Abstract
Multidrug and toxin extruder (MATE) 1 plays a central role in mediating renal secretion of organic cations, a structurally diverse collection of compounds that includes ∼40% of prescribed drugs. Because inhibition of transport activity of other multidrug transporters, including the organic cation transporter (OCT) 2, is influenced by the structure of the transported substrate, the present study screened over 400 drugs as inhibitors of the MATE1-mediated transport of four structurally distinct organic cation substrates: the commonly used drugs: 1) metformin and 2) cimetidine; and two prototypic cationic substrates, 3) 1-methyl-4-phenylpyridinium (MPP), and 4) the novel fluorescent probe, N,N,N-trimethyl-2-[methyl(7-nitrobenzo[c][1,2,5]oxadiazol-4-yl)amino]ethanaminium iodide. Transport was measured in Chinese hamster ovary cells that stably expressed the human ortholog of MATE1. Comparison of the resulting inhibition profiles revealed no systematic influence of substrate structure on inhibitory efficacy. Similarly, IC50 values for 26 structurally diverse compounds revealed no significant influence of substrate structure on the kinetic interaction of inhibitor with MATE1. The IC50 data were used to generate three-dimensional quantitative pharmacophores that identified hydrophobic regions, H-bond acceptor sites, and an ionizable (cationic) feature as key determinants for ligand binding to MATE1. In summary, in contrast to the behavior observed with some other multidrug transporters, including OCT2, the results suggest that substrate identity exerts comparatively little influence on ligand interaction with MATE1.
Introduction
The kidney, particularly the proximal tubule, plays the principal role in clearing organic cations (OCs) from the body, i.e., molecules that carry a net positive charge at physiologic pH (Hagenbuch, 2010). These OCs include approximately 40% of all prescribed and over-the-counter drugs (including cimetidine, procainamide, pindolol, and metformin) (Neuhoff et al., 2003; Ahlin et al., 2008). Thus, renal OC secretion is a critical element in the chain of processes defining the pharmacokinetics of almost half of the drugs to which people are exposed.
The secretion of OCs by the kidney is the consequence of two sequential transport processes in the renal proximal tubule. The first of these is entry of the OC from the blood into a renal proximal tubule cell across the basolateral membrane by a process that involves electrogenic-facilitated diffusion. In humans the basolateral element of OC secretion is dominated by organic cation transporter (OCT) 2 (Motohashi et al., 2002, 2013). The second step in this process involves exit of the OC into the tubular filtrate across the apical or luminal membrane of renal proximal tubule cells by a process that uses electroneutral OC/H+ exchange. In humans the luminal step is dominated by the multidrug and toxin extruders (MATEs), MATE1 and MATE2/2-K (Motohashi et al., 2013). The presence within the kidney of this common pathway for the secretion of OCs sets the stage for unwanted drug-drug interactions (DDIs) (Lepist and Ray, 2012). The clinical cost of DDIs is substantial and responsible for approximately 1% of hospital admissions (almost 5% in elderly populations) [Becker et al., 2007; U.S. Food and Drug Administration, Preventable adverse drug reactions: A focus on drug interactions (http://www.fda.gov/Drugs/DevelopmentApprovalProcess/DevelopmentResources/DrugInteractionsLabeling/ucm110632.htm)]; therefore, the ability to predict potential DDIs could lead to decreased morbidity and increased monetary savings.
MATE-mediated OC efflux is both the active and rate-limiting element of the secretory process (Schäli et al., 1983; Pelis and Wright, 2011) and has been implicated in several clinically relevant DDIs (Ito et al., 2012; Lepist and Ray, 2012). To date, a primary focus of studies of MATE function has been establishing the interaction of MATE transporters (typically, MATE1) with specific structural classes of drugs (e.g., Yonezawa et al., 2006; Nies et al., 2012; Lee et al., 2014). The increasing attention given to the clinical impact of unwanted DDIs, and the growing acceptance of the critical role played by MATE1 in renal OC secretion, has led to the development of several predictive models of ligand interaction with human MATE1 (hMATE1) (Astorga et al., 2012; Wittwer et al., 2013; Xu et al., 2015), each based on assessing profiles of ligand inhibition of MATE1 transport activity. However, little attention has been given to a critical issue relevant to understanding the influence of MATE1 on unwanted DDI: the potential impact of substrate identity on the profile of drug interaction with MATE1. Increasing evidence suggests that the effectiveness of cationic drugs as inhibitors of multidrug transporters can be significantly influenced by the substrate used to monitor transport activity (Belzer et al., 2013; Thévenod et al., 2013; Hacker et al., 2015), which may complicate the interpretation of decision tree–based assays for assessing potential DDIs (Giacomini et al., 2010; Hillgren et al., 2013). However, the extent to which MATE transporters display such behavior is not clear.
