Combined in silico and in vitro approaches to identify P‐glycoprotein‐inhibiting pesticides

The P‐glycoprotein (P‐gp) efflux pump plays a major role in xenobiotic detoxification. The inhibition of its activity by environmental contaminants remains however rather little characterised. The present study was designed to develop a combination of different approaches to identify P‐gp inhibitors among a large number of pesticides using in silico and in vitro models. First, the prediction performance of four web tools was evaluated alone or in combination using a set of recently marketed drugs. The best combination of web tools—AdmetSAR2.0/PgpRules/pkCSM—was next used to predict P‐gp activity inhibition by 762 pesticides. Among the 187 pesticides predicted to be P‐gp inhibitors, 11 were tested in vitro for their ability to inhibit the efflux of reference substrates (rhodamine 123 and Hoechst 33342) in P‐gp overexpressing MCF7R cells and to inhibit the efflux of the reference substrate rhodamine 123 in the Caco‐2 cell monolayer. In MCF7R cell assays, ivermectin B1a, emamectin B1 benzoate, spinosad, dimethomorph and tralkoxydim inhibited P‐gp activity; ivermectin B1a, emamectin B1 benzoate and spinosad were determined to be stronger inhibitors (half‐maximal inhibitory concentration [IC50] of 3 ± 1, 5 ± 1 and 7 ± 1 µM, respectively) than dimethomorph and tralkoxydim (IC50 of 102 ± 7 and 88 ± 7 µM, respectively). Ivermectin B1a, emamectin B1 benzoate, spinosad and dimethomorph also inhibited P‐gp activity in Caco‐2 cell monolayer assays, with dimethomorph being a weaker P‐gp inhibitor. These combined approaches could be used to identify P‐gp inhibitors among food contaminants, but need to be optimised and adapted for high‐throughput screening.

Pesticides include a wide variety of chemicals and are used in many applications such as agriculture and residential settings. [1]Thus, the general public is very frequently exposed to pesticides and notably to their residues, present in food. [2]However, a link has been established between exposure to pesticides and the incidence of various human diseases. [1,3]One of the contributing factors influencing pesticide toxicity is their interactions with plasma membrane transporters. [4]deed, inhibiting the activities of these transporters could lead to adverse effects not only by impairing the pharmacokinetics of drugs but also by blocking the transport of endogenous substrates. [5,6]One of the most important plasma membrane transporters is the ATP-binding cassette (ABC) efflux pump P-glycoprotein (P-gp) (ABCB1/MDR1), which limits the absorption of its substrates at the intestinal level, supports their biliary and renal elimination and protects sensitive tissues. [7]veral studies have described the modulation of P-gp activity by pesticides using in vitro assays.[17][18][19] Finally, endosulfan and diazinon remain ambiguous because studies have reported them either as a P-gp inhibitor or not depending on the in vitro assay used. [13,15,16,20,21]The diversity of the approaches used, sometimes in association with methodologies of little relevance, could partially explain the discrepancies observed between the results of these various studies.Moreover, many pesticides currently in the market are untested as to their interactions with membrane transporters, as regulatory studies do not include such tests.Therefore, current knowledge about potential P-gp inhibition by pesticides is limited and calls for extensive characterisation. [4]e present study evaluated which pesticides are P-gp inhibitors by combining different approaches.Due to the large number of these chemicals and to prioritise them for their later evaluation in vitro, we first used an in silico method to predict their potential inhibition of P-gp activity.The selected pesticides were then tested using different in vitro models to identify potential P-gp inhibitors.
Accumulation and retention assays were performed using the P-gpoverexpressing MCF7R cell subline.The apparent permeability (P app ) assay was carried out with the Caco-2 cell monolayer, a relevant intestinal model whose functional and morphological characteristics are similar to those of enterocytes.

