Advertisement
Research ArticleArticle

Calibration of In Vitro Multidrug Resistance Protein 1 Substrate and Inhibition Assays as a Basis to Support the Prediction of Clinically Relevant Interactions In Vivo

Agnès Poirier, Anne-Christine Cascais, Urs Bader, Renée Portmann, Marie-Elise Brun, Isabelle Walter, Alexander Hillebrecht, Mohammed Ullah and Christoph Funk
Drug Metabolism and Disposition September 2014, 42 (9) 1411-1422; DOI: https://doi.org/10.1124/dmd.114.057943

Abstract

The multidrug resistance protein 1 (MDR1) is known to limit brain penetration of drugs and play a key role in drug-drug interactions (DDIs). Theoretical cut-offs from regulatory guidelines are used to extrapolate MDR1 interactions from in vitro to in vivo. However, these cut-offs do not account for interlaboratory variability. Our aim was to calibrate our experimental system to allow better in vivo predictions. We selected 166 central nervous system (CNS) and non-CNS drugs to calibrate the MDR1 transport screening assay using Lewis lung cancer porcine kidney 1 epithelial cells overexpressing MDR1 (L-MDR1). A threshold efflux ratio (ER) of 2 was established as one parameter to assess brain penetration in lead optimization. The inhibitory potential of 57 molecules was evaluated using IC50 values based on the digoxin ER—IC50(ER)—or apparent permeability—IC50(Papp)—in L-MDR1 cells. Published clinical data for 68 DDIs involving digoxin as the victim drug were collected. DDI risk assessments were based on intestinal concentrations ([I2]) as well as unbound [I1u] and total plasma [I1T] concentrations. A receiver operating characteristic analysis identified an [I2]/IC50(ER) of 6.5 as the best predictor of a potential interaction with digoxin in patients. The model was further evaluated with a test set of 11 digoxin DDIs and 16 nondigoxin DDIs, resulting in only one false negative for each test set, no false positives among the digoxin DDIs, and two among the nondigoxin DDIs. Future refinements might include using cerebrospinal fluid to unbound plasma concentration ratios rather than therapeutic class, better estimation of [I2], and dynamic modeling of MDR1-mediated DDIs.

Introduction

The P-glycoprotein (P-gp), also known as multidrug resistance protein 1 (MDR1), is the best-studied and most well-characterized drug transporter. The physiological role of MDR1 is to protect the body from exposure to potentially toxic xenobiotics; it limits the absorption of drugs through the intestine, restricts access to the fetus and to critical organs such as the brain, and enhances drug excretion via biliary and renal secretion (Thiebaut et al., 1987; Zhang and Benet, 2001).

The transport of new drug candidates by MDR1 is therefore an important issue in drug development. Both the European Medicines Agency (EMA) (EMA, 2012) and the US Food and Drug Administration (FDA) (http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm292362.pdf) require that all investigational drugs are characterized for MDR1 interactions; they provide specific criteria and decision trees for assessing MDR1 substrates and inhibitors. MDR1 substrate profiling is of particular importance in the optimization of compounds that are intended for central nervous system (CNS) targets. Transcellular efflux assays are mostly used to assess in vitro MDR1 substrate interactions; compounds are selected based on their efflux ratio (ER) and bidirectional apparent permeability (Papp) (Giacomini et al., 2010; Caruso et al., 2013). Mahar Doan et al. were the first to propose cut-off values for Papp and ER parameters, based on data using MDR1-transfected MDCKII cells (Mahar Doan et al., 2002).

Drug-drug interactions (DDIs) resulting from comedication with MDR1 modulators are of particular concern when the victim drug has a narrow therapeutic index, such as digoxin (Cook et al., 2010). The EMA and FDA propose a simple, static mathematical model to quantitatively assess the risk of clinically relevant transporter-mediated DDIs. This model takes the ratio of the perpetrator drug concentration in vivo (total or unbound plasma or intestinal) against the in vitro inhibitory potential. The recent draft FDA regulations recommend considering [I1T]/IC50, [I1T] being the total plasma concentration and [I2]/IC50, where [I2] is the total gut concentration (http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm292362.pdf). Based on in vitro to in vivo extrapolations, the FDA recently proposed the use of 0.1 and 10 as reasonable cut-offs for [I1T]/IC50 and [I2]/IC50, respectively, aiming at a maximum accuracy of predictions, while minimizing the false negative (FN) rate (Zhang et al., 2008; Agarwal et al., 2013). The EMA follows a similar approach and provides thresholds for [I2]/IC50 of 10 and [I1u]/IC50 of 0.02 (http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2010/05/WC500090112.pdf).