In the current study, we screened over 400 drugs as inhibitors of the MATE1-mediated transport of four structurally distinct OC substrates: the commonly used drugs 1) metformin and 2) cimetidine; and two prototypic cationic substrates, 3) 1-methyl-4-phenylpyridinium (MPP), and 4) the novel fluorescent probe, N,N,N-trimethyl-2-[methyl(7-nitrobenzo[c][1,2,5]oxadiazol-4-yl)amino]ethanaminium iodide (NBD-MTMA). With the information gained from these screens, plus IC50 values determined for a structurally diverse subset of these compounds, we generated machine-learning and pharmacophore models, respectively. In contrast to the behavior observed with some other multidrug transporters (Ekins et al., 2002b; Garrigues et al., 2002; Westholm et al., 2009; Roth et al., 2011; Belzer et al., 2013; Hacker et al., 2015), the results suggest that substrate identity exerts comparatively little influence on ligand interaction with MATE1.
Materials and Methods
Chemicals.
[3H]MPP [specific activity (S.A.) 80 Ci/mmol] and [3H]NBD-MTMA [S.A. 85 Ci/mmol] were synthesized by the Department of Chemistry and Biochemistry, University of Arizona. [3H]Cimetidine [S.A. 80 Ci/mmol] was purchased from American Radiochemicals (St. Louis, MO), and [14C]metformin [S.A. 107 mCi/mmol] was purchased from Moravek Biochemicals (Brea, CA). Unlabeled cimetidine and metformin were purchased from Sigma-Aldrich Co. (St. Louis, MO) and AK Scientific, Inc. (Union City, CA), respectively. Unlabeled NBD-MTMA was prepared by the Synthesis Core of the Southwest Environmental Health Sciences Center, Department of Chemistry and Biochemistry, University of Arizona (Aavula et al., 2006). MPP, Ham’s F12 Kaighn’s modified medium, and Dulbecco’s modified Eagle medium were obtained from Sigma-Aldrich Co. The National Institutes of Health Clinical Collection (NCC) was acquired from Evotec (San Francisco, CA). Other reagents were of analytical grade and were commercially obtained.
Cell Culture and Stable Expression of Transporters.
Chinese hamster ovary (CHO) cells containing a single integrated flippase recombination target site were obtained from Invitrogen (Carlsbad, CA) and were used for stable expression of hMATE1 as previously described (Zhang et al., 2012). Briefly, cells were seeded in a T-75 flask following transfection via electroporation and maintained under selection pressure with hygromycin B (100 μg/ml; Invitrogen). Cells were cultured under 5% CO2-95% air in a humidified incubator (Nuaire; Plymouth, MN) at 37°C. After 2 weeks of selection the cells were used for transport studies. Subculture of the cells was performed every 3 to 4 days.
Uptake Experiments with Cultured Cells.
CHO cells expressing hMATE1 or wild-type control cells were plated in 96-well cell culture plates (Greiner/VWR International; Arlington Heights, IL) at densities sufficient for the cells to reach confluence within 24 hours (50,000 cells per well). For experiments of MATE1 transport activity the cells (MATE1-expressing and control cells) were typically preincubated for 20 minutes (room temperature) in buffer containing 20 mM NH4Cl (the first step in establishing an outwardly directed H+ gradient) (Roos and Boron, 1981). Plates were then placed in an automatic fluid aspirator/dispenser (Model 406, BioTek, Winooski, VT) and automatically rinsed/aspirated three times with room temperature Waymouth's Buffer (pH 7.4), and transport was initiated by aspirating this medium and replacing it with 60 µl of a NH4Cl-free medium (thereby rapidly establishing an outwardly directed H+ gradient) containing labeled substrate. Following the experimental incubation, the transport reaction was stopped by the rapid (∼2 seconds) addition (and simultaneous aspiration) of 0.75 ml cold (4°C) Waymouth's Buffer. Following aspiration of the cold stop, 200 μl of scintillation cocktail (Microscint 20; Perkin-Elmer, Waltham, MA) was added to each well and the plates were sealed (Topseal-A, Perkin-Elmer) and allowed to sit for at least 2 hours before radioactivity was assessed in a 12-channel, multiwell scintillation counter (Wallac Trilux 1450 Microbeta, Perkin-Elmer). Substrate uptake was typically normalized to nominal surface area of confluent cells. For the purpose of comparison with rates reported in studies that normalize transport to cell protein, we find the factor of 0.035 mg cell protein·cm−2 to be reasonably accurate (Schömig et al., 2006).