| Selection of an in silico web tool to predict P-gp-inhibiting pesticides
We selected the drugs approved by the Food and Drug Administration (FDA) over the 2010-2020 period from the FDA website (www.accessdata.fda.gov/scripts/cder/daf/index.cfm) and assessed in vitro for their P-gp activity inhibition.Data about this P-gp inhibition, including half-maximal inhibitory concentration (IC 50 ) values, were collected from drug reviews freely available on the FDA website cited above.In the present study, each drug was classified as a P-gp inhibitor if its IC 50 was lower than 100 μM, and as a P-gp noninhibitor if its IC 50 was higher than 100 μM or if no P-gp activity inhibition was observed.Thus, 64 drugs were classified as P-gp inhibitors and 31 as P-gp noninhibitors (Table S1).
For each individual or combination of web tool(s), the area under the receiver operating characteristics curve (ROC AUC) was determined to be the first prediction performance parameter using GraphPad Prism software (version 5.0; GraphPad Software Inc.) as previously described. [26]The Matthews correlation coefficient (MCC) was then calculated as a second prediction performance parameter.
The MCC measures the correlation between predictions and real values. [26]Thereafter, the prediction of P-gp activity inhibition by 762 pesticides belonging to various chemical classes was evaluated using the most efficient individual web tool or combination of tools.
The culture medium was changed three times a week, and cells were used on Day 7 postseeding.
For P app assays, Caco-2 cells were seeded at 2.7 × 10 5 cells/cm² on polyester membrane inserts (0.4 μm pore size, 12 mm diameter) purchased from Corning.The culture medium was changed three times a week, and cells were used on Days 22-24 postseeding.

| Accumulation and retention assays
P-gp activity was evaluated by measuring the cellular accumulation or retention of a reference substrate in the presence or absence of either a pesticide or a reference inhibitor used at a noncytotoxic concentration (Table S2).For accumulation assay, cells are incubated with the reference substrate and the tested pesticides or the reference inhibitor.During the period of accumulation, the substrate enters the cells but is also concomitantly effluxed by P-gp, so inhibition of P-gp by the tested pesticides or the reference inhibitor increases the substrate accumulation.The substrate accumulation is determined at the end of this period.For retention assay, cells are incubated only with the reference substrate (loading period); thereafter, cells are incubated in a substrate-free medium with the tested pesticides or the reference inhibitor (efflux period).During this efflux period, substrate is exclusively effluxed by P-gp (decrease of intracellular concentration of the reference substrate), so inhibition of P-gp by the tested pesticides or the reference inhibitor blocks this efflux and limits the decrease of the intracellular concentration of the reference substrate.The intracellular concentration of the reference substrate is determined at the end of the efflux period.Assays were performed in a transport buffer (NaCl 136 mM, KCl 5.3 mM, KH 2 PO 4 1.1 mM, MgSO 4 0.8 mM, CaCl 2 1.8 mM, HEPES 10 mM, D-glucose 11 mM, pH 7.4).Rh123 and H33342 were used as reference substrates for the accumulation and retention assays, respectively.
Elacridar and verapamil were used as reference inhibitors for the accumulation and retention assays, respectively.Assays were performed as previously described with the following modifications. [28]The cell lysate was only used for measuring fluorescence.
The retention assay using H33342 was used to measure P-gp activity (rather than breast cancer resistance protein (BCRP/ABCG2) activity).
The experiments were performed on both MCF7 and MCF7R cells.
The final concentration of solvent (DMSO, MeOH or ACN) for all the solutions was set at 0.3% (vol/vol).Vehicle controls containing DMSO 0.3%, MeOH 0.3% or ACN 0.3% were included for each experiment.The fluorescence of Rh123 (ex: 485 nm; em: 520 nm) and H33342 (ex: 355 nm; em: 460 nm) was measured with a microplate-reading spectrofluorometer (FLUOstar OPTIMA; BMG Labtech).For each independent experiment (biological replicate), the median of three technical replicates was calculated, and then this median value was expressed as a percentage of that of the vehicle control.Three independent experiments were performed.
When the accumulation or retention of the reference substrate was greater than 150% (compared with the vehicle control), the pesticide was considered to be a P-gp inhibitor.Therefore, experiments complementary to the accumulation and retention assays were performed as described above, using various concentrations of the pesticides considered to be P-gp inhibitors to determine IC 50 (Table S3).For each independent experiment (biological replicate), whatever the concentration of pesticide.Three independent experiments were performed.The efflux ratio (ER) of Rh123 was then calculated according to the following equation:

