Advertisement
Research ArticleArticles

Derivation of a System-Independent Ki for P-glycoprotein Mediated Digoxin Transport from System-Dependent IC50 Data

Aqsaa Chaudhry, Git Chung, Adam Lynn, Akshata Yalvigi, Colin Brown, Harma Ellens, Michael O’Connor, Caroline Lee and Joe Bentz
Drug Metabolism and Disposition March 2018, 46 (3) 279-290; DOI: https://doi.org/10.1124/dmd.117.075606

Abstract

It has been previously demonstrated that IC50 values for inhibition of digoxin transport across confluent polarized cell monolayers are system-dependent. Digoxin IC50 data from five laboratories participating in the P-glycoprotein (P-gp) IC50 Initiative, using Caco-2, MDCKII-hMDR1 or LLC-PK1-hMDR1 cells, were fitted by the structural mass action kinetic model for P-gp–mediated transport across confluent cell monolayers. We determined their efflux-active P-gp concentration [T(0)], inhibitor elementary dissociation rate constant from P-gp (krQ), digoxin basolateral uptake clearance (kB), and inhibitor binding affinity to the digoxin basolateral uptake transporter (KQB). We also fitted the IC50 data for inhibition of digoxin transport through monolayers of primary human proximal tubule cells (HPTCs). All cell systems kinetically required a basolateral uptake transporter for digoxin, which also bound to all inhibitors. The inhibitor krQ was cell system–independent, thereby allowing calculation of a system-independent Ki. The variability in efflux-active P-gp concentrations and basolateral uptake clearances in the five laboratories was about an order of magnitude. These laboratory-to-laboratory variabilities can explain more than 60% of the IC50 variability found in the principal component analysis plot in a previous study, supporting the hypothesis that the observed IC50 variability is primarily due to differences in expression levels of efflux-active P-gp and the basolateral digoxin uptake transporter. HPTCs had 10- to 100-fold lower efflux-active P-gp concentrations than the overexpressing cell lines, whereas their digoxin basolateral uptake clearances were similar. HPTC basolateral uptake of digoxin was inhibited 50% by 10 μM ouabain, suggesting involvement of OATP4C1.

Introduction

The P-glycoprotein (P-gp) IC50 Initiative was established to assess interlaboratory variability in P-gp IC50 determinations. The results of this study were published by Bentz et al. (2013) and Ellens et al. (2013). The data showed significant laboratory-to-laboratory variability in the reported IC50 values, even for laboratories using the same cell line (for example, MDCKII-hMDR1 or Caco-2) and the same inhibitor. This result raised concerns about the utility of in vitro data for predicting in vivo digoxin drug-drug interaction (DDI) risk (Lee et al., 2014). Despite this variability, decision criteria could be derived by receiver operating characteristic analysis, which predicted the risk for a digoxin DDI with a low false-negative rate of 12% (Ellens et al., 2013). These papers made recommendations on the most robust way to determine a P-gp IC50 value, recommended refined decision criteria, and proposed that the decision criteria be specific for the P-gp probe substrate digoxin. These recommendations were accepted by the International Transporter Consortium (Lee et al., 2014).

Based upon previous studies (Acharya et al., 2008; Lumen et al., 2010; Agnani et al., 2011), Bentz et al. (2013) hypothesized that the IC50 variability was due to the intrinsic variability in the expression levels of efflux-active P-gp and the digoxin basolateral uptake transporter. There are several approaches to analyzing transport across confluent cell monolayers (Zamek-Gliszczynski et al., 2013). In this work, we used the structural mass action kinetic model (Bentz and Ellens, 2014) for P-gp to analyze a selected subset of data generated by the P-gp IC50 Initiative to investigate the reasons for the IC50 variability and whether a cell system–independent Ki can be extracted from these variable IC50 data, as proposed by Lumen at al. (2013). The selection criteria were focused on choosing data of a higher quality than imposed by Bentz et al. (2013) that could provide unambiguous answers to these questions and serve as the input to in vivo DDI predictions. The data from five laboratories were selected for this analysis. We also analyzed new data for digoxin transport across primary culture human proximal tubule cells (HPTCs) using the same inhibitors (Brown et al., 2008).

The structural mass action kinetic model (Bentz and Ellens, 2014) was developed to obtain elementary kinetic parameters for P-gp–mediated transport. Using this kinetic model for P-gp, we previously demonstrated that digoxin transport across MDCKII-hMDR1 and Caco-2 cells cannot be fitted by using only P-gp and digoxin passive permeability across the basolateral membrane (Acharya et al., 2008). The passive permeability alone, independently measured in the presence of N-(4-[2-(1,2,3,4-tetrahydro-6,7-dimethoxy-2-isoquinolinyl)ethyl]-phenyl)-9,10-(GF120918) (Tran et al., 2005), does not allow enough digoxin to enter the cell to permit the experimentally measured amount of digoxin effluxed to the receiver chamber by both P-gp and passive permeability. We have proposed that this is most likely due to a basolateral uptake transporter for digoxin (Acharya et al., 2008; Agnani et al., 2011; Lumen et al., 2013), although this putative transporter remains unidentified. For the purpose of the kinetic analysis presented here and in our previous work, it is not crucial whether this contributor to basolateral uptake transport is due to a basolateral uptake transporter or to “something else.” For convenience, we will refer to this contributor as the basolateral digoxin uptake transporter, or occasionally, BT, throughout this manuscript.

Previously, we have fitted the elementary kinetic rate constants of digoxin transport by P-gp (Agnani et al., 2011; Lumen et al., 2013). In this work, we used these kinetic rate constants of digoxin, as the probe substrate, for the fitting of the IC50 curves using eight P-gp inhibitors obtained from the P-gp IC50 Initiative to generate values for 1) the efflux-active P-gp concentration [T(0)], 2) the clearance of digoxin by the basolateral uptake transporter (kB), 3) the inhibitor dissociation rate constant (krQ) from P-gp, and 4) the inhibitor binding affinity to the digoxin uptake transporter (KQB), as described previously (Lumen et al., 2013). Note that KQB depends upon both uptake transporter surface density and its binding constant to the inhibitor. Thus, for a given inhibitor, the KQB values for an inhibitor across different cell lines cannot be compared in any simple way.

The function of these kinetic parameters is illustrated in Fig. 1, where BT denotes the basolateral digoxin uptake transporter. The figure illustrates that all but one of the P-gp inhibitors used by Bentz et al. (2013) are shown here to also inhibit the digoxin uptake by BT in these cells (see Tables 4 and 6, as well as Supplemental Figs. 1–24 and Supplemental Tables 1 and 2). The data for amiodarone were too poor to analyze properly. Thus, all of the IC50 values in Bentz et al. (2013) are likely to be the product of the convolution of the inhibition of both P-gp and BT. This is a major contribution to the P-gp IC50 variability reported by Bentz et al. (2013).

Fig. 1.
Fig. 1.

Biologic mechanism of digoxin transport inhibition. The top portion (above the dashed line) shows digoxin transport in the absence of inhibitor, where the basolateral uptake transporter clearance is kB (s−1), making its fitted value a convolution of transporter surface density and the binding constant of digoxin to the transporter. Digoxin then diffuses within the plasma membrane (Tran et al., 2005), with an association rate constant to P-gp of k1 (M−1s−1), and binds to P-gp with a binding constant of KC (M−1). Digoxin is then either dissociated back into the bilayer, which is by far more frequent, or effluxed into the apical chamber with a rate constant k2 (s−1), which is rare. For digoxin, roughly 1 × 104 molecules bound to P-gp return to the apical bilayer for every one that is effluxed by P-gp into the apical chamber (Lumen et al., 2013). The bottom portion of the figure (below the dashed line) shows the case when there is also a P-gp inhibitor. If the inhibitor only binds to P-gp and not to the uptake transporter, then the IC50 is due solely to P-gp. However, we show in Supplemental Table 2 and Tables 4 and 6 that carvedilol, diltiazem, felodipine, isradipine, ketoconazole, mibefradil, nicardipine, nifedipine, nitrendipine, quinidine, ranolazine, sertraline, telmisartan, troglitazone, verapamil, and, of course, digoxin all bind to the basolateral uptake transporter, thereby inhibiting digoxin’s uptake into the cells. It remains to be shown whether probe substrates such as loperamide and vinblastine, which also kinetically require a basolateral uptake transporter, would likewise be inhibited by all of these P-gp inhibitors (Lumen et al., 2013). Thus, for the P-gp IC50 Initiative data (Bentz et al., 2013), the fitted IC50 was commonly a convolution of the inhibition of P-gp and inhibition of the basolateral uptake transporter.