Drug Screening.
The first five plates (400 compounds) of the NCC were used for initial inhibition screening of hMATE1 transport activity. All drugs were diluted using a VIAFLOW electronic, 96-channel pipette (Integra Biosciences, Hudson, NH) to a final concentration of 50 µM in Waymouth's Buffer at pH 7.4 with 2% dimethylsulfoxide.
Computational Modeling.
Three-dimensional (3D) quantitative structure-activity relationship pharmacophore generation was done using Discovery Studio version 4.1 (Biovia; San Diego, CA). MATE1 IC50 values were used as the indicator of biologic activity. In this approach (Ekins et al., 2002a), 10 hypotheses were generated using hydrophobic, hydrogen bond acceptor, hydrogen bond donor, and positive and negative ionizable features, as well as the conformer algorithm based on energy screening and recursive buildup generation method (Li et al., 2007). After assessing all generated hypotheses, the hypothesis with the lowest energy cost was selected for further analysis since this model possessed features representative of all of the hypotheses and had the lowest total cost. The total energy cost of the generated pharmacophore was calculated from the deviation between the estimated activity and the observed activity, combined with the complexity of the hypothesis (i.e., the number of pharmacophore features). A null hypothesis, which presumed that there was no relationship between chemical features and biologic activity, was also calculated. Therefore, the greater the difference between the energy cost of the generated and null hypotheses, the less likely it was that the generated hypothesis would reflect a chance correlation. Also, the quality of the structure-activity correlation between the predicted and observed activity values was estimated via the correlation coefficient.
We also generated and validated Laplacian-corrected naive Bayesian classifier models using Discovery Studio (Biovia). Values of the A log P; molecular weight; number of rotatable bonds, rings, aromatic rings, hydrogen bond acceptors, and hydrogen bond donors; molecular fractional polar surface area; and molecular function class fingerprints of maximum diameter 6 (FCFP6) were used as the molecular descriptors. Compounds that reduced transport to <10% of control were classed as actives and everything else was classed as inactive. Computational models were validated using leave-one-out cross validation, in which each sample was left out one at a time. A model was built using the remaining samples, and that model was used to predict the left-out sample. Each model was internally validated, receiver operator characteristic curve plots were generated, and the cross-validated receiver operator characteristic curve’s area under the curve was calculated. Then, 5-fold cross validation (i.e., leave out 20% of the data set, and repeat five times) was also performed. Bayesian models were also built with the FCFP6 descriptor only using Collaborative Drug Discovery (CDD, Burlingame, CA) models in the CDD vault (Clark et al., 2015; Clark and Ekins, 2015), and 3-fold cross validation was performed.
Data Analysis.
Results are presented as mean ± S.E. Unless otherwise noted, statistical analyses were performed using a two-tailed unpaired Student’s t test. In some cases data sets were compared using one- or two-way analysis of variance (with Bonferroni post tests). Curve-fitting used algorithms found in Prism 5.03 (GraphPad Software Inc.; San Diego, CA).
Results
Kinetic Characterization of MATE1-Transported Substrates.
Four compounds shown previously to be substrates of MATE1 were selected for study. Selection criteria included: 1) structures that differed substantially from one another and 2) rates of transport sufficiently large to permit accurate kinetic analyses of inhibition. The selected compounds were [3H]MPP, [3H]NBD-MTMA, [14C]metformin, and [3H]cimetidine. The first two are model substrates for OC transport research (Lazaruk and Wright, 1990; Bednarczyk et al., 2000; Aavula et al., 2006), whereas metformin and cimetidine are therapeutic agents in wide use in the United States and other countries, both of which are secreted by the renal OCT2-MATE1/2K pathway (Nies et al., 2011). Figure 1 shows the structures of these substrates with comparisons of similarity, as assessed by Tanimoto similarity coefficients (Bajusz et al., 2015) (Discovery Studio), emphasizing their structural diversity.
