1. Introduction
Chemical pesticides constitute an important group of substances, exhibiting notably insecticide, rodenticide, fungicide, nematicide, or herbicide properties, which are used to control and repel pests in different fields. They can be grouped into various chemical families, including the pyrethroids, organochlorines, carbamates, organophosphorus compounds, triazines, and neonicotinoids. Due to their large use for agricultural, industrial, or domestic purposes, and also due to the strong persistence of some of them, pesticides appear as widely-distributed environmental pollutants [
1]. Humans are consequently highly exposed to these chemicals—notably by an oral way—through the consumption of food or water contaminated by pesticides [
2,
3,
4]. This is probably a major health issue, owing to the well-established toxicity of pesticides, which can trigger or favor various diseases including cancers and neurodegenerative, metabolic, reproductive, or developmental pathologies [
5,
6]. Such adverse effects are thought to reflect specific toxicodynamic properties of pesticides, such as agonistic or antagonistic effects towards endocrine receptors, genotoxicity, the blockage of ion channels, the inhibition of enzymes such as acetylcholinesterase, the impairment of the redox balance, or oxidative stress generation [
7,
8,
9].
Before exerting toxicodynamic effects, pesticides have however to be absorbed by exposed humans, notably at the gastro-intestinal level in response to oral exposure; neurotoxic ones have additionally to cross the blood-brain barrier [
10]. Such a consideration for intestinal absorption and brain distribution is also a major concern for drugs, notably during their industrial development [
11,
12]. In this context, the passage of drugs across the intestinal barrier and the blood-brain barrier is usually extensively characterized using
in vitro cellular models, animal experimentation, and clinical human pharmacokinetics studies before marketing authorization [
13,
14,
15]. The intestinal absorption and brain disposition of drugs can moreover be
in silico predicted with good accuracy, using various computer models [
16,
17,
18,
19]. Among them, SwissADME is a recent free web tool designed for predicting pharmacokinetics parameters [
20]. It notably concomitantly evaluates the intestinal and brain permeation of drugs through an accurate predictive model that works by computing the lipophilicity and polarity of drugs with good accuracies (83–93% for intestinal absorption and 90% for brain permeation [
21]). This physicochemical descriptors-based method, termed the BOILED-Egg method, permits a clear and informative graphical representation of the data, and can be applied in a variety of settings, from the filtering of chemical libraries at the early steps of drug discovery, to the evaluation of drug candidates for development [
21].
In contrast to drugs, pesticides are very poorly characterized with respect to their intestinal and brain passage in humans. For a very limited number of them, intestinal absorption has been characterized from pharmacokinetics studies performed with human volunteers [
22]. Some information may additionally be extrapolated from animal studies or
in vitro studies [
23]; however, the animal versus human bioavailability of chemicals is known to be poorly correlated [
24], making it difficult to precisely and unambiguously predict the human absorption of pesticides from animal data. The intestinal and brain permeations of pesticides are nevertheless important to evaluate, in order to fully apprehend their potential toxicity, as discussed above. In this context, the use of
in silico tools may be considered a valuable and robust approach. Indeed, the great number and variety of pesticides which have to be evaluated precludes the use of cost-effective and labor-consuming classical pharmacokinetics
in vitro or
in vivo methods. Clinical pharmacokinetics studies in human subjects with pesticides are moreover now hampered by appropriate ethical considerations [
25,
26]. The present study was therefore designed to evaluate
in silico the intestinal absorption and brain permeation of 338 pesticides belonging to various chemical classes, using the SwissADME web tool.
4. Discussion
Human intestinal absorption and human brain distribution correspond to key steps of pesticide toxicokinetics, but remain poorly characterized. In the present study, we have used a recent and original web tool, i.e., SwissADME, previously developed and validated for drugs [
20], for investigating
in silico the human intestine and brain permeation of 338 various pesticides belonging to the main chemical classes of pesticides. Our results showed that a large proportion of pesticides (more than 80%) is predicted to be highly absorbed by the human gastro-intestinal tract. This supports the idea that human exposure to pesticides through the oral way, which commonly occurs via the ingestion of contaminated food or water, is likely to result in an intake of these environmental contaminants. This may ultimately favor their deleterious effects towards human health, including endocrine disruption and carcinogenic effects [
5]. In addition, a notable, although minor, proportion of pesticides (around 38%) is predicted to permeate the brain, which may support the well-established idea that pesticides have neurotoxic effects [
39].