The k1 for association of substrates to P-gp is essentially the same for all compounds used for model validation (Lumen et al., 2013; Meng et al., 2017a), consistent with a large open binding site on P-gp (Li et al., 2014). The inhibitor dissociation constant Ki = krQ/k1, with respect to the inhibitor concentration in the membrane, so the system independence of k1 allows Ki to be calculated for each inhibitor from krQ alone. Relative to the cytosolic concentration, Ki = (krQ/k1)*KQPC, where KQPC is the partition coefficient of inhibitor Q into a liposome mimic of the cytosolic inner lipid monolayer of these eukaryotic cells (Tran et al., 2005; Lumen et al., 2013).

The fitting of the values for T(0), kB, KQB, and krQ for the inhibitors from the five laboratories used in this work allowed us to simulate IC50 curves for “virtual” cell lines, with individual kinetic parameters defined within the ranges of these parameters. This estimates how much IC50 variability the ranges of these parameters can create. The laboratory-to-laboratory variability in transporter expression levels for just these five laboratories can explain more than 60% of the IC50 variability found in the principal component analysis (PCA) plot in Bentz et al. (2013). This supports the hypothesis that the observed IC50 variability in Bentz et al. (2013) was primarily due to laboratory-to-laboratory differences in expression levels of P-gp and the basolateral digoxin uptake transporter.

Materials and Methods

Experimental.

For the generation of IC50 data for the P-gp IC50 Initiative, the methods were reported by Bentz et al. (2013).

Materials and Methods for HPTCs.

Cell culture reagents, including high-glucose Dulbecco’s modified Eagle’s medium, Ham’s F-12 Nutrient Mixture, RPMI 1640 medium, fetal calf serum, penicillin, streptomycin, l-glutamine, trypsin (with 0.02% EDTA), and Dulbecco’s phosphate-buffered saline, were obtained from Sigma-Aldrich, UK. SingleQuot kit renal epithelial growth medium supplements and growth factors were purchased from Lonza (Basel, Switzerland). Percoll was bought from GE Healthcare (UK), type 2 collagenase from Worthington Biochemicals (Lakewood, NJ), and 10X Hanks’ balanced salt solution from Invitrogen (Carlsbad, CA). Radiolabeled substrates were sourced from Hartmann Analytics (Braunschweig, Germany) and (PerkinElmer, UK). All other chemicals were from Sigma-Aldrich, and were of the highest quality available.

HPTC Cell Culture.

Primary HPTCs used in this study were isolated from human kidney donors that were not suitable for transplant. Informed consent and ethical approval for the use of human renal tissue for primary culture and drug safety studies was approved by the ethical review board of the tissue bank supplying the tissue. These kidneys were kept on ice after removal from the body and processed within 18 hours. All cell culture work was performed in a class II vertical laminar flow hood to ensure sterility. The protocol for human HPTC isolation was adapted from Brown et al. (2008). The procedure was as follows. Under sterile conditions, macroscopically normal tissue was decapsulated, and the cortex and outer stripe of the outer medulla (if present) were dissected, cut into pieces of about 1 mm3, and digested in collagenase solution (Worthington), with a final concentration 300 U/mg tissue in RMPI 1640 medium. The suspension was shaken vigorously for 2 hours at 37°C, then passed through a 120-µm sieve. The resulting cell suspension was loaded on top of a discontinuous Percoll (Pharmacia) gradient made up in RMPI 1640 medium with densities of 1.04 and 1.07 g/ml. After centrifugation at 3000 rpm for 25 minutes in a 4× 200-ml swing-out rotor, cells from the intersection were carefully aspirated, washed, and brought into culture as a mixed population of proximal tubular and distal tubular and cortical collecting duct cells seeded directly onto 6.5-mm 0.4-μm pore size polycarbonate Transwell filter supports (Costar) at a density of 50,000 cells/filter.

HPTC Transport Measurements.

Transepithelial flux measurements of digoxin across monolayers of human tubular epithelial cells were measured essentially as previously described (Brown et al., 2008). Cell monolayers grown on permeable filter supports were extensively washed 4× in a modified Krebs’ buffer (millimoles per liter): NaCl 140, KCl 5.4, MgSO4 1.2, KH2PO4 0.3, NaH2PO4 0.3, CaCl2 2, glucose 5, and Hepes 10 buffered to pH 7.4 at 37°C with Tris base. Filters were then placed in 12-well plastic plates, each well containing 1 ml of prewarmed Krebs’ or Krebs’ plus inhibitor with a further 0.5 ml of identical solution added to the apical chamber. Monolayers were preincubated for 1 hour at 37°C. Basolateral to apical fluxes of digoxin and mannitol were measured in paired resistance-matched monolayers. Monolayers were paired according to their transepithelial resistance; additionally, monolayers were excluded if the transepithelial resistance of the monolayer corrected for the resistance of the filter was less than 60 Ω.cm2. Flux was initiated by adding [3H]digoxin (1 µCi/ml) and [14C]mannitol (0.1 µCi/ml) to the basolateral chamber. A 250-μl sample was removed from the apical chamber after a 60-minute flux period. [3H] or [14C] activity in the samples was determined by liquid scintillation spectrophotometry using a Beckman liquid scintillation counter. At the end of the flux period, the remaining solutions were aspirated off and the filters were washed 4× in a 500-ml volume of ice-cold Krebs’ buffer at pH 7.4 to remove extracellular isotope. The cell monolayers were then excised from the filter insert, and the cell-associated isotope was determined by liquid scintillation counting.

Kinetic Fitting for All Cells.

Unless specifically noted, all calculations, including statistics, were performed using a 64-bit installation of MATLAB version 7.11 (release 2010b; MathWorks, Natick, MA). Logistic regressions (logistic fits), parameter, and S.E. estimates were fitted using nonlinear least-squares regression from MATLAB’s statistics toolbox. S.E.s of log(IC50) estimates were calculated as recommended by Lyles et al. (2008). Linear least-squares regressions were performed using MATLAB (Quinn and Keough 2002; Press et al., 2007). Analysis of variance and analysis of covariance were calculated via general linear models (Rao 1998; Quinn and Keough 2002). The transport kinetics fittings used a MATLAB program published by Agnani et al. (2011), and the program is freely available.

Calculation of Ki.

The binding constant of the inhibitor Q in the plasma membrane to P-gp is defined as Ki = k1Q/krQ, where k1Q is the association rate constant of the inhibitor Q from the membrane to P-gp, and krQ is the dissociation rate constant of the inhibitor from P-gp back into the membrane. Based upon the system independence of k1, we set k1Q = k1 (Lumen et al., 2013). Ki = KQPCk1/krQ is the system-independent dissociation constant of the inhibitor relative to the aqueous concentration of the inhibitor in the cytosol. We used a partition coefficient of KQPC = 350 for all inhibitors used in this work, which is the value we measured for quinidine binding to 0.1 μm of liposomes composed of a phosphatidylethanolamine/phosphatidylserine/cholesterol (1:1:1) mole ratio (Lumen et al., 2013). This lipid composition roughly mimics the cytosolic face of the plasma membrane (van Meer et al., 2008). Verapamil had a measured partition coefficient of 650, which would give a Ki roughly half as large (Lumen et al., 2013). None of the other inhibitors have known partition coefficients measured using this system. k1Q has been measured for MDCKII-hMDR1-NKI cells for several P-gp substrates, including quinidine and verapamil, and was found to be well fitted as 1 e8 M−1s−1 (Agnani et al., 2011; Lumen et al., 2013). The same value has been assumed for the LLC-PK1-hMDR1-NKI cells. However, for Caco-2 cells, Meng et al. (2017a) found that k1Q was about 1.7-fold larger. This means that the Ki for an inhibitor with the Caco-2 cells would be 1.7-fold smaller than with MDCKII-hMDR1-NKI cells. Since mammalian plasma membranes appear to be similar with respect to lipid acyl chain composition (van Meer et al., 2008), the elementary rate constants of P-gp should not depend strongly on which plasma membrane it inhabits, which appears to be the case.