Figure 2 shows time courses of MATE1-mediated uptake of the four test substrates, each corrected for uptake into wild-type CHO cells. Under the condition of the outwardly directed H+ gradient used in these experiments, uptake of all four substrates was nearly linear for almost 60 seconds, and a 30-second time point was used to provide an estimate of the initial rate of transport for all substrates in the subsequent experiments. Figure 3 shows the kinetics of MATE1-mediated transport of the four test substrates. The transport of each was adequately described by the Michaelis-Menten equation for competitive interaction of labeled and unlabeled substrates as described previously (Malo and Berteloot, 1991):(1)where J* is the rate of transport of the radiolabeled substrate from a concentration of the labeled substrate equal to [S*]; Jmax is the maximal rate of mediated substrate transport; Ktapp is the apparent Michaelis constant of the transported substrate; and [S] is the concentration of unlabeled substrate (note: uptakes at each substrate concentration were corrected for the nonsaturable component of labeled substrate accumulation that reflected the combined influence of diffusion, nonspecific binding, and incomplete rinsing of labeled substrate from the cell culture well). The different substrates exhibited a wide range of kinetic values. The transporter had the highest apparent affinity but lowest transport capacity for cimetidine (Ktapp of 2.2 µM and Jmax of 4.9 pmol/cm2·min−1) and the lowest apparent affinity but highest capacity for metformin (Ktapp of 336 µM and Jmax of 344 pmol/cm−2·min−1). The kinetic parameters for MPP and NBD-MTMA transport were distributed between these extremes (see Table 1). Transport efficiency (the ratio of Jmax to Ktapp) provides a comparative measure of carrier-mediated permeability (Schömig et al., 2006) varied by a factor of 5, with MPP transport being most efficient and NBD-MTMA transport being least efficient (Table 1).
Screening of Inhibition of MATE1-Mediated Transport.
Figure 4 shows the inhibitory influence of each of the four test substrates on transport of the other three. As expected, increasing concentrations of each compound resulted in increasing inhibition of transport activity. This inhibition was described by the following relationship:(2)where J* is the rate of MATE1-mediated transport of labeled substrate from a concentration of substrate equal to [S*] (which was selected to be much less than the Ktapp value for transport of that substrate); IC50 is the concentration of inhibitor that reduces mediated (i.e., blockable) substrate transport by 50%; and Japp is a constant that includes the maximal rate of substrate transport times the ratio of the inhibitor IC50 and the Ktapp values for transport of the labeled substrate (Groves et al., 1994) (note: uptakes at each inhibitor concentration were corrected for uptake measured in wild-type CHO cells). If the four test substrates compete with one another for a common binding site, then one may expect that each will have a single IC50 value that is equal to its Ktapp value for transport (Segel, 1975). This proved to be the case; for each compound there was no significant difference between its Ktapp value and the IC50 values it produced against transport of the other test molecules (Fig. 4; Table 1).
To assess the influence of inhibitor structure on inhibitory effectiveness we used the NCC (http://www.nihclinicalcollection.com/). Our examination began with a low-resolution screen of inhibition of MATE1-mediated transport of the four test substrates produced by a single concentration (50 µM) of each of 400 compounds from the NCC (Supplemental Material). These compounds included a broad array of physicochemical characteristics, including a structurally diverse set of OCs, organic anions, and neutral compounds, i.e., compounds that carried net positive, negative, or zero charge at physiologic pH. Figure 5 shows the profile of inhibition of all the test drugs against MATE1-mediated transport of MPP, NBD-MTMA, cimetidine, and metformin (see also Supplemental Material). The order of test agents is the same for each substrate and reflects the order of (top to bottom) increasing inhibition of MPP transport. For the purpose of comparison, compounds were considered to be comparatively effective inhibitors if the 50 µM test concentration reduced MATE1-mediated transport by 50% or more. By this criterion about 30% of the test compounds were effective inhibitors (MPP, 34.3%; NBD-MTMA, 32.5%, cimetidine, 25.3%; metformin, 36.3%). Moreover, as shown in the inhibitory profiles presented in Fig. 5, the overall profile of inhibition was similar for the four test substrates, although the rank order of effectiveness differed somewhat between the four. The top 30 most effective inhibitors of transport of each substrate included 14 compounds in common (alosetron, amisulpride, azasetron, donepezil, 6-([2-(1h-imidazol-4-yl)ethyl]amino)-n-[4-(trifluoromethyl)phenyl]heptanamide (2z)-2-butenedioate (1:1), lofexidine, midazolam, ormetoprim, perospirone, risperidone, rosiglitazone, topotecan, tropisetron, and ondansetron). The overall similarity of inhibitory effectiveness displayed by the NCC compounds is evident in the series of pairwise comparisons shown in Fig. 6, in which the percent inhibition by each test agent is compared for each pair of substrates, e.g., inhibition of MATE1-mediated MPP transport versus inhibition of NBD-MTMA transport (upper left-hand panel of Fig. 6). For each paired comparison a simple regression of the data is shown (in red), as well as the line of identity (blue) that depicts equal inhibition of transport of both substrates by all compounds. The similarity of inhibition profiles between the four substrates is evident. Furthermore, Bland-Altman analysis provided no support for the presence of significant systematic differences (fixed bias) in inhibitory profiles between any of the substrate pairs (Supplemental Fig. 1).
