Structure-activity relationships for xenobiotic transport substrates and inhibitory ligands of P-glycoprotein.

The multixenobiotic resistance phenotype is characterized by the reduced accumulation of xenobiotics by cells or organisms due to increased efflux of the compounds by P-glycoprotein (P-gp) or related transporters. An extensive xenobiotic database, consisting primarily of pesticides, was utilized in this study to identify molecular characteristics that render a xenobiotic susceptible to transport by or inhibition of P-gp. Transport substrates were differentiated by several molecular size/shape parameters, lipophilicity, and hydrogen bonding potential. Electrostatic features differentiated inhibitory ligands from compounds not catagorized as transport substrates and that did no interact with P-gp. A two-tiered system was developed using the derived structure-activity relationships to identify P-gp transport substrates and inhibitory ligands. Prediction accuracy of the approach was 82%. We then validated the system using six additional pesticides of which tow were predicted to be P-gp inhibitors and four were predicted to be noninteractors, based upon the structure-activity analyses. Experimental determinations using cells transfected with the human MDR1 gene demonstrated that five of the six pesticides were properly catagorized by the structure-activity analyses (83% accuracy). Finally, structure-activity analyses revealed that among P-gp inhibitors, relative inhibitory potency can be predicted based upon the surface area or volume of the compound. These results demonstrate that P-gp transport substrates and inhibitory ligands can be distinguished using molecular characteristics. Molecular characteristics of transport substrates suggest that P-gp may function in the elimination of hydroxylated metabolites of xenobiotics. ImagesFigure 1. AFigure 1. BFigure 1. CFigure 1. DFigure 1. EFigure 1. FFigure 1. GFigure 1. HFigure 2.Figure 2.Figure 2.Figure 2.Figure 2.Figure 2.Figure 3. AFigure 3. B

P-glycoprotein (P-gp) is a 170 kDa protein, encoded by the human MDRI gene, which has been shown to efflux a wide variety of drugs such as the chemotherapeutic agents doxorubicin, vincristine, and vinblastine (1). P-gp in humans is located on the secretory surfaces of organs of elimination, such as the brush border of the renal proximal tubules, the canalicular membrane of hepatocytes, and the apical surface of mucosal cells in the small and large intestines (2). This tissue localization and direct experimental evidence (3) indicates that P-gp plays a role in eliminating xenobiotics from the whole organism. P-gp is also localized in hormone-producing and hormone-responsive organs such as the adrenal gland, the testes, and the placenta (2). Steroid hormones such as cortisol (4), corticosterone (5), and aldosterone (6) have been shown to be transport substrates for P-gp.
The involvement of P-gp in pesticide toxicity was shown with the anthelmintic agent ivermectin. Mdrla gene-knockout mice treated with the neurotoxic pesticide ivermectin demonstrated increased sensitivity to this pesticide and increased accumulation of this pesticide in the brain as compared to control mice (7). When C57BL/6 mice were cotreated with ivermectin and the P-gp inhibitors SDZ PSC833 or SDZ 280-446, the mice were hypersensitive to the neurotoxic effects of ivermectin compared to treat-ment with ivermectin alone (8). Additionally, rats treated with the pesticide chlorpyrifos showed increased P-gp protein levels in the bile duct, adrenal gland, stomach, jejunum, and kidney proximal tubules (9). Thus, P-gp appears to play a role in both the efflux of chemicals from cells and the elimination ofchemicals from the body.
Many compounds have been shown to inhibit P-gp function, and the use of P-gp inhibitors has received significant attention as a possible means of enhancing the efficacy of drugs transported by this protein by decreasing their rate of efflux (10). Holfsli and Nissen-Meyer (11) observed that 73% of 26 drugs assessed inhibited P-gp in vitro. Studies in our laboratory have shown that many pesticides can inhibit P-gp function (12). Although the structures of P-gp inhibitors differ significantly, common characteristics include a molecular mass greater than 300 (13,14) and lipophilicity (11). Additional characteristics associated with some P-gp inhibitors include amphipathic properties (15) and a positive charge at physiological pH (1,10,15). Molecular characteristics of transport substrates of Pgp are less well understood; however, P-gp inhibitors have been typically assumed to competitively inhibit transport substrates (11,16), suggesting that inhibitory ligands and transport substrates share similar molecular characteristics.