Unstirred Water Layer.

There has been renewed interest in the unstirred water layer (UWL) enveloping the plasma membrane as a kinetic factor in total substrate permeation (Ghosh et al.,2014; Shibayama et al., 2015). The UWL is important when total transport is measured. Our kinetic model avoids this complication because the UWL contribution to transport is isolated to the passive permeability component, measured in the presence of GF120918 (Tran et al., 2005). This assumes that 2 μM GF120918 does not significantly affect the UWL, which is reasonable. This allows us to measure specifically the kinetics of P-gp and BT transport and inhibition.

Results

Acharya et al. (2008) and Lumen et al. (2013) previously showed that a basolateral digoxin uptake transporter was required in two of the three overexpressing cell lines analyzed here. Therefore, this uptake transporter is incorporated in the kinetic model used here (Fig. 1). If a data set did not kinetically require a basolateral digoxin uptake transporter, the fitted value of kB would be zero.

The parameters required to fit the IC50 curves for inhibition of probe-substrate transport in this model are as follows:

  1. T(0) [M, moles of P-gp per liter of acyl chains in the bilayer (Tran et al. 2005)] is the initial efflux-active concentration of P-gp prior to drug binding, which depends strongly on the microvilli morphology (Meng et al., 2017b). We fit the concentration of P-gp in the apical membrane needed to efflux the digoxin/probe substrate concentration into the apical chamber over time, which is the efflux-active concentration of P-gp prior to drug binding. Efflux from the rest of the P-gp in the apical membrane is reabsorbed back into the same or adjacent microvilli prior to reaching the apical chamber in a futile cycle.

  2. kB (s−1) is the digoxin clearance rate constant across the basolateral membrane due to the BT. The identity of BT is currently not known, nor is its surface density.

  3. krQ (s−1) is the elementary dissociation rate constant of the inhibitor (Q) from P-gp back into the apical membrane. A smaller value of krQ corresponds to a stronger binding of the inhibitor to P-gp.

  4. KQB (M−1) is the binding affinity of the inhibitor to the BT from the basolateral compartment. KQB is a convolution of the surface density of BT and the binding constant for the inhibitor to BT. The term affinity, rather than binding constant, is used since the identity and the value of the surface density of BT is unknown.

These four kinetic parameters are necessary and sufficient to fit all of the IC50 curves analyzed in this work, as shown later. We can calculate the inhibitor dissociation constant, Ki, from P-gp into the membrane from the inhibitor krQ and k1 (see Materials and Methods). We have found previously that k1 is essentially the same for all of the drugs we have studied in both MDCKII-hMDR1-NKI and Caco-2 cells (Agnani et al., 2011; Lumen et al., 2013; Meng et al., 2017a). %CV is the coefficient of variation as a percentage between the data and the fitted data points. It gives a quantitative rank order for the quality of the fit to the data.

Criteria for Choosing Data Sets to Fit.

In Bentz et al. (2013) and Ellens et al. (2013), the sole qualifying criterion for a data set was that the t-statistic tαβ > 3, which yields a 95% confidence that the measured IC50 is within 4-fold of the true IC50 (O’Connor et al., 2015). In this work, we have replaced tαβ with tβ defined by O’Conner et al., (2015), which is simpler to calculate and a nearly perfect approximation for tαβ.

The choice of data sets from Bentz et al. (2013) used here for fitting the previously described four parameters from IC50 curves in this work was based on more rigorous criteria derived from preliminary fittings focused on obtaining unambiguous data fits. At least four inhibitors out of the eight tested by the laboratory must satisfy the following two criteria:

  1. tβ > 5, which implies a 95% confidence that fitted IC50 is within 3-fold of the true IC50 (O’Connor et al., 2015).

  2. Laboratory average negative control (NC, no inhibitor) and positive control (PC, maximum inhibition) must both have CV < 20%. This is crucial because two of the four essential kinetic parameters, i.e., T(0) and kB, are specific to the cells and cannot depend on the inhibitor. Thus, two crucial parts of the IC50 curve depend on the laboratory average value NC and PC for the cells grown in that laboratory. Variations of NC and/or PC across inhibitors, which was more common for the NC data in Bentz et al. (2013), suggest that the cells in that laboratory varied in transporter expression levels from experiment to experiment, so the IC50 fits would vary. Note that the value of NC will depend on the initial concentration of digoxin used, which varied widely across laboratories, ∼100-fold [see Supplemental Table 1 for examples or Bentz et al. (2013).

Table 1 shows those laboratories that matched the quality criteria in terms of the laboratory average NC and PC controls for four or more inhibitors, and only these inhibitor data were fitted in this work. The average of NC and of PC, their S.D.s, and coefficients of variation are shown. It was important to have at least one LLC-PK1-hMDR1 laboratory in this analysis, which determined the maximum NC %CV allowed. The inhibitors left out for a laboratory did not show any consistent estimates for kinetic parameters.

View this table:
TABLE 1 

Average values for NC and PC values in the different overexpressing cell systems for the qualified inhibitors with %CV < 20% for both NC and PC, and tβ > 5

Table 2 shows the fitted values for the parameters that characterized these IC50 data. The IC50, tβ, laboratory average, tβ and β, and the IC50 fitted slope factor, which will be discussed later, are shown for each chosen laboratory and the chosen inhibitors for that laboratory.

View this table:
TABLE 2 

P-gp–overexpressing cell IC50 values, tβ-statistic values for inhibitors, and IC50 slope factor β

Fitting Protocol.

The values for T(0) and kB for a particular cell system were fixed first by simultaneously fitting all qualified IC50 data sets from the laboratory in question, as described previously, using an exhaustive fitting approach (Agnani et al., 2011; Lumen et al., 2013). Thus, for each laboratory, there is one consensus value for T(0) and for kB, both of which are only cell-dependent, not inhibitor-dependent. The second round of fits for each laboratory used these fixed values for T(0) and kB for each laboratory, but refitted the values for krQ and KQB for each inhibitor. Since there were four to eight qualified inhibitors for each chosen laboratory and five chosen laboratories, several independent krQ values were fitted at least twice (mibefradil) and up to five times (ranolazine). The other inhibitors had three to four independent fits for krQ.

Across all laboratories, the independent krQ values for each inhibitor were relatively clustered, indicating that this parameter appeared essentially cell-independent. These individual krQ values are shown in Supplemental Table 7. Table 3 shows the average, low, and high estimates for krQ at the 95% confidence level and the statistically significant one-digit consensus value for each krQ for all overexpressing cell lines, i.e., Caco-2, MDCKII-hMDR1-NKI and LLC-PK1-hMDR1-NKI.

View this table:
TABLE 3 

Average fitted values for krQ for dissociation from P-gp

A complete refit of the data was done using the fixed values of the consensus krQ values for each inhibitor. Using this consensus krQ, rather than the independently fit values, made little difference in the goodness of fit, as measured by the coefficient of variation for the fit relative to the data. This was because the refits allowed T(0), kB, and KQB for each laboratory to adjust slightly to optimize the fit with the now fixed krQ value. As shown in more detail by Agnani et al. (2011), the global minima for these multiparameter kinetic fits lie within shallow “multidimensional valleys,” which makes mass action kinetic and evolutionary sense. These same consensus krQ values were used later to fit the IC50 data curve for the primary culture HPTCs. This provides strong support for the hypothesis that the elementary rate constant krQ is essentially system-independent.