Inhibitory Profiles of Selected Compounds.
To obtain a more precise understanding of the structural characteristics associated with inhibition of MATE1-mediated transport of the four test substrates, a subset of the NCC collection (22 compounds) was selected to determine the IC50 values. Principal component analysis was used to compare the molecular descriptor space (A log P; molecular weight; number of hydrogen bond donors, hydrogen bond acceptors, rotatable bonds, rings, and aromatic rings; molecular polar surface area; and FCFP6) of 80 high affinity (effective) and 80 modest-to-low affinity (ineffective) inhibitors of MATE1 transport. Supplemental Fig. 2 shows 3D principal component analysis plots of effective and ineffective inhibitors of MPP transport (as determined from the 50 µM screen of the NCC). The yellow symbols show the distribution within the principal component analysis–defined chemical space of a subset of structurally diverse effectives and ineffectives from which 22 compounds (Supplemental Fig. 2C) reflecting a broad range of inhibitory effectiveness were selected to generate IC50 values for inhibition of each test substrate.
Figure 7 gives an example of five structurally distinct drugs that displayed a broad range of inhibitory effectiveness, with IC50 values that ranged over three orders of magnitude from ∼300 nM (famotidine) to ∼300 µM (venlaxafine), which shows the range of inhibition of MATE1-mediated transport produced by the broad array of structures used in the high-resolution screen. Substrate identity had comparatively little effect on the IC50 values for these five compounds; the IC50 values measured against the four test substrates did not vary by more than 60% from the average determined for each inhibitor.
The general agreement between IC50 values measured against transport of the four test substrates is evident in the pairwise comparisons presented in Fig. 8, which directly compares the log of the IC50 values for the test inhibitors generated against each substrate with those determined for the other substrates (Table 2). Regression analysis of these log-log relationships revealed that none of the slopes were different from 1 (P > 0.05). The average ratio of individual IC50 values for each set of comparisons did not vary by more than 30%, and of the 156 individual comparisons only two varied by more than 2-fold. These observations show that there was no systematic, i.e., consistent, tendency for the transport of any of the four test substrates to be inhibited with more or less effectiveness by the test inhibitors.
The set of substrates used in the current study did not include the fluorescent OC, 4-(4-dimethylamino)styryl)-N-methylpyridinium (ASP), which has been used as a test substrate to assess selectivity of both OCT2 (Kido et al., 2011) and MATE1 (Wittwer et al., 2013). In the study of MATE1 selectivity, Wittwer et al. (2013) screened 900+ compounds for inhibition of MATE1-mediated ASP transport and noted, as discussed subsequently in the present study, that cationic charge and hydrophobicity were positively correlated with inhibition of MATE1 activity. Eighty-six compounds in the set of ligands used in the current study were included in the Wittwer report and Supplemental Fig. 3A compares the degree of inhibition of MPP transport reported here with the inhibition of ASP transport reported in that study. There was a clear correlation between the inhibitions produced by this common set of ligands. Although it appeared that, in general, there was a greater degree of inhibition of MPP transport than of ASP transport (particularly evident for the higher affinity inhibitors distributed toward the left side in Supplemental Fig. 3A), this probably reflected the use of a 50 µM screening concentration in our study compared with a 20 µM screening concentration in the study by Wittwer et al. (2013). For five compounds, Supplemental Fig. 3B compares the IC50 values for inhibition of MPP or metformin transport that we determined to the values obtained by Wittwer et al. (2013) for inhibition of ASP transport. Within the limits of resolution provided by this small sample, there was little evidence for a systematic variation in IC50 values obtained for the two substrates.