The use of these cut-off values by all companies in a generic manner ignores the impact of interlaboratory variability, which might lead to inappropriate predictions. To help address this issue, the P-gp Inhibition Working Group has refined these cut-off values to account for interlaboratory variability in IC50 values (Bentz et al., 2013; Ellens et al., 2013). The use and variability of the different equations used to estimate MDR1 IC50 have been recently discussed and kinetically described (Sugimoto et al., 2011b; Bentz et al., 2013). Significant interlaboratory variability in IC50 estimations, partly due to the different experimental systems used, indicate that a universal cut-off is not optimal (http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm292362.pdf). Therefore, different groups have individually refined the cut-off values, mainly focusing on gut absorption and renal secretion of victim drugs (mainly digoxin; summarized in Table 1). There is very little information on the appropriate threshold to apply with regard to MDR1 inhibition, in the context of brain distribution. An in vitro to in vivo correlation has been developed in rat and human to evaluate the impact of MDR1-mediated DDI at the blood-brain barrier (Sugimoto et al., 2011a; Sugimoto et al., 2013). In addition, a recent review by the International Transporter Consortium has focused on the clinical relevance of DDIs at the blood-brain barrier (Kalvass et al., 2013).

View this table:
TABLE 1

[I1T]/IC50 and [I2]/IC50 threshold calibrations in different laboratories/companies and regulatory agencies

The in vitro tool and substrate, as well as the output parameter used to estimate the IC50 are listed in the second column. The fourth column contains the number of clinical studies used in the training set to establish the threshold in each group.

As described above, MDR1 substrate and inhibition assays have already been thoroughly evaluated in a number of academic and industrial laboratories. Therefore, the first purpose of our study was to calibrate, within our own laboratory, suitable cut-off criteria for MDR1 substrate and inhibition assays. The second purpose was to evaluate alternatives to the suggested methods by the FDA/EMA for assessing the need for MDR1-mediated clinical DDIs, such as the type and number of [I]/IC50 ratios and the use of more integrated approaches. Overall, we aimed at setting up and validating a high-throughput assay coupled with some efficient predictive models to select compounds that are 1) appropriate for CNS indication(s) by assessing MDR1 substrate properties, and 2) safe with respect to digoxin and related DDIs by assessing MDR1-inhibition properties.

Materials and Methods

The compounds used in the in vitro experiments were all of highest purity, at 80% from Sigma (Buchs, Switzerland), Fluka (Buchs, Switzerland), or Apin Chemicals (Abingdon, Oxfordshire, UK).

Cell Origin and Culture.

The parent Lewis lung cancer porcine kidney 1 cell line was transfected with human P-gp, resulting in Lewis lung cancer porcine kidney 1 epithelial cells overexpressing MDR1 (L-MDR1) which were obtained from Dr. Alfred Schinkel, The Netherlands Cancer Institute (Amsterdam, The Netherlands) and used under a license agreement. Cells were cultivated at 37°C, 5% CO2, and air-saturation humidity. Standard tissue culture flasks were obtained from Falcon and 96-well plates from Millipore (Darmstadt, Germany).

MDR1 96-Insert Plate Automated In Vitro Experiment.