Table 4 shows the final fits of the qualified data, together with the CV of the fit and the Ki. Of course, when the true partition coefficient KQPC of the inhibitors is measured, the Ki values for each inhibitor may change somewhat from the values shown in Table 4.

View this table:
TABLE 4 

P-gp–overexpressing cells, P-gp efflux-active concentration, and kinetic parameters for the qualified inhibitors

Bentz et al. (2013) also obtained data from seven other P-gp inhibitors used in the P-gp IC50 Initiative, called Group 2 in their Table 9, that were lesser quality than the eight inhibitors in Tables 3 and 4 with respect to tβ values, as explained there. Data from selected laboratories, with tβ > 3 for six of these P-gp inhibitors (felodipine, nifedipine, nitrendipine, sertraline, telmisartan, and troglitazone), were fitted using their consensus krQ values. Initial fittings of these data showed that using their consensus values for each of these inhibitors, calculated as shown in Table 3, as opposed to individual fits, made no significant difference in the quality of the fits. All fits and fitted kinetic parameters are shown in Supplemental Figs. 1–24 and Supplemental Tables 1–3. All B > A amiodarone data from Bentz et al. (2013) were too poor to fit for unknown reasons.

All Four Kinetic Parameters Are Required to Fit the IC50 Curves.

We performed fitting studies to test whether all of these fitted parameters are necessary and sufficient to fit the data. The best fits (line) for carvedilol inhibition of digoxin transport across a confluent cell monolayer of MDCKII-hMDR1-NKI cells are shown in Figs. 2 and 3. Figure 2A shows the fit to the carvedilol IC50 curve for digoxin transport with a model that only contained P-gp, and not the basolateral digoxin uptake transporter, so kB = 0 (Fig. 2). krQ was fixed at the value shown in Table 4. This model clearly does not fit the data, especially at the NC. Changing the P-gp efflux-active concentration 10-fold from 1 e−3 M (Fig. 2A) to 1 e−2M (Fig. 2B) or 1/10-fold to 1 e−4M (Fig. 2C) did not improve the fit of digoxin transport kinetics. Therefore, P-gp efflux-active concentration does not significantly affect the fit to NC.

Fig. 2.
Fig. 2.

Inhibition of digoxin transport through a confluent cell monolayer of MDCKII-hMDR1-NKI cells by carvedilol without a basolateral digoxin uptake transporter. The squares are the data points with S.D.s. The lines are the fits to the data. All figures use the same format. The inhibitor dissociation constant from P-gp, krQ (s−1), is fixed at the consensus value found for all cells used (Table 4). (A) [T(0) = 1 e−3 M, kB = 0 second−1, krQ = 1 e4 (s−1), and KQB = 0 M−1] The best fit without the basolateral uptake transporter, which is very poor. The NC, with little or no inhibitor, cannot be reached. Can changing the efflux-active P-gp concentration change this fit? (B) The fit with a higher P-gp efflux-active concentration of T(0) = 1 e−2 M, which is the value for closely packed P-gp, i.e., the maximum possible (Tran et al., 2005; Agnani et al., 2011); the other kinetic parameters are as in (A). The best fit is poor. (C) A lower P-gp efflux surface active density of T(0) = 1 e−4 M, which remains poor. Clearly, altering the P-gp concentration, by itself, cannot significantly alter the fit to reach the NC.

Fig. 3.
Fig. 3.

Transport of digoxin through a confluent cell monolayer of MDCKII-hMDR1-NKI cells with a basolateral uptake transporter, without and with inhibitor binding to the uptake transporter. The same data as in Fig. 2 are fitted for the basolateral uptake clearance, kB(s−1). (A) [T(0) = 1 e−3 M, kB = 30 second−1, krQ = 1 e4 (s−1), and KQB = 0 M−1] A good fit to NC, but a poor fit to the PC plateau. The PC fit without binding of the inhibitor to the uptake transporter is about 15% too high. (B) [KQB = 1 e5 M−1 for inhibitor binding to the BT and all other parameters the same as in (A)] A good fit to PC. KQB is called an affinity constant since it is a convolution of the uptake transporter surface density and the inhibitor’s binding constant to the uptake transporter. Although the PC correction may not look large, that smallness is largely due to these cells overexpressing P-gp (see Fig. 5 for the HPTCs).

When the basolateral digoxin uptake transporter was added to the model, the predicted curve for carvedilol inhibition of digoxin transport kinetics fit much better. Figure 3A shows that the NC, where inhibitor concentration goes to zero, can now be fitted. That is the primary contribution of kB to these fits, i.e., allowing enough digoxin uptake into the cells for the measured amount of P-gp–mediated efflux to occur in the absence of inhibitor. At high inhibitor concentrations, the complete inhibition of P-gp efflux is inadequate to fit the full extent of inhibition of digoxin reaching the apical chamber. BT allows digoxin into the cell faster than would passive permeability. Allowing the inhibitor to bind to the BT with the affinity KQB allows the inhibitor to inhibit both P-gp and BT. This gives a good fit at the PC, as shown in Fig. 3B.

Carvedilol required all four kinetic parameters to fit the digoxin inhibition IC50 curve. The same was true for all other inhibitors. All final fits are shown in Supplemental Figs. 25–53 and Supplemental Table 4.

PCA Analysis of IC50 Variability.

We examined the variability of T(0), kB, and KQB measured here as potential causes for the variability in IC50 values across the chosen five laboratories from the P-gp IC50 Initiative and the HPTCs. krQ was fixed at the consensus value for each inhibitor (Table 4) and does not contribute significant variability in this calculation. Figure 4 shows the PCA plot for IC50 data simulated for virtual cell lines based on the ranges of kinetic parameters [T(0), kB, krQ, and KQB] shown in Tables 4 and 6. PCA axis 1 is essentially the average of log10{IC50 (M)} over all qualified inhibitors within each laboratory, as was the case in Bentz et al. (2013). The amplitude of the second axis of this PCA is very small since, in the simulation, the only remaining variabilities are the fixed krQ values for each inhibitor, since no simulated experimental error was added.

Fig. 4.
Fig. 4.

PCA plot for the variability of the IC50 values with simulated data for “virtual” cells using the kinetic parameters from the five laboratories from the P-gp IC50 Initiative (Table 4) and the HPTC data (Table 6). T(0), the efflux-active P-gp concentration, ranged from 1 e−5 to 5 e−3 M; kB, the basolateral uptake transporter clearance, ranged from 5 to 30 second−1; and KQB, the affinity constant of the inhibitor to the basolateral uptake transporter, ranged from 2 e4 to 1 e6 M−1. For each inhibitor, the consensus values of the inhibitor dissociation constant from P-gp, krQ, from Table 4 were used. These combinations of these ranges were used to simulate IC50 data curves that were fit for the IC50 values that were then used to make the PCA plot. The color of the symbols denotes the value of kB, and their shape denotes the value of KQB, as indicated on the right-hand legend. The value for T(0) was not indicated, as a third element embedded into the symbols makes the plot unintelligible. The important point for PCA axis 1 is that its range, i.e., the minimum and maximum values, is determined by the data for T(0) ≥ 1 e−4 M. The data points for the smaller T(0) values depicting the HPTCs lay between these extremes.

Fitting Transport Kinetics for Primary Cell Culture Monolayers of HPTCs.