| Apparent permeability assay
The integrity of the Caco-2 cell monolayer was monitored by measuring its transepithelial electrical resistance (TEER) at 37°C both before and after the P app assay with a Millicell Electrical Resistance System supplied by Merck.For this, ATB was loaded in the A (700 μL) and B (1500 μL) compartments.The Caco-2 cell monolayer was then equilibrated for 30 min at 37°C.Thereafter, the TEER was measured twice per insert, and a cell-free insert was included.The TEER value was expressed as Ω cm 2 and calculated according to the following equation: with R M (Ω) the resistance mean value of the Caco-2 cell monolayer on insert, R 0 (Ω) the resistance mean value of the cell-free insert and S (cm²) the surface area of the insert.Only Caco-2 cell monolayers with a TEER > 250 Ω cm 2 before and after the P app assay were used because this threshold value indicates adequate integrity of the Caco-2 cell monolayer. [29,30]Three independent experiments were performed with one technical replicate for each independent experiment.

| Statistical analysis
Statistical analyses were performed using GraphPad Prism software.
For accumulation and retention assays with only one applied concentration, data were analysed using the one-sample t test, with '100' as the theoretical mean.Means were considered significantly different from 100 at p < 0.05.For IC 50 , P app and ER values, an analysis of variance was performed.When the experimental condition effect was significant (p < 0.05), the values were compared with each other using Bonferroni's test.Differences were considered significant at p < 0.05.The values presented are mean ± SEM.

| Selection of an in silico web tool and application to predict P-gp-inhibiting pesticides
We used 64 drugs classified as P-gp inhibitors and 31 drugs classified as P-gp noninhibitors according to FDA drug reports to evaluate the prediction performance of four web tools as to P-gp activity inhibition by determining the ROC AUC and the MCC as parameters.
We obtained ROC AUC values of the web tools individually and in combination (Table 1).Therefore, MCC values of the AdmetSAR 2.0/PgpRules/pkCSM combination were determined for the different cut-offs (Table 2).The MCC value was equivalent between cut-offs 1 and 2, and higher than that obtained with cut-off 3.
Based on the ROC AUC and MCC values, we used the AdmetSAR 2.0/PgpRules/pkCSM combination of web tools with cut-offs of 1 and 2 to evaluate the prediction of P-gp activity inhibition by 762 pesticides.The results indicated that 421 and 187 pesticides were predicted to be P-gp inhibitors with cut-offS 1 or 2, respectively.Therefore, we selected the results obtained with cut-off 2 to limit the number of pesticides to evaluate for the in vitro assays (Table S4).Among these 187 pesticides, we selected nine (cyflumetofen, dimethomorph, fenpicoxamid, ivermectin B1a, oxathiapiprolin, pinoxaden, profoxydim, silafluofen, spinosad) based on their representativeness of the different chemical families, their solvent solubility, their market availability and the availability of P-gprelated data in the literature.We also selected two more pesticides (emamectin B1 benzoate and tralkoxydim) as they inhibited P-gp activity (own laboratory data) yet had been predicted as noninhibitors with the AdmetSAR 2.0/PgpRules/pkCSM combination.

| P-gp activity inhibition by pesticides based on accumulation and retention assays
There were no morphological changes in the cells after exposure to pesticides (image not shown).Results on MCF7 cells indicated no variation in the intracellular concentration of Rh123 or H33342 in the presence of either the tested pesticides or the reference inhibitor compared with the vehicle control (data not shown).This indicated that there was little or no P-gp activity in MCF7 cells.However, in MCF7R cells, with the reference inhibitor elacridar, we observed a significant increase in Rh123 intracellular concentration (270 ± 2% compared with the vehicle control) (Figure 1A).With a concentration greater than 150%, this result confirmed P-gp activity in MCF7R cells and elacridar's ability to inhibit it.Five of the 11 tested pesticides significantly increased Rh123 intracellular concentration: dimethomorph (165 ± 4%), ivermectin B1a (308 ± 19%), spinosad (297 ± 19%), emamectin B1 benzoate (298 ± 31%) and tralkoxydim (178 ± 11%).
IC 50 values were determined for the five pesticides being considered based on the Rh123 accumulation assay (Figure 2A).