The method used has been reported previously (Schwab et al., 2003). Briefly, cell lines were cultured on semipermeable inserts (surface area 0.11 cm2, pore size 0.4 µm; Millipore), and transport measurements were performed at day 3 or 4 after seeding. Cell monolayer integrity was assessed via the permeability of the extracellular marker, Lucifer yellow (10 µM), with a Papp cut-off of 25 nm/s. All test substrates were dosed at 1 µM. The assays were performed on the Freedom Evo 200 Base Workstation (Tecan, Männedorf, Switzerland) with an integrated Storex incubator (Liconic Instruments, Mauren, Liechtenstein). The medium was removed from the apical (100 µl) and basolateral (240 µl) compartments and replaced on the receiver side by culture medium without phenol red and with or without inhibitor. Transcellular transport was initiated by addition of media to the donor compartment containing test substrate and Lucifer yellow. The inhibitor (elacridar 1 µM or test compound) was added to both compartments. The transport experiment was performed in both directions in triplicate. Plates containing the inserts were incubated at 37°C and 5% CO2 under continuous shaking (100 rpm). Samples were taken from the donor and receiver compartments after 3.5 hours of incubation (2 hours for the inhibition study). The concentration of substrate in both compartments was determined by scintillation counting or high-performance liquid chromatography with tandem mass spectrometry. Lucifer yellow was quantified using a Spectrafluor Plus Reader (Tecan) set at 430/535 nm (excitation/emission). Triplicate inserts were used for each condition. Inserts with a Lucifer yellow permeation greater than 1%/hour were rejected. Digoxin (5 µM) with and without inhibitor (elacridar) was included on each 96-insert plate as a positive control. The acceptance range for mass balance (recovery) was 70% to 120%. For inhibition studies, probe substrates digoxin and talinolol were dosed at 5 µM and 1 µM, respectively. IC50 values were determined using seven concentration points along with a vehicle control, with the top concentration of each inhibitor selected based on the limit of solubility.

Analytics.

Analytical standards were prepared during the sample incubation as part of the assay. At the end of the experiment, samples were quenched with three volumes of acetonitrile containing the internal standards. Analyses of nonlabeled compounds were performed by high-performance liquid chromatography with tandem mass spectrometry. Briefly, fast reverse-phase liquid chromatography was conducted on a 10ADvp pump system (Shimadzu, Kyoto, Japan) coupled with a PAL HTS autosampler (CTC Analytics, Zwingen, Switzerland). The injection volume, mobile phase composition, analytical column, and gradient profile were optimized for each compound. Mass spectrometric detection was performed on an API 4000 or QTrap4000 system equipped with a TurboIonspray source (AB Sciex, Framingham, MA). Detection by tandem mass spectrometry was based on precursor ion transition to the strongest intensity product ion. Key instrumental conditions were optimized to yield best sensitivity. The typical run time was 1.5 minutes. The calibration range was typically 1 nM to 4 μM for each analyte. Analyst 1.4.2 software (AB Sciex) was used for data analysis. Concentration of compound in the samples was calculated from the peak area ratio between the analyte and the internal standard.

Data Evaluation.

The following equation was used for the evaluation of transcellular transport data:Embedded Imagewhere Papp, A, C0, and dQ/dt represent the apparent permeability, filter surface area, initial concentration, and amount transported per time period, respectively. Papp values were calculated on the basis of a single terminal time point. Transport efflux ratios were calculated as follows: Embedded Imagewhere PappBA is the permeability value in the basolateral-to-apical direction and PappAB the permeability value in the apical-to-basolateral direction. The average passive permeability value (Pappi) was calculated as follows:Embedded Imagewhere PappABi and PappBAi represent the apparent permeability in the apical-to-basolateral and basolateral-to-apical directions, respectively, in the presence of an inhibitor. A modified four-parameter Hill’s equation was used to estimate the IC50:Embedded Imagewhere X is either the ER, PappAB, or PappBA; Xmax is the maximum value without inhibitor; Xmin is the value with elacridar (maximum inhibition); [I] is the inhibitor concentration in µM; and s the slope factor (or Hill coefficient). Origin v7 (OriginLab Corporation, Northampton, MA) was used to perform the nonlinear fitting of the data and evaluation of IC50 and s parameters.

Clinical DDI Study Parameters.