The inhibition of digoxin transport through confluent cell monolayers composed of primary HPTCs using the same inhibitors was fitted as well, using the fixed consensus krQ values in Table 3. Figure 5 shows the best fits for ketoconazole inhibition of digoxin transport across a confluent monolayer of HPTCs. Figure 5A shows that when a BT was not incorporated, kB = 0, the fit was poor, as digoxin transport is not fitted at the NC. Figure 5B shows that when BT is added to the model without binding/inhibition by the inhibitor, KQB = 0, digoxin can be fitted at the smaller inhibitor concentrations, but not the data at the high inhibitor concentrations. Figure 5C shows that when inhibition of BT is accounted for by KQB, then the observed data are fit well by the model. KQB for ketoconazole is responsible for about a 50% inhibition of digoxin transport. The primary culture HPTCs, with smaller efflux-active P-gp concentrations, show a much greater role of inhibitor binding to BT on transport than in the P-gp–overexpressing cell lines (see Fig. 5, B and C).

Fig. 5.
Fig. 5.

Digoxin transport through a confluent cell monolayer of primary HPTCs with inhibition by ketoconazole. (A) [T(0) = 1.5 e−5 M, kB = 0 second−1, krQ = 3 e4 (s−1), and KQB = 0 M−1] Fitted with just P-gp, i.e., no basolateral uptake clearance. The fit is very poor, as in Fig. 2. (B) [kB = 45 s−1 and all other parameters the same as (A)] A good fit to NC, but a poor fit to PC, notably worse than Fig. 3A. The PC fit without binding of the inhibitor to BT is about 50% too high. (C) [KQB = 2 e6 M−1 and all other parameters the same as (B)] A good fit to PC.

The kinetic parameters of inhibition of digoxin transport conducted in HPTCs are shown in Tables 5 and 6. Table 5 shows the IC50, tβ, and the slope factor β for these data, which came from 10 out of a total of 13 kidneys evaluated. These data were not filtered through the same quality criteria used for the overexpressing cells, but instead just used tβ > 3, except for ranolazine, where tβ = 2.4 (Table 5). These ranolazine data were included because they were the only acceptable data for this inhibitor for the HPTCs. The IC50 values were typically lower than the values observed with the overexpressing cells, mostly due to the lower efflux-active P-gp concentration in these primary cells, which correlates with the IC50 (Lumen et al., 2010).

View this table:
TABLE 5 

HPTC IC50 values, tβ statistic values for inhibitors, and IC50 slope factor β

View this table:
TABLE 6 

HPTC P-gp efflux-active concentration and elementary kinetic parameters for the data in Table 5

The slope factor β was also smaller than those for the overexpressing cells. Although estimated β values for the HPTCs include a few values near or greater than 1, the mean of the estimates is approximately 0.71 (S.E. = 0.062), and the 95% confidence interval for the values (0.58–0.85) does not include 1.0. Thus, the HPTCs have a lower average β < 1 estimate than the cultured cells. A hypothesis for this behavior is given in the Discussion.

Table 6 shows this final fitting of the qualified HPTC data. These cells have roughly 10- to 100-fold less efflux-active P-gp than the overexpressing cell lines used in the P-gp IC50 Initiative (Table 4). The fraction of total transport due to the basolateral digoxin uptake transporter is greater (Fig. 5B), thus the inhibition of BT has a greater impact on the IC50. The consensus values for krQ (Table 4) worked well with these HPTCs, expanding the system independence of this elementary kinetic parameter. KQB values were similar to those shown in Table 4 for the overexpressing cells.

We note that the kinetic parameters can vary significantly between different kidney samples for the same inhibitor. For example, the IC50/Ki ratio with carvedilol varied 25-fold between two kidney preps, which was entirely due to the variability in the IC50 values (Table 5). Interestingly, this IC50 variability was due more to the variability of kB and KQB rather than the efflux-active P-gp. The same is true for the ketoconazole, quinidine, and verapamil data. Nicardipine, on the other hand, shows IC50 variability between kidney samples due to differences in efflux-active P-gp concentrations. This highlights the importance of the basolateral digoxin uptake transporter and its inhibition in any in vivo DDI predictions and the need to know which parameters are the primary drivers of the IC50. All fits are shown in Supplemental Figs. 54–67 and Supplemental Table 5.

Potential Involvement of OATP4C1 in Basolateral Uptake of Digoxin in HPTC Monolayers.

Figure 6A shows that digoxin uptake into the HPTCs was 50% inhibited with about 10 μM of the OATP4C1 inhibitor ouabain. The inhibition followed a roughly linear inhibition curve versus log[inhibitor concentration], i.e., not a logistical curve as observed for the other inhibitors studied here. Digoxin uptake into the cells is significantly greater in the presence of GF120918, where P-gp is fully inhibited, than in the control cells, where P-gp is fully active. Figure 6B shows that digoxin transport across the HPTCs, JB➔A, was only about 20%–25% inhibited with 30 μM OATP4C1 inhibitor ouabain, also with a roughly linear, not logistic, inhibition curve.

Fig. 6.
Fig. 6.

Digoxin uptake in and transport through a confluent cell monolayer of primary HPTCs inhibited by ouabain. (A) The inhibition of digoxin uptake into the HPTCs by ouabain. Ouabain (10 μM) inhibits digoxin uptake by about 50%. The decreased transport is mostly linear rather than logistic, as discussed. (B) The ouabain inhibition of digoxin transport, JB > A, through the HPTCs. Ouabain (30 μM) only inhibits digoxin B > A transport by 20%–25%. The decreased transport is mostly linear rather than logistic, as discussed.

The inhibition of digoxin uptake into and transport through the HPTCs by T3, 0–30 μM, was only somewhat reduced at and above 10 μM T3. This was similar to the uptake results by Mikkaichi et al. (2004). The data for T3 inhibition of digoxin uptake and digoxin transport across the HPTC confluent cell monolayer are shown in Supplemental Figs. 68–69 and Supplemental Table 6.

Discussion

We have used the structural mass action kinetic model for digoxin transport through a confluent monolayer of P-gp–overexpressing polarized cells to 1) derive system-independent P-gp inhibitor dissociation rate constants for calculation of system-independent Ki values and 2) explore potential mechanistic factors that contribute to the variability in IC50 values observed in the P-gp IC50 Initiative (Bentz et al., 2013). A subset of the IC50 data generated by the P-gp IC50 Initiative participants was selected for this work, based on data quality criteria described in the Results section. We used the data from two Caco-2 laboratories, two MDCKII-hMDR1-NKI laboratories, and one LLC-PK1-hMDR1-NKI laboratory. Newly generated data for digoxin transport inhibition across primary HPTC monolayers were also included in this analysis using the same inhibitors, as well as ouabain and T3.

The structural mass action kinetics model for P-gp–mediated transport has been extensively validated as a reliable diagnostic tool using several cell lines to determine the efflux-active P-gp concentrations and to identify kinetically required uptake transporters for the transport of P-gp substrates across confluent cell monolayers (Acharya et al., 2008; Agnani et al., 2011; Lumen et al., 2013; Bentz and Ellens, 2014; Meng et al., 2017a,b). The IC50 curves of digoxin transport across confluent cell monolayers studied here were analyzed using this model. The kinetic parameters needed to fit the data shown in Figs. 2, 3, and 5 were the efflux-active P-gp concentration [T(0)], uptake clearance of digoxin by the basolateral digoxin uptake transporter (kB), the dissociation constant of the inhibitor from P-gp (krQ), and the affinity of the inhibitor to the basolateral uptake transporter (KQB), as well as the relevant kinetic parameters for the probe substrate, digoxin in this case, which were obtained from Lumen et al. (2013). As shown in the Results section, these kinetic parameters were necessary and sufficient to fit IC50 data in P-gp–overexpressing cell lines and the primary culture HPTCs to within experimental error. Importantly, the krQ (and therefore, the calculated Ki) for a given inhibitor was found to be system-independent over the cell lines used here.

Our kinetic analysis found that the eight P-gp inhibitors used here in all overexpressing cell lines and the HPTCs, as well as six other P-gp inhibitors used with the overexpressing cell lines were kinetically required to bind to and inhibit the digoxin basolateral uptake transporter (see Tables 4 and 6 and Supplemental Tables 1–3). This means that any IC50 value reported by Bentz et al. (2013) could be due to inhibitor binding to P-gp, or inhibitor binding to the digoxin basolateral uptake transporter, or more likely a complex convolution of both binding events.