In a previous study, we assessed the ability of 38 pesticides to inhibit P-gp function (12). Based upon these results, we concluded that chemicals with the greatest potential to inhibit P-gp contained a cyclic structure within the molecule, a molecular mass of 391-490 Da, and a log Kow of 3.6-4.5.
Seven of the inhibitory pesticides were evaluated for susceptibility to transport by Pgp, but only one of the pesticides was transported by P-gp. These results demonstrated that many pesticides are capable of inhibiting P-gp function; however, inhibition does not imply that the pesticide is transported by P-gp. Thus, transport substrates of P-gp may have significantly different molecular characteristics than inhibitory ligands.
In the present study, we significantly refined our structure-activity analysis of Pgp inhibition by pesticides. In addition, the structural attributes of P-gp inhibitors were compared to those of known P-gp substrates in order to define characteristics that dictate whether a chemical will either inhibit or be transported by P-gp. Six new pesticides were then subjected to the structure-activity analyses in order to predict the level of interaction with P-gp; these predictions were then validated experimentally.

Materials and Methods
Structure-activity analyses. Molecular attributes used in the structure-activity analyses were calculated using Molecular Modeling Pro software (NorGwyn Montgomery Software, Inc., North Wales, PA). Known P-gp transport substrates and pesticides previously shown to inhibit P-gp (12 were selected for analysis. In addition, 15 pesticides that had previously been shown not to interact with P-gp (12 were randomly selected to provide an approximately equal number of transport substrates, inhibitory ligands, and noninteracting compounds. These compounds were analyzed for the following molecular descriptors as described (with citations) in the software manual: Molecular weight was calculated as the sum of the individual weights of the atoms comprising the molecule (values are exact) Compounds were grouped into three classes: transport substrates, inhibitory ligands, or noninteractors. For each molecular descriptor, means and standard deviations were calculated within each class, and significant differences between classes was evaluated by the Student's t-test.
Cell lines. The parental murine melanoma cell line B16/F1O was transfected with the human MDRI gene by a replication-defective amphitrophic retrovirus to generate the P-gp-overexpressing B16/hMDR1 cell line. The cells were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (Gibco BRL, Gaithersburg, MD) at 37°C in a humidified atmosphere of 5% CO2/95% air. The expression of P-gp, as well as the characterization of the multixenobiotic resistance phenotype (MXR), in these cells has been described previously (12,14). The MXR phenotype of the B16/hMDR1 cells was maintained by intermittently culturing the cells in 6 pM vinblastine (Sigma Chemical Company, St. Louis, MO).
Transport ofpesticides by P-glycoprotein. B16/F10 and B16/hMDR1 cells were plated in 35-mm petri dishes at a density of 0.8-1.0 X 106 cells/plate and were allowed to attach overnight. The next morning, the medium was aspirated off and replaced with medium containing 100 pM of the pesticide of interest. Plates were incubated at 370C for 4 hr after which the medium was removed and the cells were washed twice with ice-cold phosphate-buffered saline (PBS) and were overlaid with fresh medium to allow for efflux. After defined efflux periods, the medium was aspirated off and the cells were washed again with ice-cold PBS. The cells were scraped into PBS and lysed by pulse sonication, and the suspension was extracted twice with ethyl acetate. Extraction efficiencies of the six pesticides were 81-98%. The extract was evaporated under nitrogen and the residue dissolved in HPLC mobile phase. The amount of pesticide remaining in the cells was detected by HPLC on a Waters LC Module 1 Plus (Waters, Milford, MA) equipped with a reverse-phase Nova-Pak C18 column (3.9 x 150 mm, 4 pm pore size). The mobile phase for phosalone was 80% methanol/20% water at 1 ml/min with detection at 235 nm (17). The mobile phase for coumaphos was 85% acetonitrile/15% water at 2 ml/min with detection at 313 nm (18). The mobile phase for azinphos-ethyl was 80% methanol/20% water at 1 ml/min with detection at 220 nm. The mobile phase for dialifos was 67% acetonitrile/33% water at 1.4 ml/min with detection at 225 nm.
The mobile phase for fluazifop p-butyl was 71% acetonitrile/29% water acidified to pH 3.0 with phosphoric acid at 1 ml/min and detection at 270 nm (19). The mobile phase for triforine was 80% methanol/20% water at 1 ml/min with detection at 202 nm. Results from three replicates per cell line are presented for at least five time points during the efflux period. Each data point represents the percentage of pesticide remaining in the cells relative to the amount present in the cells at the beginning of the efflux period.