The prediction of intestinal absorption of pesticides performed with the SwissADME/BOILED-egg method was interestingly validated with absorption data from human toxicokinetics studies, with an accuracy of 96%. Even if the available measured data from the human studies are unfortunately limited (
n = 25), and rather unbalanced, i.e., most of them (
n = 23) reporting high intestinal absorption of pesticides (
Table 4), this validation is worth noting. A similar high accuracy (83–93%) has also been reported for the prediction of human intestine absorption by the SwissADME tool for a large data set of drugs [
21]. In addition, pesticides predicted to display high intestinal permeation exhibit a profile of physicochemical parameters, i.e., reduced molecular weight, van der Waals volume, and polarizability, and increased solubility, which is well-known to favor the high intestinal absorption of drugs and chemicals [
70,
71,
72]. This supports the relevance of the
in silico prediction. In particular, the size of the drug molecule affects absorption. Smaller molecular weight drugs are consequently absorbed better compared to larger ones [
73]. Indeed, as molecular size increases, a larger cavity must be formed in water to be soluble. Increasing the size also impedes passive diffusion through the tightly packed aliphatic side chains of the lipid bilayer membrane. Moderately lipophilic drugs, i.e., drugs with LogP < 5, are additionally more absorbed than highly lipophilic drugs [
74]; in agreement with this rule,
in silico predicted high intestine-permeant pesticides have significantly lower LogP (estimated notably by the XlogP3 value in the present study) than their non-permeant counterparts. It should however be kept in mind that each physicochemical parameter, when separately considered, may fail to unambiguously predict intestine permeation, thus highlighting the relevance of combining two parameters as is done in the BOILED-Egg method. Finally, the
in vitro measurement of transport across human intestinal Caco-2 cells for some pesticides, such as the neonicotinoids acetamiprid and imidacloprid, have revealed high apparent permeability coefficients [
75,
76] consistent with
in vivo absorption with 100% efficiency, and thus with the
in silico prediction of high absorption for these two pesticides (
Table S1).
The prediction of brain permeation for pesticides cannot, unfortunately, be validated by human experimental pharmacokinetics data, owing to the quasi-absence of such data in the scientific literature and pesticide database. Nevertheless, the fact that the BOILED-Egg method predicts drug transfer across the blood-brain barrier with high accuracy [
21] supports the relevance of this method for other chemicals such as pesticides. In addition, some physicochemical parameters significantly associated with the prediction of brain permeation for pesticides, such as a reduced tPSA and the number of nitrogen and oxygen atoms, are well-established factors favoring the permeation of drugs into the brain. Indeed, drugs entering the brain typically have a tPSA value of less than 60–70 Å
2 [
77] and a number of nitrogen and oxygen atoms less than or equal to five [
38].
In silico predictions of brain passage were moreover supported by
in vitro or
in vivo data for some pesticides. The pyrethroid deltamethrin, predicted to be brain non-permeant (
Table S1), has thus been shown to accumulate relatively poorly in human brain endothelial hCMEC/D3 cells [
78]; whereas paraquat, also predicted to be brain non-permeant (
Table S1), is excluded by the blood-brain barrier in the primate rhesus macaque [
79].
Even if most of the pesticides are predicted to exhibit high intestine absorption, some of them, notably the organochlorines aldrin, chlordane, DDT, heptachlor, and lindane, are predicted to display low intestine absorption. Such intestine non-permeant organochlorines are highly lipophilic non-polar chemicals; they do not comprise H-bond acceptors or donors, in contrast to intestine-permeant organochlorines (
Figure S1). This lack of potential H-bonds for the non-permeant organochlorines is consistent with their low solubility, probably contributing to their predicted poor intestinal absorption. It is noteworthy that the marked lipophilicity of the organochlorines also results in long-term storage in fat tissue and low excretion [
80]. Therefore, the organochlorines, although some of them poorly absorbed, may nevertheless accumulate in an organism. These persistent organic pollutants may consequently reach cellular levels for which toxic effects, notably metabolic or endocrine effects, are suspected to occur [
81,
82]. In addition, the fact that the dietary intake of lipophilic organic chemicals such as organochlorines may occur primarily via adsorption to mixed dietary lipid micelles has to be considered [
83,
84]. Such a process may notably contribute to the intestine permeation of TCDD, thus explaining the discordance between the high measured intestine absorption of this organochlorine [
51] and the prediction of low absorption by the BOILED-Egg method (
Table 4).