| P-gp activity inhibition by pesticides based on the apparent permeability assay
The pesticides affected the P app of Rh123 to different degrees (Table 3).The P app A-B of Rh123 was similar for all the tested compounds (ranging from 7.8 ± 0.2 × 10 −7 to 9.5 ± 0.4 × 10 −7 cm s −1 ).
In contrast, the effects on P app B-A of Rh123 were stronger.
With verapamil, the P app B-A of Rh123 was significantly lower −47%) but still significant, while no difference was observed with tralkoxydim.The compounds tested affected the ER (Figure 3).In comparison with MeOH (8.3 ± 0.5) and DMSO (8.0 ± 0.3), there was a significant decrease (−70%) with verapamil (2.4 ± 0.1), thus confirming verapamil's ability to inhibit P-gp activity.We also observed a significant decrease in ER with spinosad (2.4 ± 0.   Using a combination of three web tools, our results indicated that out of 762 pesticides, 187 were predicted to be P-gp inhibitors.Among the 11 pesticides tested in vitro, between four and five (depending on the assay used) were determined to be P-gp inhibitors.
T A B L E 3 P app of Rh123 in Caco-2 cell monolayers in the presence of pesticides.An evaluation of the web tools' prediction performance revealed that the best combination was AdmetSAR 2.0/PgpRules/pkCSM; this combination had ROC AUC and MCC values of, respectively, 0.791, 0.47 or 0.46 (depending on the cut-off applied).These results indicated a correct prediction, as in the literature an ROC AUC value between 0.7 and 0.8 is considered acceptable, and an MCC value of 0.5 corresponded to 75% of cases being correctly predicted. [31,32]deed, the correct positive and negative predictions represented 78% and 77% of the 95 drugs with either cut-off 1 or 2, respectively.
Nevertheless, a nonnegligible number of drugs was not correctly predicted as already reported in a study using web tools to predict Pgp substrates. [26]Using cut-off 1, the false positives and false negatives represented 20% and 2% of the drugs, respectively, while they represented 12.5% and 10.5% when using cut-off 2. This relative poor prediction performance by the web tool combination was confirmed in our experiments.Indeed, among the nine pesticides predicted to be P-gp inhibitors and evaluated in vitro, six (cyflumetofen, fenpicoxamid, oxathiapiprolin, pinoxaden, profoxydim and silafluofen) failed to inhibit P-gp activity in MCF7R cells.We were not able to corroborate our results with other studies as these six pesticides had never been evaluated previously.Other pesticides were predicted to be P-gp inhibitors by the web tool combination, but data in the literature reported that they were not able to inhibit P-gp activity in vitro.This was notably the case of numerous pyrethroids such as deltamethrin, tested using a variety of assays. [16,28]Moreover, the two pesticides predicted to be P-gp noninhibitors (emamectin B1 benzoate and tralkoxydim) and tested in vitro in our study were in fact able to inhibit P-gp activity in MCF7R cells.Emamectin B1 benzoate also inhibited P-gp activity using the P app assay with the Caco-2 cell monolayer.Our results confirmed previous experiments describing this pesticide as a P-gp inhibitor in vitro. [14]The case of tralkoxydim is more ambiguous, as this pesticide inhibited P-gp activity when using the MCF7R cell model but not when using the Caco-2 cell monolayer model.Moreover, it had not been previously studied in vitro.Another pesticide, phosalone, was predicted to be a P-gp noninhibitor by our combination of web tools, but a previous study had reported that this chemical inhibited P-gp activity in vitro. [15]Only three pesticides (ivermectin B1a, dimethomorph and spinosad) predicted in silico to be P-gp inhibitors were confirmed by our in vitro assays with both MCF7R cell and Caco-2 cell monolayer models.Nevertheless, predictions by our combination of web tools were correlated for other pesticides such as tebufenozide and methoxyfenozide, described as P-gp inhibitors by a previous study, [33] but also imidacloprid and nitenpyram reported in vitro to be P-gp noninhibitors. [18] report here for the first time that cyflumetofen, fenpicoxamid, oxathiapiprolin, pinoxaden, profoxydim and silafluofen were not able to inhibit P-gp activity in vitro, these pesticides never having been evaluated previously to our knowledge.The logP value (ranging from 3 to 6.5) of these pesticides being equivalent to that of the P-gp inhibitor pesticides, we could exclude a nonentry into the cells to explain this absence of P-gp inhibition.In the present study, the pesticides dimethomorph, ivermectin B1a, spinosad and emamectin B1 benzoate were found to be P-gp inhibitors using two complementary in vitro models, while tralkoxydim was able to inhibit P-gp activity only in MCF7R cells.36] These discrepancies between the studies could be explained by the different assays and reference substrates used.Nevertheless, our results were in accordance with these data.The stronger inhibition of P-gp activity by ivermectin B1a, spinosad and emamectin B1 benzoate was also observed with the P app assay.Indeed, these three pesticides had ER values (on average 2.4) close to those of the reference inhibitor, verapamil.The ER value was slightly higher for dimethomorph (3.9), and no different from the negative control for tralkoxydim (7.1).Previous studies have also reported a lower ER with ivermectin and spinosad in Caco-2 and MDCK-MDR1 cell monolayers. [11,34]This lower ER value was basically explained by a decrease in P app B-A for ivermectin B1a, spinosad and emamectin B1 benzoate, while P app A-B remained stable whatever the experimental conditions.Indeed, P app B-A decreased for the vehicle control from 63 × 10 −7 to on average 20 × 10 −7 cm s −1 for these three pesticides, our results being in agreement with the literature. [11]For dimethomorph, the P app B-A value was, respectively, 33 × 10 −7 cm s −1 , thus confirming its lower potency in inhibiting the P-gp activity.
In the context of environmental exposure, P-gp activity was unlikely to be inhibited by the tested pesticides according to the FDA criteria for in vivo drug-drug interactions.Indeed, intestinal P-gp activity in vivo is likely to be inhibited if the human luminal gut concentration of the compound exceeds or equals 10 times the IC 50 (Table 4).As IC 50 values are usually over 1 μM, this would require greater pesticide concentrations, which would be unlikely to be reached for the tested pesticides in light of their environmental concentrations.The prediction of a potential inhibition of the activity of intestinal transporters in vivo has also been evaluated for other pesticides, such as allethrin, tetramethrin, malathion, parathion and chlorpropham. [19,28,37]These studies also predicted that intestinal transporter activity in vivo would not be inhibited by these pesticides in environmentally exposed humans.Nevertheless, both these previous studies and our own study evaluated each pesticide individually.However, this evaluation needs to be performed for mixtures (not only pesticide combinations but also pesticides with other environmental pollutants, drugs or compounds that naturally interact with P-gp).Pesticide combinations have already been shown to affect P-gp activity. [15]Exposure to such mixtures may consequently have to be considered when judging the in vivo relevance of transporter inhibition by pollutants, including pesticides. [4]Moreover, the study of P-gp expression regulation upon chronic exposure to pesticides mixtures is worthy of interest and would deserve further experiments.