The efflux of two known P-gp substrates, doxorubicin and vincristine, were assessed in the two cell lines as positive controls. Doxorubicin was incubated with the cells at a concentration of 50 pM. To extract the doxorubicin associated with the cells, they were suspended in 1 ml PBS containing 100 pl of 0.1 M Tris buffer, pH 8.4, and extracted into 5 ml chloroform. The mobile phase for doxorubicin consisted of 52% water/48% acetonitrile, 10 mM SDS, 20 mM phosphoric acid at a flow rate of 1.5 ml/min with detection at 233 nm (20). Cells were incubated in 7 pM vincristine and were extracted from the cells into 5 ml chloroform. The mobile phase for vincristine detection was 60% acetonitrile/40% 25 mM phosphate buffer, pH 2.7, adjusted with phosphoric acid, containing 0.25 g/l SDS at a flow rate of 0.9 ml/min with detection at 238 nm (21).
Inhibition ofP-glycoprotein by pesticides. B16/F10 and B16/hMDR1 cells (0.2-1.0 x 106 cells/assay) were placed in microfuge tubes containing 50 pM doxorubicin (Pharmacia, Columbus, OH) in RPMI 1640 medium and either various concentrations of the pesticide or 10 pM verapamil (Sigma), the positive control. Cells were incubated at 370C for 4 hr before being pelleted by centrifugation and washed with ice-cold PBS. The pellet was resuspended and the cells washed again before extraction solution (0.6 N HCl in 50% water/50% ethanol) was added to each tube. The pellet was then vortexed and the tubes were refrigerated overnight. The amount of doxorubicin associated with the cells was measured fluorimetrically at 470 nm excitation and 585 nm emission (22). The results are presented for at least six concentrations of pesticide using three replicates per cell line. The effective concentration (EC50) of the pesticide required to increase the level of doxorubicin associated with the B16/ hMDR1 cells to half of that obtained with cells containing 10 pM verapamil (maximum inhibition) was calculated by Spearman-Karber regression analysis (23).

Results
Sture-activity analysis ofP-glycoprotein transport substrates and inhibitory ligands. Previous work in our laboratory has shown that the optimal characteristics for a pesticide to interact with P-gp are at least one cyclic structure, a molecular weight of391-490 Da, and a log K., 3.6-4.5. While those character-istics provide optimal binding to P-gp, only one of seven pesticides that show evidence of binding to P-gp was actually transported by this protein. Forty-four compounds were analyzed in the present study, including 11 transport ligands, 18 inhibitors, and 15 noninteractors (Table 1). Compounds were evaluated for various molecular attributes including size/shape parameters (molecular weight, surface area, volume, kappa shape, connectivity indices 0-3), electrostatic parameters (dipole moment, valence indices 0-3), solubility parameters (log K , solubiity, polarity, percent hydrophilic surface, hydrophiliclipophiic balance), hydrogen bonding potential, cyclic components, and others ( Table 2).
Predominant molecular characteristics that significantly differ among the com- Table 1. Compounds used in structure-activity analyses Class/compound Reference pound classes are size/shape, with molecular weight and cyclic structures being the most informative attributes; solubility, with log Ko being the most informative attribute; hydrogen bonding potential; and electrostatic parameters, with dipole moment being most informative. Based upon these analyses, criteria were developed for the differentiation of transport substrates, inhibitory ligands, and noninteracting compounds using a two tiered approach. The first tier was used to differentiate transport substrates from inhibitory ligands and noninteracting compounds. Compounds were classified as transport substrates if they met all of the following tier 1 criteria: 1) the compound had at least one .6-membered cyclic structure; 2) the molecular weight was >399; 3) the log K was <2; and 4) hydrogen bonding ow potential was >8. Remaining compounds were classified as inhibitory ligands if they met all of the following tier 2 criteria: 1) the compound had at least one .6-membered cyclic structure; 2) the molecular weight was >247; and 3) the dipole moment was .3.3.
Compounds that met neither the tier 1 nor tier 2 criteria were catagorized as noninteracting compounds. Using these criteria, transport substrates, inhibitory ligands, and noninteracting compounds were detected with 82%, 72%, and 89% accuracy, respectively.