Victim area under the curve (AUC), Cmax, and renal clearance fold ratios were extracted either from previous publications in which digoxin, fexofenadine, and talinolol clinical studies were already grouped and evaluated or from the clinical DDI original publications (Supplemental Table 3) (Zhang et al., 2008; Fenner et al., 2009; Tachibana et al., 2009; Cook et al., 2010; Sugimoto et al., 2011b). [I2] was calculated as the perpetrator dose in the clinical study of interest divided by 250 ml and converted into µM. [I1T] is defined as the maximum total plasma concentration of the perpetrator drug in the clinical DDI study. [I1T] was extracted either from previous publications in which digoxin clinical studies were already grouped and evaluated (Zhang et al., 2008; Fenner et al., 2009; Cook et al., 2010; Sugimoto et al., 2011b) or from the clinical DDI original publications. In cases where the perpetrator maximum plasma concentration was not measured in the specific clinical DDI, [I1T] was extracted from the Metabolism and Transport Drug Interaction Database from the University of Washington, (http://www.druginteractioninfo.org) and, if needed, levels were adjusted to the dose used in the clinical DDI study assuming dose linearity. The unbound fraction of drug in plasma (fup) was extracted either from the literature (Davit et al., 1999; Sugimoto et al., 2011b) or from the University of Washington (http://www.druginteractioninfo.org). The value of fup was approximated to 0.01 if the published value was below this. In addition, in cases of conflicting values or concentration-dependent protein binding, the most conservative value was used; i.e., the highest fup associated with the highest [I1u]. The unbound maximum plasma concentration of the perpetrator, [I1u], was calculated using [I1T] and fup.

Statistical Analysis.

A classic binary classification analysis was performed on all CNS and non-CNS drugs tested for MDR1 substrate properties. S was the total number of compounds tested, P the number of compounds marketed for CNS indications (CNS+), and N the number of compounds marketed for non-CNS indications (CNS−). The four possible outcomes were: 1) true positive (TP) when the in vitro data predicted penetration into brain for a compound marketed for a CNS indication, 2) false negative (FN) when the in vitro data did not predict penetration into brain for a compound marketed for a CNS indication, 3) false positive (FP), when the in vitro data predicted penetration into brain for a compound marketed for a non-CNS indication, and 4) true negative (TN) when the in vitro data predicted lack of penetration into brain for a compound marketed for a non-CNS indication (Tables 2 and 3).

View this table:
TABLE 2

First contingency description of the classification nomenclature for the marketed drugs screened for MDR1 transport and inhibition used within the manuscript

View this table:
TABLE 3

Second contingency description of the classification nomenclature for the marketed drugs screened for MDR1 transport and inhibition used within the manuscript

A classic binary classification analysis was also performed on all clinical digoxin DDI studies. S was the total number of studies, P the number of DDI studies with clinically significant digoxin interactions (AUCi/AUC > 1.25 or Cmax,i,ss/Cmax,ss > 1.25), and N is the number of insignificant interactions with digoxin (AUCi/AUC < 1.25 and Cmax,i,ss/Cmax,ss < 1.25). The four possible outcomes were: 1) true positive: the in vitro data predict a significant DDI with digoxin, and the compound exhibits this interaction in clinics; 2) false negative: the in vitro data do not predict a significant DDI with digoxin, but the compound causes a significant interaction in clinics; 3) false positive: the in vitro data predict a significant DDI with digoxin, but it is not observed in clinics; and 4) true negative: the in vitro data predict an absence of significant interaction with digoxin, and the compound shows this behavior in clinics.

Performance metrics used in the analyses are defined as follows:Embedded ImageThe performance metrics outlined above depend upon the choice of a specific threshold for the predictor (assay readout; e.g., ER). A receiver operating characteristic (ROC) curve plots the true positive rate against the false positive rate at all possible thresholds (Fawcett, 2004). The AUCROC allows for a threshold-independent comparison of classification performance between several predictors. For the DDI data set, a standardized partial AUC (pAUCROC) focusing the analysis on the high sensitivity region between 90% and 100% sensitivity is also reported. In either case, an area of 0.5 corresponds to a random classifier, whereas 1.0 indicates perfect class separation. All calculated metrics are based on the empirical ROC curve. For the calculation of thresholds yielding a given sensitivity, the value with the highest possible specificity is used. Thresholds are calculated by linear interpolation; further algorithmic details are described elsewhere (Robin et al., 2011). ROC analysis was performed with the package pROC 1.5.4 (Robin et al., 2011) using the R language for statistical computing version 2.15.1 (RCoreTeam, 2012; http://cran.r-project.org/src/contrib/Archive/pROC/).