| CONCLUSION
In the present study, we applied a combination of in silico and in vitro approaches to identify pesticides that inhibit P-gp activity.The combination of different in silico web tools was helpful in prioritising the pesticides to be tested in vitro.However, the prediction performance of the in silico models needs to be improved by b According to the European Union Pesticides database and to the Joint FAO-WHO expert committee on food additives.
c Defined for one meal out of two daily meals and a 70 kg body weight.
d Calculated as the ratio oral dose/250 mL.
e The lowest of the two calculated IC 50 values was retained.
the three technical replicates were expressed as a percentage of the median of three technical replicates of the vehicle control.The IC 50 was determined from a nonlinear regression based on a dose-response stimulation equation (log(agonist) vs. response [three parameters]) using GraphPad Prism software through the following equation: percentage of reference substrate in the cells (relative to the vehicle control) for a given concentration of pesticide, 100 the percentage of reference substrate in the cells for the vehicle control (arbitrarily set at 100%) and Y max the maximal percentage of reference substrate in the cells (relative to the vehicle control) app 0 with dC (μM) the final concentration of Rh123 in the acceptor compartment, V (cm 3 ) the volume of the acceptor compartment, C 0 (μM) the initial concentration of Rh123 in the donor compartment, dt (s) the time of incubation and S (cm²) is the surface area of the insert.
For individual web tools, values ranged from 0.555 (for vNN-ADMET) to 0.737 (for pkCSM).When the web tools were combined, values ranged from 0.622 (for AdmetSAR 2.0/vNN-ADMET) to 0.791 (for AdmetSAR 2.0/PgpRules/pkCSM).The combination of AdmetSAR 2.0/PgpRules/pkCSM had the best ROC AUC value, and different cut-offs can be applied: a drug can be considered a P-gp inhibitor if one, two, or three of the web tools predicted it to be a P-gp inhibitor.The term cut-off 1, 2 or 3 used hereinafter refers to this classification by one, two or three web tools.