In an attempt to validate these criteria, six previously untested pesticides were subjected to the two tier structure-activity analysis ( Table 3). None of the pesticides were predicted to be transport substrates based upon molecular weight, log K , or hydrogen bonding considerations. Coumaphos and phosalone were predicted to be P-gp inhibitors while azinphos-ethyl, dialifos, fluazifop p-butyl, and triforine were predicted to be noninteractors. These catagorizations were predicated by differences in dipole moments of the pesticides.   Articles -P-glycoprotein structure-activity relationships Transport ofpesticides by P-glycoprotein. The susceptibility of the six pesticides, in addition to the known P-gp substrates doxorubicin and vincristine, to P-gp-mediated transport then was determined experimentally. B16/F1O and B16/hMDR1 cells were incubated with each potential substrate for 4 hr; the cells were then washed and allowed to efflux the accumulated compound into clean medium. The amount of potential substrate remaining in the cells at various times during efflux was measured by HPLC. The known P-gp substrates doxorubicin and vincristine were eliminated from the P-gp-expressing B16/hMDR1 cells at a significantly greater rate than from the parental B 16/F 10 cells (Figs. 1 A,  B). In contrast, the six pesticides were eliminated from both cell lines at comparable rates (Figs. 1 C-H). Consistent with the structure-activity predictions, the six pesticides were not P-gp transport substrates.
Inhibition ofP-glycoprotein bypesticides. The six pesticides were then evaluated experimentally to identify inhibitors of P-gp. Azinphos-ethyl, coumaphos, and phosalone completely inhibited the P-gp-mediated efflux of doxorubicin as indicated by the increased accumulation of doxorubicin in the B16/hMDR-1 cells (Fig. 2). EC50 values were calculated to determine the relative potency of each of these pesticides at inhibiting the P-gp-mediated doxorubicin efflux ( Table 4). None of these pesticides were nearly as potent as verapamil (EC50 = 3.2 11M), but were within the range of EC50 values measured previously for pesticides (12. Dialifos, fluazifop p-butyl, and triforine had no effect on P-gp function at the highest concentrations tested and were therefore considered to be noninteracting compounds. Structure-activity analyses accurately predicted the interactions of five of the six pesticides (83%). Azinphos-ethyl was shown experimentally to inhibit P-gp, but was predicted to be a noninteractor. Interestingly, azinphos-ethyl was the least potent among the three inhibitory pesticides.
Predicting potency ofinhibitory ligands towards P-glycoprotein. This and previous studies (12) have shown that the potency with which some xenobiotics inhibit P-gp function varies widely. Therefore, relationships among molecular characteristics and potency of P-gp inhibitors were examined using the inhibitors chlorpyrifos, 'clotrimazole, dicapthon, hydramethylnon, parathion, and verapamil for which we had previously measured EC50 values (12. Both molecular surface area (r = 0.93) and volume (r = 0.97) were highly correlated (p = 0.01) to EC50 values (Fig. 3). As either the surface area or volume of the molecule increased, the potency of the pesticide to inhibit P-gp also increased. Exponential linear regression equations derived from these analyses were EC50 = 8060 x 10-0.098(surface area) (1) EC50 = 10937 x 10-O.013(volume) (2) From these equations, the predicted EC50 values for each of the newly identified inhibitory pesticides, azinphos-ethyl, coumaphos, and phosalone were calculated (Table 5). Both surface area and volume accurately predicted the order of potency of the pesticides and predicted the actual EC50 values within a factor of 2. Discussion P-gp is recognized primarily for its ability to efflux a variety of structurally diverse anticancer drugs from cells (25). P-gp may also be important in the cellular clearance of environmental chemicals, including some pesticides and polycyclic aromatic hydrocarbons (12,26,27). Accordingly, this phase III elimination process may prove to rival phase I and II biotransformation processes in the protection of organisms against chemical toxicity. Considering the diversity of compounds to which organisms are exposed via food, water, and environment, an understanding of the structural attributes that render a compound susceptible to elimination by P-gp would greatly facilitate the characterization of the role of P-gp in protecting against toxicant insult. Results from the present analyses indicate that size, shape, solubility, and hydrogenbonding characteristics are all important determinants of chemical transport by P-gp.   (12). Molecular surface area and volume of the pesticides are presented in Table 2.