It is noteworthy that the percentage of brain-permeant pesticides (less than 40%) is approximately twice as low as that of their intestine-permeant counterparts (more than 80%). This probably reflects the fact that the physicochemical criteria for entering the brain are more stringent than those required for intestinal absorption. In particular, concerning polarity, the tPSA value has to be ideally less than 60–70 Å
2 for brain permeation, whereas the tPSA threshold is higher, i.e., 120 Å
2, for oral absorption [
77]. Organophosphorus pesticides are consequently predicted, for most of them, to be not brain-permeant, owing to their relatively high tPSA values (
Figure 2); the bioactive metabolites of organophosphorus compounds, such as oxons-related molecules, are also rather poorly brain-permeant (
Table 1). Similarly, the carbamates and triazines exhibiting substantial tPSA values are predicted to not cross the blood-brain barrier (
Figure 2). It should be however kept in mind that the organophosphorus pesticides are suspected to cause neurotoxicity, including delayed neuropathies and long-term effects [
85,
86], which may appear as contradictory with the fact that they are predicted to be not brain-permeant. Such an apparent discrepancy may be due to the threshold retained for predicting a brain non-permeant chemical by the BOILED-egg method, i.e., LogBB < 0, whereas other studies have retained a less stringent threshold (LogBB < −1) for compounds poorly distributed in the brain [
87,
88,
89]. In this context, it may be hypothesized that at least some of the organophosphorus pesticides can in fact partly access the brain, with LogBB values comprised between −1 and 0 (−1 < LogBB < 0), which may be compatible with their neurotoxicity. The fact that a LogBB equal to −0.09 has been measured for the organosphosphorus parathion in pigs probably supports this hypothesis [
90]. Besides, organophosphorus pesticides such as chlorpyrifos and malathion directly alter the functional integrity of the blood-brain barrier [
91,
92,
93]. Such a disruption of this blood-tissue barrier may secondarily result in the enhanced passage of organophosphorus pesticides in the central nervous system.
It is noteworthy that the BOILED-Egg method-based prediction of drug passage across the intestine and the blood-brain barrier does not directly take into account the potential implication of the influx or efflux of drug transporters [
21]. However, there is now increasing evidence that such membrane transporters may interact with various pesticides [
94], including organophosphorus compounds, organochlorines, paraquat, and pyrethroids [
33,
95,
96,
97,
98]. In particular, the putative transport of pesticides at the intestinal level by ATP-binding cassette (ABC) efflux transporters such as P-glycoprotein (
ABCB1) and the breast cancer resistance protein (BCRP
, ABCG2) has probably to be taken into account [
99]. Indeed, the low doses of pesticides contained in food are unlikely to saturate these efflux pumps, in contrast to the high doses of administrated drugs. The P-glycoprotein and/or BCRP-mediated efflux of pesticides at the apical pole of intestinal cells may consequently efficiently prevent their absorption by actively repelling them into the lumen of the intestine. Solute carrier (SLC) transporters acting as influx transporters may also have to be considered for the intestinal transport of some pyrethroids [
100].
If intestinal absorption and brain penetration constitute key steps of pesticide toxicokinetics in response to oral exposure, other parameters have to be taken into account for more globally considering pesticide disposition in the human organism. This is notably the case for metabolism and/or excretion processes, which may occur rapidly and extensively, notably for first-pass hepatic metabolism, thus resulting in the relatively short half-lives of some pesticides. This may concern pyrethroids and organophosphorus insecticides, which are subjected to many metabolic pathways mediated by various xenobiotic metabolizing enzymes [
101,
102]. By contrast, organochlorine pesticides are thought to be less metabolized, and therefore appear to be more stable in the organism. Storage in some compartments, such as adipose tissue, may additionally play a major role in the toxicokinetics of pesticides, notably for the most lipophilic of them, such as the organochlorines [
103], as discussed above. Binding to proteins, including plasma proteins, is also a key parameter to consider, as only the free fraction of pesticides is presumed to exert toxicodynamic effects and to be handled by pharmacokinetics effectors such as drug transporters and drug metabolizing enzymes. The chirality of pesticides may also have to be taken into account, as reported for drugs [
104]. Entero-hepatic circulation, which occurs for some pesticides such as the carbamates and organochlorines [
103,
105], may additionally deserve attention. Moreover, the formulation of pesticides may constitute a feature influencing their intestinal absorption, notably for the organosphosphorus insecticides [
52,
106]. The fact that humans are usually exposed to a mixture of pesticides [
107] may also affect their intestinal absorption, owing to possible pesticide–pesticide interactions with respect to drug transporters, as already described for drug–drug interactions [
108]. Finally, the nature of the diet may interfere with the passage of pesticides across the intestinal barrier, as previously shown for the intestinal absorption of drugs [
109].
Computer-based methods based on molecular descriptors such as lipophilicity and polarity, initially developed for drugs [
16], may be useful for predicting not only pesticide intestine and brain permeation, as described in the present study, but also for evaluating other pharmacokinetics parameters such as metabolism and clearance, as well as the biological effects of pesticides. More globally,
in silico methods and computational systems biology probably represent promising ways for assessing pesticide toxicity and risk, notably in humans [
110,
111,
112,
113]. However, further studies are required to fully validate
in silico approaches, notably with respect to the realistic conditions of environmental exposure to pesticides, i.e., low and/or chronic doses of mixtures of pesticides. In particular,
in silico methods are likely to be compared with other non-animal approaches, such as
in vitro methods, for human toxicokinetics in risk evaluations of pesticides [
10,
114].