Bentz et al. (2013) used PCA to show that the largest variability was due essentially to the differences in the average log10{IC50 (M)} over the inhibitors between the different laboratories, which was PCA axis 1 in Fig. 7 of that paper. This result raised concerns about the utility of in vitro data for predicting in vivo digoxin DDI risk (Lee et al., 2014). Here, IC50 curves were simulated using all combinations of efflux-active P-gp and the kinetic parameters from within the ranges shown in Tables 4 and 6. Figure 4 shows the PCA analysis on these simulated curves, which had no added random error in the data. Here, PCA axis 1 was essentially the average of the log10{IC50}, the same as in Bentz et al. (2013). The range for PCA axis 1, i.e., the minimum and maximum values, is determined by the data for T(0) ≥ 1 e−4 M. The data points for the smaller T(0) values depicting the HPTC lay between these extremes. This range covers more than 60% of the range of axis 1 found in Bentz et al. (2013). Thus, most of the IC50 variability found in Bentz et al. (2013) can be explained by the cells in each laboratory, regardless of origin, expressing a range of efflux-active P-gp concentrations and expression levels of basolateral digoxin uptake transporter. The remaining variability in Bentz et al. (2013) is most likely due to a combination of experimental error, laboratories with 3 < tβ ≤ 5 with NC and PC variations in excess of %CV < 20%, and finally, the convolution of P-gp and BT inhibition within the fitting of the inhibition data to a single logistic IC50 curve.

The kinetic need for a basolateral uptake transporter to explain transport of P-gp substrates across a P-gp–expressing polarized cell monolayer is not unique to digoxin. We have found that both loperamide and vinblastine kinetically require a basolateral uptake transporter in MDCKII-hMDR1-NKI, MDCKII-hMDR1-NIH, and Caco-2 cells (Acharya et al., 2008; Lumen et al., 2013). If a P-gp substrate uses a BT, then how can that be shown? Lumen et al. (2013) showed by simulations that a basolateral uptake transporter could be observed when the passive permeability, in the presence of 2 μM GF120918, was less than about 320 nm/s. When this is so, the experimental probe substrate reaching the apical chamber is significantly greater than what can be fitted by the kinetic model without a mechanism for increasing the permeation of the probe substrate through the basolateral membrane beyond the passive permeability through the bilayer. When the passive permeability is larger than this threshold of 320 nm/s, then the model fitted deficit due to the absence of BT becomes insignificant compared with the total probe-substrate transport. This passive permeability threshold matched our findings that digoxin, loperamide, and vinblastine kinetically required the basolateral uptake transporter, whereas amprenavir, ketoconazole, quinidine, and verapamil did not (Lumen et al., 2013).

Mechanistically, it is very interesting that Lumen et al. (2013) showed for Caco-2 and MDCKII-hMDR1-NIH cells that ketoconazole and verapamil inhibited digoxin transport through the basolateral digoxin uptake transporter, as opposed to just inhibiting P-gp. The same result was found here in Tables 4 and 6. Thus, although these two P-gp inhibitors were not kinetically required to use a basolateral uptake transporter for their own transport, this does not mean that they do not bind to that transporter and, thereby, inhibit the transport of another P-gp substrate.

The identity of the basolateral digoxin uptake transporter in all these cells remains unknown. The possibility of a digoxin transporter in Caco-2 cells has been reported (Lowes et al., 2003). Taub et al. (2011) showed that digoxin is not a substrate of OATP1A2, 1B1, 1B3, and 2B1, but was a substrate of a sodium-dependent transporter endogenously expressed in HEK293 cells. The kinetic modeling of the basolateral digoxin uptake transporter in MDCKII-hMDR1-NKI cells by Agnani et al. (2011) found that digoxin’s uptake was somewhat better fitted as a bidirectional passive transporter as compared with an active importer.

In Mikkaichi et al. (2004), MDCK cells whose “endogenous expression of OATP4C1 in MDCK cell was not detected” were transfected with human OATP4C1. Confluent monolayers of these cells were used to examine digoxin uptake into and the digoxin flux across these monolayers as a function of the OATP4C1 inhibitors ouabain and T3. Their Fig. 5A showed that digoxin uptake into the cells was inhibited logistically to about 50% at about 0.4 μM ouabain. T3 showed no significant reduction of digoxin uptake at 0–30 μM (Mikkaichi et al., 2004). In our study of ouabain, digoxin uptake into HPTCs was inhibited essentially linearly, not logistically, at ouabain concentrations between 0.1 and 30 μM, with an inhibition of about 50% at 10 μM ouabain. Thus, OATP4C1 may be a basolateral digoxin uptake transporter in this system.

The fact that ouabain’s inhibition curve of basolateral uptake in the HPTCs is basically linear, rather than logistical, suggests that there could be a second basolateral digoxin uptake transporter involved in this complex system of primary isolated cells. Given that GF120918 caused the digoxin uptake into the HPTCs to be larger than found without ouabain (Fig. 6A), whereas digoxin’s JB➔A flux was smaller than that caused by ouabain (Fig. 6B), suggests that one of the hypothesized two (or more) basolateral digoxin uptake transporters in HPTCs is not sensitive to GF120918 inhibition. The basolateral digoxin uptake transporter in the overexpressing cells used here is completely inhibited by 2 μM GF120918 (Tran et al., 2005; Acharya et al., 2008).

Whether ouabain directly affects P-gp transport function has not been reported. It has been reported that ouabain had no effect on P-gp ATPase activity (Shapiro and Ling, 1994). Brouillard et al. (2001) reported that ouabain induced P-gp expression, but that is unlikely to effect the immediate inhibition of digoxin transport by P-gp in this work.

The β in the IC50 logistic or Hill equation (Hill, 1913) is called the slope factor, because it is fitted to the slope of the IC50 curve at the IC50 (O’Connor et al., 2015). Historically, β > 1 has also been considered an indicator of cooperativity in substrate binding to a protein, e.g., hemoglobin (Hill, 1913). The slope factor is used in all commercial IC50 fitting packages, so it is part of every published IC50 fit, yet it is rarely reported or discussed. In Bentz et al. (2013), the β values averaged near 1, but there was a wide range between the minimum (0.3) and maximum (3.6) in Table 7 here. For the higher-quality data sets used in this work, with roughly 4-fold fewer fitted data points, the fitted values were still centered on β = 1, but the ranges were significantly smaller, ranging from 0.6 to 1.8 in Table 7.

View this table:
TABLE 7 

Effect of data quality on the slope factor β

There have been very many papers on the cooperativity of substrate binding to P-gp, mostly based upon the ATPase activity, but there has been no clear consensus (Lumen, 2011). There are several citations to works that claim multiple substrates being bound simultaneously to P-gp (Lumen, 2011), but there is no evidence that all of these binding sites actually efflux substrate. The structural mass action kinetic model has only a single substrate binding site per P-gp, explained in detail by Tran et al. (2005), so this kinetic model cannot support cooperativity. To date, no kinetic need for efflux cooperativity has been observed using this kinetic model for P-gp efflux kinetics. Together, these data suggest that the fitted β values have little, if anything, to do with the number of cooperative binding sites in P-gp. The slope factor β does provide an additional degree of freedom for fitting IC50 data, without any mechanistic control for β values greater than 1.

On the other hand, for the HPTCs, the average of the slope factor β is significantly less than 1, with a mean of the estimates of < β >= 0.71 and a 95% confidence interval of 0.58–0.85 (Table 6). The simplest hypothesis that explains this difference from the overexpressing cells is that the HPTCs express two different basolateral digoxin uptake transporters. The interaction of their respective binding sites would reduce the slope of the inhibition curve at the IC50, i.e., yielding β < 1. This speculation is also consistent with the linear, rather than logistical, inhibition curve of digoxin uptake by ouabain in HPTCs, described above (Fig. 6A).