Development of MATE1 Pharmacophores and Bayesian Machine-Learning Models.
Figure 9 shows the 3D pharmacophores developed from data on the inhibition produced by the 22 test drugs of the NCC plus the test substrates when used as inhibitors against MATE1-mediate transport of the four test substrates (total = 26 molecules). Each is shown overlaid with the structure of gabexate, which was a particularly good inhibitor of all four substrates. Given the relative independence of substrate identity on the profile of inhibition evident in Fig. 8, it was not unexpected that the four pharmacophores were generally quite similar to one another. Figure 10 shows the observed versus expected IC50 values calculated using these pharmacophores (MPP, r = 0.80; NBD-MTMA, r = 0.81; cimetidine, r = 0.81; metformin, r = 0.79). For MPP, NBD-MTMA, and cimetidine, each pharmacophore included two hydrogen bond acceptor features (green), one hydrophobic region (cyan), and an ionizable (i.e., cationic) feature (red). The pharmacophore developed for metformin (Fig. 9D) included only one hydrogen bond acceptor feature, two hydrophobic regions, and one ionizable feature; however, cluster analysis revealed little or no statistical difference between the pharmacophores, which is evident in the spatial alignment of the four pharmacophores (Fig. 9E).
Six molecules, 1-methyl-3-butylimidazolium (IC50 of 178.7 µM), N-butylpyridinium (26.5 µM), alosetron (0.1 µM), levofloxacin (51.6 µM), nifekalant (2.9 µM), and terbinafine (1209 µM), were used as a test set and IC50 data were generated for inhibition of MPP transport (predicted versus measured values are shown in Fig. 10; predictions based on all four pharmacophores are presented in Supplemental Table 1). N-butylpyridinium, alosetron, and nifekalant were consistently predicted as less potent inhibitors than the measured values revealed. The six compounds were added to the MPP set and this resulted in a model with the same features but a different arrangement (Supplemental Fig. 4).
Discussion
Decision tree–based predictions of potential DDIs with multidrug transporters are complicated when the quantitative profile of inhibition of transport by a potential perpetrator is influenced by the choice of substrate used to assess transport activity (e.g., Hacker et al., 2015). Although increasingly viewed as an issue for OCTs, P-glycoprotein, and organic anion-transporting polypeptides (Garrigues et al., 2002; Roth et al., 2011; Belzer et al., 2013; Hacker et al., 2015), the extent to which ligand interaction with MATE1 displays a similar substrate dependence is not clear. The two screens of inhibitor interaction with MATE1 reported to date focused on profiles generated against transport of single substrates, i.e., MPP (Astorga et al., 2012) or ASP (Wittwer et al., 2013). However, we did recently report that two structurally distinct ionic liquids (1-methyl-3-butylimidazolium and N-butyl-N-methylpyrrolidinium) had IC50 values for inhibition of MATE1-mediated transport of [3H]MPP that were about 4-fold lower than the values observed for inhibition of transport of [3H]triethylmonomethylammonium, consistent with the concept of substrate-dependent ligand interaction with MATE transporters (Martínez-Guerrero and Wright, 2013). However, the current results suggest that substrate identity exerts comparatively little influence on ligand interaction with MATE1.
This conclusion was based on the assessment of transport of four structurally diverse MATE1 substrates, two drugs in common clinical use (metformin and cimetidine) and two probe OCs (MPP and NBD-MTMA) (Fig. 1). When tested as inhibitors of each other’s transport, there were no significant differences between each substrate’s Ktapp value and the IC50 values they displayed against transport of the others (Figs. 3 and 4). Thus, within the limits of this restricted list of compounds, there was no evidence of substrate dependence in the interaction of these structurally distinct ligands with MATE1. This was followed by a low-resolution screen of 400 compounds from the NCC that provided a broadly based assessment of the influence of structural diversity on ligand interaction with MATE1. Although the rank order of inhibitory effectiveness varied slightly for the four test substrates (Fig. 5), no systematic differences were noted. In other words, the results of the low-resolution screen revealed no indication that transport of one of the test substrates was more efficiently reduced by exposure to inhibitory ligands than any of the other substrates (Fig. 6; Supplemental Fig. 1). Finally, substrate-to-substrate pairwise comparisons of IC50 values determined for the structurally diverse subset of the NCC also revealed no differences for the inhibitory interaction of the test compounds against transport of the test substrates (Fig. 8). These data are consistent with the four test substrates and the set of test inhibitors competing for interaction at a common binding site (or a set of mutually exclusive sites) at the external face of the transporter.