Results

Refinement of an MDR1 Transport Exclusion Criterion for CNS Indications.

Results for 76 out of the tested 79 drugs marketed for CNS indications are reported (Supplemental Table 1). For three drugs (baclofen, gabapentin, and sumatriptan) the samples were all below the limit of quantification. Using the limit of quantification to estimate an upper bound of passive permeability resulted on average in permeability below 15 nm/s. The passive permeability values (mean values of apparent bidirectional permeabilities in presence of inhibitor; Pappi) ranged from 9–445 nm/s with a mean of 232 nm/s, and the ER values ranged from 0.5–28 with a mean value around 1.9. The results obtained for 91 out of 117 non–CNS indicated drugs that were tested are reported (Supplemental Table 2). Nine drugs were either unstable, not soluble, or there was no suitable analytical method; for 17 drugs the samples were all below the limit of quantification (Papp < 15 nm/s). The passive permeability values ranged from 6–351 nm/s with a mean of 135 nm/s, and the ER values ranged from 0.5–56 with a mean value around 8.4.

The results obtained in this report reveal some differences from a previous publication by Mahar Doan et al. (Mahar Doan et al., 2002). Both the mean and range for the permeability are about 2-fold below the mean values reported by Mahar Doan et al. for both groups (CNS and non-CNS drugs) and in addition, the highest ERs were on average below the results reported previously (Mahar Doan et al., 2002). From the 167 results (Supplemental Tables 1 and 2), equivalent transcellular permeability and MDR1-related ERvalues have already been described for 139 drugs (Seelig, 1998; Polli et al., 2001; Mahar Doan et al., 2002; Wandel et al., 2002; Chen et al., 2003; Singh et al., 2003; Summerfield et al., 2007; Kanaan et al., 2009; Berginc et al., 2010; Gertz et al., 2010; Gnoth et al., 2011; Moons et al., 2011; Hellinger et al., 2012; Wittgen et al., 2012). The passive permeability of five drugs (chlorprothixene, trazodone, astemizole, terfenadine, and saquinavir) were from 6- to 20-fold lower than the values previously reported, although their overall data categorizing MDR1 substrate interaction remained the same (Mahar Doan et al., 2002). Conversely, sertraline showed a 6-fold higher permeability than previously reported values (Summerfield et al., 2007). Olanzapine and citalopram have been described as substrates in MDR1 overexpressing cells but tested negative in the present study, whereas sildenafil, which was previously described as a non-MDR1 substrate in MDCKII-MDR1 and Caco-2 cells, tested positive here (Summerfield et al., 2007; Gertz et al., 2010). Oxycodone and aripiprazole have been described as MDR1 substrates in Caco-2 cells but tested negative in the present study; thus, another efflux transporter might be responsible for the observed efflux in Caco-2 cells (Hassan et al., 2007; Moons et al., 2011). Finally, transcellular permeability measurements with MDR1 overexpressing cells are reported for the first time for 17 drugs (paroxetine, codeine, alprazolam, moclobemide, pentobarbital, oxymorphone, atropine, lorazepam, enalaprilat, isotretinoin, fumitremorgin C, pioglitazone, nimodipine, ezetimibe, piroxicam, cilazapril, and amlodipine). Results for most of the 20 drugs that had Pappi values less than 15 nm/s are consistent with previously reported low permeabilities (Polli et al., 2001; Mahar Doan et al., 2002; Summerfield et al., 2007; Berginc et al., 2010; Hellinger et al., 2012).

Figure 1A shows the data with ER and Pappi grouped into CNS and non-CNS indications. The ER values of 86% of the CNS drugs were below 2; only two compounds had an ER above 5 (bromocriptine and eletriptan), and 89% showed a permeability above 100 nm/s, with 67% above 200 nm/s. The drugs from the non-CNS indication group showed a different distribution; the ERs of 39% were below 2 and 39% above 5, while 52% showed a permeability below 100 nm/s.