Finally, P-gp vesicles have been suggested as a simpler and possibly superior system for determination of P-gp IC50 values. However, the vesicle IC50 data generated by the P-gp IC50 Initiative using N-methyl-quinidine as a probe substrate were also very variable from laboratory to laboratory as the confluent cell monolayer data (Bentz et al., 2013). Currently, these vesicles are typically derived from mammalian or insect cell membranes, which contain a variety of endogenous transporters, just like the polarized cell systems, and must be assumed to vary from laboratory to laboratory in expression levels of P-gp and other transporters, just as the polarized cell monolayers.

Conclusions

We have previously demonstrated that the elementary rate constants for P-gp–mediated transport (k1, k2, and kr) obtained using our structural mass action kinetics model are essentially the same in both MDCKII-hMDR1-NKI and Caco-2 cells for amprenavir, quinidine, loperamide, and digoxin, i.e., they are essentially system-independent. Here, we have shown that krQ is also system-independent for the cells used, including with the primary cell culture HPTCs. We can therefore calculate a system-independent P-gp–specific Ki using highly variable IC50 data, even when another transporter or two are involved in the transport of the probe substrate.

Here, the criteria for selecting qualified data sets were based upon having IC50 data curves from multiple laboratories for multiple inhibitors. These criteria can be adapted to a single lead compound. It requires at least three completely independent IC50 data curves, e.g., for different cell preparations, for the inhibitor of interest. For these curves, the average of the negative and of the positive controls must each have a %CV < 20%. Each of the IC50 curves must have tβ > 5. It is best to simultaneously fit all data curves to obtain the consensus elementary rate constants, but they can be fitted separately and then averaged, as was done for krQ in Table 3. The k1, kr, k2, and kB for the probe substrate used must be known, since they are part of the IC50 data-fitting program. These are already available for amprenavir, digoxin, ketoconazole, loperamide, quinidine, verapamil, and vinblastine (Lumen et al., 2013).

For the purpose of the kinetic analysis presented here, and in our previous work, it is not crucial whether the basolateral uptake clearance of the probe substrate is due to a basolateral uptake transporter or to “something else.” However, any proposed “nontransporter” mechanism for this enhanced basolateral membrane permeability of the probe substrate must have a plausible mechanism for the inhibition of this enhanced permeability as a function of increased inhibitor concentration to account for the significant effect of KQB on the fit.

The final point is how to use the structural mass action kinetic model to obtain in vivo DDI predictions. Basically, our MATLAB program can replace the linearized Michaelis-Menten programs within existing in vivo Physiologically Based PharmacoKinetics (PBPK) computer programs. At present, this appears most easily accomplished within the MATLAB SimBiology framework, which has access to other required MATLAB programs and which can run the Particle Swarm global optimization program we use, but other options are possible. Our analysis provides the necessary elementary rate constants from in vitro experiments, all of which appear system-independent. Although there are good biophysical reasons for expecting the elementary rate constants for P-gp transport to be essentially system-independent, that is not the case for the microvilli morphology–dependent efflux-active P-gp and the as-yet unidentified basolateral and apical uptake transporter(s) kinetic parameters. The need for in vivo data fitting is seen clearly here in the HPTC data, with large kidney-to-kidney variability in IC50 values for the same inhibitor that was due mostly to variability in the basolateral digoxin uptake transporter clearance and inhibition, not P-gp activity. Therefore, using the in vitro–derived elementary rate constants, the in vivo PBPK model can fit in vivo data for the efflux-active P-gp, the basolateral and apical uptake transporter clearances, and the inhibitor affinities for these uptake transporters. These fits would allow the formulation of in vivo DDI predictions.

Authorship Contributions

Participated in research design: Brown, O’Connor, Lee, Ellens, Bentz.

Conducted experiments: Chung.

Contributed new reagents or analytic tools: O’Connor, Bentz.

Performed data analysis: Chaudhry, Chung, Lynn, Yalvigi, Brown, O’Connor, Lee, Ellens, Bentz.

Wrote or contributed to the writing of the manuscript: Chung, Brown, O’Connor, Lee, Ellens, Bentz.

Footnotes

Abbreviations

BT
basolateral digoxin uptake transporter
CV
coefficient of variation
DDI
drug-drug interaction
GF120918
N-(4-[2-(1,2,3,4-tetrahydro-6,7-dimethoxy-2-isoquinolinyl)ethyl]-phenyl)-9,10-
HPTC
human proximal tubule cell
NC
negative control
PC
positive control
PCA
principal component analysis
P-gp
P-glycoprotein
UWL
unstirred water layer