Volume 105, Number 8, August 1997 * Environmental Health Perspectives 40 I 1 Articles -P-gIycoprotein structure-activity relationships Using the criteria defined in the tier 1 identification of P-gp transport substrates, the steroids cortisol and aldosterone meet the criteria for transport with respect to log K., and hydrogen-bonding potential, although both compounds have molecular weights that are slightly smaller than the 399 molecular weight requirement. Progesterone, on the other hand, is significantly smaller than the other steroids and does not meet the lipophilicity and hydrogen-bonding requirements for transport, and would be predicted to be an inhibitor of P-gp according to the tier 2 requirements. Cortisol and aldosterone have been shown experimentally to be transported by P-gp, while progesterone is not a transport substrate (6). Rather, progesterone has been shown experimentally to be a potent inhibitor of P-gp function (28). These observations further support the validity of the derived structure-activity relationships and suggest that additional analyses may reveal that compounds having a molecular weight of slightly less than 400 are also transport candidates. Interestingly, 400 is typically considered the approximate molecular weight limit above which chemicals are susceptible to biliary elimination (29). Perhaps P-gp and related transporters located on the canalicular membrane of hepatocytes are responsible for this size limitation. The present study demonstrates that Pgp transport substrates have higher molecular weights, lower log Ko, values, and greater hydrogen-bonding potential than compounds that are not transport substrates. Further, the hydrogen-bonding potential of transport substrates is predominantly as hydrogen donors rather than hydrogen acceptors (data not shown). All of these characteristics are conferred to a molecule as a result of hydroxylation. Hydroxylation is a major biotransformation process leading to the elimination of many xenobiotics (29). Hydroxylation historically has been considered to facilitate xenobiotic elimination by increasing the aqueous solubility of the compound and thus mobilizing it in urine or bile. Present results suggest that hydroxylation may actually target xenobiotics for active elimination by P-gp. The example of the steroid hormones provided above lends support to this hypothesis in that both cortisol and aldosterone are hydroxylated derivatives of progesterone (30). Therefore, phase I hydroxylation enzymes may function in concert with phase III transport proteins to eliminate toxic compounds from cells and from the body. Precedents for similar interactions between biotransformation enzymes and active transport proteins exist. The multidrug resistance-associated protein (MRP) and the organic anion transporter (MOAT) are both involved in the cellular efflux of compounds following glucuronic acid or glutathione conjugation (31,32).
Transport substrates of P-gp tend to be hydrophilic, based upon calculated log I §w values, while inhibitory ligands were relatively lipophilic. We noted previously that lipophilicity increased the ability of pesticides to inhibit P-gp up to a maximum potency at a log KIw of 3.6-4.5 (12).
Inhibitory potency progressively decreased as log Kow values of pesticides increased beyond this optimum range. It is possible that compounds having a log Kw of greater than 4.5 are so highly partitioned into cell membranes as to limit interaction with Pgp. It is currently not clear whether the P-gp inhibitors compete with transport substrates for the same binding site. Should these compounds interact at the same binding site, the present data suggests that increasing lipophilicity impedes transport while hydrogen bonding characteristics enhance transport potential.
Some inhibitors of P-gp had molecular weights that were somewhat lower than the minimum molecular weight of transport substrates, though potency of the inhibitors increased with increasing molecular volume or surface area. Assuming that the inhibitors competed for the same binding site as the transport substrates, the data indicates that compounds having a suboptimum molecular weight have the potential to bind to the transport site; binding affinity increases with increasing molecular mass, as indicated by molecular surface area or volume, but transport occurs only when a minimum molecular weight requirement is met.
The susceptibility of P-gp to inhibition by a variety of structurally diverse compounds implicates this protein as a possible target site of toxicity associated with many environmental chemicals. For example, a murine P-gp (mdr2 gene product) has been implicated in the biliary elimination of phospholipids (33). Drugs including amitryptyline, chlorpromazine, promethazine, and propranolol have been shown to both inhibit P-gp function (11) and cause excess accumulation of phospholipids in tissues (34). Thus, inhibition of P-gp by some xenobiotics may result in phospholipid storage disorders. Disruption of another murine P-gp (mdrla gene product) compromises the blood-brain barrier, resulting in increased accumulation of some xenobiotics in the brain (7). Accordingly, P-gp inhibitors may potentiate the toxicity of central nervous system toxicants. Continuous exposure to some xenobiotics results in decreased serum levels of corticosteroids (35). The inhibition of P-gp in the adrenal gland by the xenobiotics could be responsible for the reduced secretion of corticosteroid by this organ. Clearly, much is to be learned of the role of P-gp in the elimination of xenobiotics and as a target site for xenobiotic toxicity. An understanding of the characteristics that render xenobiotics as transport substrates or inhibitory ligands of P-gp will provide a foundation upon which additional experimental evidence can be generated to elucidate the role of this protein in chemical toxicity and elimination.