Fig. 1.
Fig. 1.

Number of compounds versus ER or mean Pappi in CNS and non-CNS indications for marketed drugs (A), (detailed tabulated data in Supplemental Data), and for Roche compounds entering phase I clinical trials (B). Pappi, mean substrate permeability in presence of inhibitor.

Figure 1B shows the same analysis with internal compounds in development having reached clinical phase I trials for CNS and non-CNS indications. The ERs of 81% of the CNS-indicated drugs were below 2, with only three compounds with an ER between 2 and 10. Most (75%) showed permeability above 100 nm/s, with 62% above 200 nm/s. The drugs from the non-CNS indication group showed a different distribution, with only 11% of ER values below 2, and 68% above 5, while 58% showed permeability below 100 nm/s.

The results of the ROC analysis performed on MDR1 substrate properties are presented in Fig. 2. Estrone-3-sulfate (not a marketed drug) was excluded from the data analysis. The graphs do not show a conventional ROC curve, but instead sensitivity, specificity, and accuracy are plotted as a function of the threshold for ease of visualization, whereas area calculations are based on the standard ROC curves. A similar analysis was performed for passive permeability (Pappi, data not shown). The area under the ROC curve was 0.778 for ER and 0.754 for Pappi, respectively. If the ER threshold is set at 1.5, specificity, sensitivity, and accuracy are all approximately 70%. Increasing the ER threshold to 2 decreases the specificity to 60% (equivalent to 40% FP), while sensitivity and accuracy increase to reach 88% and 73%, respectively. With respect to the permeability threshold, the accuracy was stable (∼70%) from 100 to 230 nm/s. In this permeability range, the sensitivity decreases, and the specificity increases (both between 60% and 80%) with increasing permeability.

Fig. 2.
Fig. 2.

ROC analysis of the CNS data set. The curves show the sensitivity (dotted lines), specificity (dashed lines), and accuracy (solid lines) as a function of the efflux ratio cut-off. All compounds below the threshold are predicted to permeate into the CNS. The plot in panel (A) shows the full ER range, whereas the plot in panel (B) zooms into the ER range between 0.5 and 2.5, respectively.

Refinement of MDR1 Inhibition Exclusion Criteria for Clinically Relevant Drug-Drug Interactions.

Fifty-seven drugs identified from selected clinical DDI studies were tested in vitro for MDR1 inhibition, and the results were evaluated for IC50 using the ER or Papp parameters and are reported in Table 4. IC50 values based on the ER were calculated for 36 drugs that showed complete inhibition, while for compounds that did not show more than 50% inhibition, the IC50 was designated as the highest concentration tested. The three MDR-modulating agents: zosuquidar, elacridar, and valspodar, showed the strongest inhibition (IC50 < 0.5 µM). Among the significant inhibitors (IC50 < 100 µM), 60% belong to the cardiovascular therapeutic class, whereas they represented less than 20% of the noninhibitors (IC50 > 100 µM). The IC50 values based on Papp were systematically higher, irrespective of whether PappAB or PappBA were used, making it impossible to fit an IC50, even though an IC50 based on ER could be estimated for 14 and 17 compounds, respectively.

View this table:
TABLE 4

Inhibition in vitro results of digoxin MDR1-mediated efflux: IC50s based on ER, PappAB, or PappBA, ordered from the lowest IC50(ER) to the highest

For 29 compounds (Table 4), the inhibitory potential on digoxin efflux has been reported previously by Sugimoto et al. using L-MDR1 cells. Of these compounds, 22 of our IC50 values were within 2-fold of the previously reported data (Sugimoto et al., 2011b). For the seven outliers, the IC50 values reported in Table 4 are mostly higher than those reported by Sugimoto et al. by up to 43-fold (simvastatin). Further comparison with data from Cook et al., who previously used Caco-2 cells to test 25 of the drugs listed in Table 4, show that 18 of our IC50 values are within 2-fold of their IC50 values based on ER (Cook et al., 2010). For the seven outliers, values reported in Table 4 are all above the ones reported by Cook et al.; up to 6-fold higher (omeprazole).