F I G U R E 1
Intracellular concentration of rhodamine 123 (A) and Hoechst 33342 (B) in MCF7R cells in the presence of either pesticide (white bars) or a reference inhibitor (black bars).The pesticides were tested at 100 μM (for cyflumetofen, dimethomorph, fenpicoxamid, oxathiapiprolin, pinoxaden, profoxydim, silaflulofen and tralkoxydim), or 10 μM (for ivermectin B1a, spinosad and emamectin B1 benzoate), while the reference inhibitors elacridar and verapamil were tested at 10 and 100 μM, respectively.Values are shown as mean ± SEM and expressed as percentages of the vehicle control.Dotted line: 150%.When the concentration of rhodamine 123 or Hoechst 33342 was higher than 150%, the tested compound was considered to be a P-glycoprotein inhibitor.Three independent experiments were performed.*, **, ***: significantly different from the vehicle control (p < 0.05, p < 0.01 and p < 0.001, respectively).

F I G U R E 2
IC 50 of pesticides determined from the intracellular concentration of rhodamine 123 (A) and Hoechst 33342 (B) in MCF7R cells.From various concentrations of pesticides, intracellular concentrations of rhodamine 123 and Hoechst 33342 were expressed as percentages of the vehicle control, and the IC 50 was determined from a nonlinear regression based on a dose-response stimulation equation.Values are shown as means ± SEM.Three independent experiments were performed.a, b : bars without a common letter differ (p < 0.001).EC 50 , half-maximal effective concentration; IC 50 , half-maximal inhibitory concentration.
indicated a stronger inhibition of P-gp activity by ivermectin B1a, spinosad and emamectin B1 benzoate than by dimethomorph and tralkoxydim.These observations were confirmed by the IC 50 values, which ranged from to 2.5 to 14.4 μM for ivermectin B1a, spinosad and emamectin B1 benzoate, while IC 50 values for dimethomorph and tralkoxydim ranged from 79.6 to 102.1 μM.Previous experiments have indicated IC 50 values with respect to P-gp activity for ROC AUC values of web tools considered alone or in combination for predicting P-gp inhibiting FDA-approved drugs.
T A B L E 4 Prediction concerning inhibition of intestinal P-gp activity in vivo by luminal gut concentrations of pesticides according to FDA criteria.a ADI, admissible daily intake; FDA, Food and Drug Administration; I, intestinal luminal gut concentration; IC 50 , half-maximal inhibitory concentration; P-gp, P-glycoprotein.a The inhibition of intestinal P-gp activity in vivo can be predicted if the ratio I gut /IC 50 ≥ 10.