The qualifier, the external face of the transporter, is important. The present observations, indeed those from virtually all studies on MATE transport to date (Wright, 2014), focused on the kinetic characteristics of the transporter operating in an uptake mode. However, in its normal physiologic role as the second step in OC secretion, MATE1 mediates efflux of its organic substrates. The emphasis on influx largely reflects the technical challenges associated with accurate assessment of rates of efflux. Cytoplasmic substrate activity is difficult to quantify, and because cells are small the cytoplasmic substrate concentration during efflux changes very rapidly; the combination of these issues typically confounds efforts to measure the kinetics of efflux. It should be acknowledged that although there are thermodynamic constraints on the kinetic properties of influx versus efflux, they need not be symmetrical (Stein, 1986); in other words, under so-called zero-trans conditions the apparent affinity for substrate (or inhibitor) of the cytoplasmic face of MATE1 need not be the same as that of the extracellular face. Thus, whereas the rank ordering of ligand affinity may be expected to be qualitatively similar at the two faces of the membrane (e.g., both membrane faces of OCT2 display much higher affinity for tetrabutylammonium and corticosterone than for TEA and choline) (Volk et al., 2003), the few studies that have made such measurements suggest that the absolute Kt or IC50 value can differ by 10-fold (or more) (Stein, 1986).
The absence of systematic substrate dependence of ligand inhibition for MATE1 was in rather marked contrast to the evidence for such effects with OCT2. Two studies that examined the influence of substrate on inhibition of OCT2 transport included MPP and metformin as test substrates (Belzer et al., 2013; Hacker et al., 2015). In both studies the test inhibitors exerted a significantly greater inhibition of metformin transport than of MPP transport. These data were cited as being consistent with the view expressed by others (Zhang et al., 2005; Koepsell, 2011; Egenberger et al., 2012; Harper and Wright, 2013) that ligand interaction with OCT transporters may involve interaction at a binding surface that can support binding of two or more ligands at once. The observation here of inhibitor interactions with MATE1 that consistently displayed the same apparent inhibitor constants, regardless of substrate identity, suggests that substrates and inhibitory ligands typically interact at a kinetically common binding site at the external face of MATE1. Therefore, it is interesting to note that crystal structures of the prokaryotic MATE transporter, NorM, bound to three distinct ligands (ethidium, rhodamine 6G, and tetraphenylphosphonium) show these ligands occupying a common binding locus at the external face of the protein (Lu et al., 2013). These authors noted the presence of multiple acidic residues in the binding region that may enable versatile orientation and charge complementation of structurally dissimilar cationic drugs in NorM without the need to revamp the drug binding site. Given its multispecificity, it is intriguing to speculate that a similar strategy may exist for hMATE1.
Common feature 3D pharmacophores for MATE1 were generated previously for inhibition of MATE1-mediated MPP transport and consisted of multiple hydrophobic, hydrogen bonding, and positive ionizable features (Astorga et al., 2012). In this study we identified these same features when we generated pharmacophores for the 26 compounds screened as inhibitors of four distinct substrates (Fig. 9) using a quantitative 3D pharmacophore approach. We had also previously used Bayesian machine learning with the MATE1 inhibitor data for 46 molecules (Astorga et al., 2012), which suggested nitrogen-containing heterocycles are positively correlated with MATE1 interaction. In the current study, we used the data for 400 compounds screened as inhibitors to generate four models as well as a consensus model; these all showed that nitrogen-containing rings were again shown as important for activity, while hydroxyl, carboxylic acids, and chlorine substitutions were unfavorable for MATE1 inhibition (Supplemental Fig. 5). The independent computational approaches using either the complete data set or a subset of 26 molecules pointed to minimal differences in the models created for each substrate probe. Our hMATE1 models are also in good agreement with those we observed previously (Astorga et al., 2012). Xu et al. (2015) recently used a combinatorial pharmacophore approach with the data from Wittwer et al. (2013) and described four unique pharmacophores for inhibitors of MATE1. However, our results suggest that one pharmacophore is likely sufficient to explain inhibitory binding to MATE1. Using pharmacophores alone to score compounds fitting to a discrete pharmacophore may not be ideal, as we have shown using a small test set of six molecules—while three were reasonably well-predicted (1-methyl-3-butylimidazolium, levofloxacin, and nifekalant), three were not (N-butylpyridinium, alosetron, and terbinafine) (Supplemental Table 1). Perhaps adding some van der Waals shape restrictions to the pharmacophores may help to limit prediction error. An additional approach that uses the full extent of the screening data generated also may be a useful addition. We recently described how Bayesian models can be generated with open source FCFP6 descriptors and a Bayesian algorithm to enable transporter models to be shared and used in mobile apps (Ekins et al., 2015), and we used the data from Wittwer et al. (2013) and our own previous study (Astorga et al., 2012) as examples. This produced Bayesian models with 5-fold receiver operator characteristic curve values of 0.65 and 0.75. When we used the consensus MATE1 data set in the current study, containing 12 actives across all 4 substrates and the remaining inactives, the 3-fold cross validation was 0.82 using the open FCFP6 descriptor only (Supplemental Fig. 5; Supplemental Tables 2 and 3). These values of the area under the curve obtained using commercial or open source modeling approaches are comparable to those obtained by Wittwer et al. (2013) and their random forest model for over 800 molecules as inhibitors of ASP. While pharmacophores can produce compelling images that help explain the 3D nature of the ligand-protein interaction, machine learning may be more useful for classifying compounds and their potential for DDI at MATE1.