References

  1. ↵
    1. Acharya P,
    2. O’Connor MP,
    3. Polli JW,
    4. Ayrton A,
    5. Ellens H, and
    6. Bentz J
    (2008) Kinetic identification of membrane transporters that assist P-glycoprotein-mediated transport of digoxin and loperamide through a confluent monolayer of MDCKII-hMDR1 cells. Drug Metab Dispos 36:452–460.
    OpenUrlAbstract/FREE Full Text
  2. ↵
    1. Agnani D,
    2. Acharya P,
    3. Martinez E,
    4. Tran TT,
    5. Abraham F,
    6. Tobin F,
    7. Ellens H, and
    8. Bentz J
    (2011) Fitting the elementary rate constants of the P-gp transporter network in the hMDR1-MDCK confluent cell monolayer using a particle swarm algorithm. PLoS One 6:e25086.
    OpenUrlCrossRefPubMed
  3. ↵
    1. Nagar S,
    2. Argikar UA, and
    3. Tweedie DJ
    1. Bentz J and
    2. Ellens H
    (2014) A structural model for the mass action kinetic analysis of P-gp mediated transport through confluent cell monolayers, in Enzyme Kinetics in Drug Metabolism – Fundamentals and Applications (Nagar S, Argikar UA, and Tweedie DJ eds) pp 289–316, Humana Press, New York.
  4. ↵
    1. Bentz J,
    2. O’Connor MP,
    3. Bednarczyk D,
    4. Coleman J,
    5. Lee C,
    6. Palm J,
    7. Pak YA,
    8. Perloff ES,
    9. Reyner E,
    10. Balimane P, et al.
    (2013) Variability in P-glycoprotein inhibitory potency (IC50) using various in vitro experimental systems: implications for universal digoxin DDI risk assessment decision criteria. Drug Metab Dispos 41:1347–1366.
    OpenUrlAbstract/FREE Full Text
  5. ↵
    1. Brouillard F,
    2. Tondelier D,
    3. Edelman A, and
    4. Baudouin-Legros M
    (2001) Drug resistance induced by ouabain via the stimulation of MDR1 gene expression in human carcinomatous pulmonary cells. Cancer Res 61:1693–1698.
    OpenUrlAbstract/FREE Full Text
  6. ↵
    1. Brown CD,
    2. Sayer R,
    3. Windass AS,
    4. Haslam IS,
    5. De Broe ME,
    6. D’Haese PC, and
    7. Verhulst A
    (2008) Characterisation of human tubular cell monolayers as a model of proximal tubular xenobiotic handling. Toxicol Appl Pharmacol 233:428–438.
    OpenUrlCrossRefPubMed
  7. ↵
    1. Ellens H,
    2. Deng S,
    3. Coleman J,
    4. Bentz J,
    5. Taub ME,
    6. Ragueneau-Majlessi I,
    7. Chung SP,
    8. Herédi-Szabó K,
    9. Neuhoff S,
    10. Palm J, et al.
    (2013) Application of receiver operating characteristic analysis to refine the prediction of potential digoxin drug interactions. Drug Metab Dispos 41:1367–1374.
    OpenUrlAbstract/FREE Full Text
  8. ↵
    1. Ghosh A,
    2. Scott DO, and
    3. Maurer TS
    (2014) Towards a unified model of passive drug permeation I: origins of the unstirred water layer with applications to ionic permeation. Eur J Pharm Sci 52:109–124.
    OpenUrlCrossRefPubMed
  9. ↵
    1. Hill AV
    (1913) The combinations of haemoglobin with oxygen and with carbon monoxide. I. Biochem J 7:471–480.
    OpenUrlFREE Full Text
  10. ↵
    1. Lee CA,
    2. Kalvass JC,
    3. Galetin A, and
    4. Zamek-Gliszczynski MJ
    (2014) ITC commentary on the prediction of digoxin clinical drug-drug interactions from in vitro transporter assays. Clin Pharmacol Ther 96:298–301.
    OpenUrl
  11. ↵
    1. Li J,
    2. Jaimes KF, and
    3. Aller SG
    (2014). Refined structures of mouse P-glycoprotein. Protein Sci 23:34–46.
    OpenUrlCrossRefPubMed
  12. ↵
    1. Lowes S,
    2. Cavet ME, and
    3. Simmons NL
    (2003) Evidence for a non-MDR1 component in digoxin secretion by human intestinal Caco-2 epithelial layers. Eur J Pharmacol 458:49–56.
    OpenUrlCrossRefPubMed
  13. ↵
    Lumen AA (2011). Mechanistic Understanding of P-gp Mediated Transport and Inhibition Kinetics across a Confluent Monolayer of MDCKII-hMDR1 cells using a Mass Action Kinetic model. Ph.D. thesis, Drexel University, Philadelphia.
  14. ↵
    1. Lumen AA,
    2. Acharya P,
    3. Polli JW,
    4. Ayrton A,
    5. Ellens H, and
    6. Bentz J
    (2010) If the KI is defined by the free energy of binding to P-glycoprotein, which kinetic parameters define the IC50 for the Madin-Darby canine kidney II cell line overexpressing human multidrug resistance 1 confluent cell monolayer? Drug Metab Dispos 38:260–269.
    OpenUrlAbstract/FREE Full Text
  15. ↵
    1. Lumen AA,
    2. Li L,
    3. Li J,
    4. Ahmed Z,
    5. Meng Z,
    6. Owen A,
    7. Ellens H,
    8. Hidalgo IJ, and
    9. Bentz J
    (2013) Transport inhibition of digoxin using several common P-gp expressing cell lines is not necessarily reporting only on inhibitor binding to P-gp. PLoS One 8:e69394.
    OpenUrl
  16. ↵
    1. Lyles RH,
    2. Poindexter C,
    3. Evans A,
    4. Brown M, and
    5. Cooper CR
    (2008) Nonlinear model-based estimates of IC(50) for studies involving continuous therapeutic dose-response data. Contemp Clin Trials 29:878–886.
    OpenUrlCrossRefPubMed
  17. ↵
    1. Meng Z,
    2. Ellens H, and
    3. Bentz J
    (2017a) Extrapolation of elementary rate constants of P-glycoprotein-mediated transport from MDCKII-hMDR1-NKI to Caco-2 cells. Drug Metab Dispos 45:190–197.
    OpenUrlAbstract/FREE Full Text
  18. ↵
    1. Meng Z,
    2. Le Marchand S,
    3. Agnani D,
    4. Szapacs M,
    5. Ellens H, and
    6. Bentz J
    (2017b) Microvilli morphology can affect efflux active P-glycoprotein in confluent MDCKII-hMDR1-NKI and Caco-2 cell monolayers. Drug Metab Dispos 45:145–151.
    OpenUrlAbstract/FREE Full Text
  19. ↵
    1. Mikkaichi T,
    2. Suzuki T,
    3. Onogawa T,
    4. Tanemoto M,
    5. Mizutamari H,
    6. Okada M,
    7. Chaki T,
    8. Masuda S,
    9. Tokui T,
    10. Eto N, et al.
    (2004) Isolation and characterization of a digoxin transporter and its rat homologue expressed in the kidney. Proc Natl Acad Sci USA 101:3569–3574.
    OpenUrlAbstract/FREE Full Text
  20. ↵
    1. O’Connor M,
    2. Lee C,
    3. Ellens H, and
    4. Bentz J
    (2015) A novel application of t-statistics to objectively assess the quality of IC50 fits for P-glycoprotein and other transporters. Pharmacol Res Perspect 3:e00078.
    OpenUrl
  21. ↵
    1. Press WH,
    2. Teukolsky SA,
    3. Vetterling WT, and
    4. Flannery BP
    (2007) Numerical Recipes: The Art of Scientific Computing, 3rd ed, Cambridge Univ. Press, Cambridge, UK.
  22. ↵
    1. Quinn GP and
    2. Keough MJ
    (2002) Experimental Design and Data Analysis for Biologists, Cambridge Univ. Press, Cambridge, UK.
  23. ↵
    1. Rao PV
    (1998) Statistical Research Methods in the Life Sciences, Duxbury Press, Pacific Grove, CA.
  24. ↵
    1. Shapiro AB and
    2. Ling V
    (1994) ATPase activity of purified and reconstituted P-glycoprotein from Chinese hamster ovary cells. J Biol Chem 269:3745–3754.
    OpenUrlAbstract/FREE Full Text
  25. ↵
    1. Shibayama T,
    2. Morales M,
    3. Zhang X,
    4. Martínez-Guerrero LJ,
    5. Berteloot A,
    6. Secomb TW, and
    7. Wright SH
    (2015) Unstirred water layers and the kinetics of organic cation transport. Pharm Res 32:2937–2949.
    OpenUrl
  26. ↵
    1. Taub ME,
    2. Mease K,
    3. Sane RS,
    4. Watson CA,
    5. Chen L,
    6. Ellens H,
    7. Hirakawa B,
    8. Reyner EL,
    9. Jani M, and
    10. Lee CA
    (2011) Digoxin is not a substrate for organic anion-transporting polypeptide transporters OATP1A2, OATP1B1, OATP1B3, and OATP2B1 but is a substrate for a sodium-dependent transporter expressed in HEK293 cells. Drug Metab Dispos 39:2093–2102.
    OpenUrlAbstract/FREE Full Text
  27. ↵
    1. Tran TT,
    2. Mittal A,
    3. Aldinger T,
    4. Polli JW,
    5. Ayrton A,
    6. Ellens H, and
    7. Bentz J
    (2005) The elementary mass action rate constants of P-gp transport for a confluent monolayer of MDCKII-hMDR1 cells. Biophys J 88:715–738.
    OpenUrlCrossRefPubMed
  28. ↵
    1. van Meer G,
    2. Voelker DR, and
    3. Feigenson GW
    (2008) Membrane lipids: where they are and how they behave. Nat Rev Mol Cell Biol 9:112–124.
    OpenUrlCrossRefPubMed
  29. ↵
    1. Zamek-Gliszczynski MJ,
    2. Lee CA,
    3. Poirier A,
    4. Bentz J,
    5. Chu X,
    6. Ellens H,
    7. Ishikawa T,
    8. Jamei M,
    9. Kalvass JC,
    10. Nagar S, et al., and International Transporter Consortium
    (2013) ITC recommendations for transporter kinetic parameter estimation and translational modeling of transport-mediated PK and DDIs in humans. Clin Pharmacol Ther 94:64–79.
    OpenUrlCrossRefPubMed
View Abstract
Back to top

In this issue

Download PDF
Article Alerts
Email Article
Citation Tools

Research ArticleArticles

Generation of System-Independent Ki from IC50 Curves

Aqsaa Chaudhry, Git Chung, Adam Lynn, Akshata Yalvigi, Colin Brown, Harma Ellens, Michael O’Connor, Caroline Lee and Joe Bentz
Drug Metabolism and Disposition March 1, 2018, 46 (3) 279-290; DOI: https://doi.org/10.1124/dmd.117.075606
Reddit logo Twitter logo Facebook logo Mendeley logo

Jump to section

Advertisement