirm or reject these data; however, preliminary data from our laboratory suggests that the stimulatory response is neither via stimulation of the receptor, because we have not observed evidence of androgenic effects in the YAS for any compounds, nor due to cell proliferation, as evidenced by transient transfection of cells with a nonandrogenic responsive element. Cyprodinil and chlorpropham were more potent AR agonists (EC20 = 1.91 and 2.67, respectively) than antagonists (IC20 = 15.1 and 7.66, respectively) in the MDA-kb2 assay.

18In addition to identifying new candidate antiandrogens, our findings highlight important data gaps that prevent accurate assessment of male reproductive health risks from pesticides. The most important of these are the absence of in vivo studies and human biomonitoring data for environmentally relevant pesticides. In addition, fungicides typically had high exposure scores and were thus well represented in the testing set, presumably because they are often applied just before or after harvest to food commodities. They are typically applied as mixtures in order to increase effectiveness and prevent development of resistant strains, and therefore, human exposure to mixtures of these in vitro antiandrogens may be considerable. The contribution of pesticides to declining male reproductive health requires further investigation, particularly to clarify the relationship between effective concentrations in vivo and exposure.