In conclusion, our experimental and computational data using structurally diverse substrate probes and over 400 diverse molecules tested as potential inhibitors suggest that, unlike the situation with OCT2, the interaction of inhibitory ligands with MATE1 is not systematically influenced by the structure of the substrate used to assess transport activity. Thus, in general, our observations support the conclusion that broad screening for DDIs can use a single substrate, (arguably metformin, given its utility in both in vitro and in vivo testing) and that decision trees promoted by the International Transport Consortium and the United States Food & Drug Administration can be applied without concern for the complicating influence of substrate structure for MATE1.
Lechner et al. (2016) recently determined IC50 values for inhibition of MATE1- and MATE2-K-mediated transport of five substrates (MPP, metformin, thiamine, estrone-3-sulfate (E3S), and rhodamine 123) produced by 10 inhibitors. Consistent with the results of the present study, no pronounced substrate dependency was found in IC50 values between transport of metformin, MPP, thiamine, and E3S. Inhibition of rhodamine 123 transport by both proteins, however, consistently resulted in IC50 values >4-fold greater than those for the other substrates.
Acknowledgments
The authors kindly acknowledge Biovia for providing Discovery Studio; Collaborative Drug Discovery, Inc., for providing CDD vault and CDD models; and Dr. Alex Clark for collaborations on open source descriptors and machine learning.
Authorship Contributions
Participated in research design: Martínez-Guerrero, Ekins, Wright.
Conducted experiments: Martínez-Guerrero, Morales.
Performed data analysis: Martínez-Guerrero, Morales, Ekins, Wright.
Wrote or contributed to the writing of the manuscript: Martínez-Guerrero, Ekins, Wright.
Footnotes
- Received May 13, 2016.
- Accepted June 30, 2016.
This work was supported by the National Institutes of Health National Institute of Diabetes and Digestive and Kidney Diseases [Grant 1R01DK080801]; the National Institutes of Health National Institute of Environmental Health Sciences [Grant 5P30ES006694]; and the National Institutes of Health National Heart, Lung, and Blood Institute [Grant 5T32HL07249].
Portions of this work were a part of a dissertation that was submitted by L.J.M.-G. to the University of Arizona in accordance with academic requirements.
↵This article has supplemental material available at molpharm.aspetjournals.org.
Abbreviations
- ASP
- 4-(4-dimethylamino)styryl)-N-methylpyridinium
- CHO
- Chinese hamster ovary
- DDI
- drug-drug interaction
- FCFP6
- function class fingerprints of maximum diameter 6
- hMATE
- human multidrug and toxin extruder
- MATE
- multidrug and toxin extruder
- MPP
- 1-methyl-4-phenylpyridinium
- NBD-MTMA
- N,N,N-trimethyl-2-[methyl(7-nitrobenzo[c][1,2,5]oxadiazol-4-yl)amino]ethanaminium iodide
- NCC
- National Institutes of Health Clinical Collection
- OC
- organic cation
- OCT
- organic cation transporter
- S.A.
- specific activity
- 3D
- three-dimensional
- Copyright © 2016 by The American Society for Pharmacology and Experimental Therapeutics