Characteristics of 68 clinical studies involving digoxin as a victim were collected and are summarized (Supplemental Table 3); 33 studies were categorized as relevant DDI (AUCi/AUC > 1.25 or Cmax,i,ss/Cmax,ss > 1.25) and 35 as nonsignificant (AUCi/AUC < 1.25 and Cmax,i,ss/Cmax,ss < 1.25) (Ellens et al., 2013). The predictability of different [I]/IC50 ratios were tested, using IC50 estimations based on either ER, PappAB, or PappBA combined with either [I2], [I1T], or [I1u] (nine combinations in total). When the IC50 exceeded the tested concentration range, the highest tested concentration was used for the calculation of the [I]/IC50 ratios (e.g., repaglinide IC50(ER) > 100 µM and [I2] = 18 µM, then [I2]/IC50(ER) = 0.18). The accuracy, sensitivity, and specificity of each ratio applying the FDA thresholds (0.1 for [I1]/IC50 and 10 for [I2]/IC50) are presented in the upper panel of Fig. 3. The accuracy was best using [I2] (76% to 82%) and worst using [I1u] (54% to 60%). The objective was to obtain a low FN rate, thus a higher sensitivity, to avoid missing a potential safety issue during drug development. The highest sensitivity by far was obtained using [I2]/IC50(ER) (91% corresponding to three FN predictions) and the lowest sensitivity with [I1u] (6% to 18%).

Fig. 3.
Fig. 3.

Accuracy, sensitivity, and specificity of the different [I]/IC50 models for the prediction of digoxin MDR1-mediated clinical DDI: free and total [I1]/IC50 and [I2]/IC50 (using different IC50s). Raw data are reported in Supplemental Table 5. PappAB, substrate (digoxin) permeability from apical to basolateral; PappBA, substrate (digoxin) permeability from basolateral to apical; IC50(ER), IC50 values based on digoxin ER; IC50(Papp), IC50 values based on digoxin permeability.

The results of the ROC analysis on the clinical studies are presented in Table 5 and Fig. 3. The area under the ROC curve is reported for each of the nine [I]/IC50 predictors. [I2]/IC50 showed a higher discriminatory power (AUCROC ∼ 0.85) compared with [I1T]/IC50 or [I1u]/IC50 (AUCROC 0.75–0.80, respectively). Furthermore, [I]/IC50(ER) ratios showed a consistently higher discriminatory power over [I]/IC50(PappAB) or [I]/IC50(PappBA) (Table 5). The same trend holds when focusing the analysis on the high sensitivity region (90% to 100% sensitivity) as indicated by the partial AUC (pAUCROC) (Table 5).

View this table:
TABLE 5

Results of the ROC analysis on nine [I]/IC50 ratios for the digoxin DDI data set

The total AUC (AUCROC), the standardized partial AUC (pAUCROC) integrated over the 90% to 100% sensitivity interval (focusing on the region of FN rate 10% to 0%) and the threshold yielding 6% FN rate are given.

Of the 33 clinically relevant DDIs, captopril and talinolol were impossible to predict and showed IC50s greater than 1000 and 600 µM, respectively, resulting in very low [I]/IC50 ratios. Thus, 6% (two FN studies) was considered as the lowest realistically achievable FN rate within this data set, so we calculated the thresholds yielding 94% sensitivity. The resultant refined thresholds are presented in Table 5. These refined thresholds are all more conservative than the FDA proposed cut-offs; 60- to 100-fold lower for [I1T], 970- to 3300-fold lower for [I1u], and,1.5- to 5-fold lower for [I2]. Using the refined thresholds, accuracy, sensitivity, and specificity were estimated (Fig. 3). The sensitivity was 94% for all nine [I]/IC50 predictors by design; the accuracy and specificity were best for all [I2]-derived ratios and increased when using IC50(ER).

A test set was selected containing four relevant and seven nonrelevant clinical DDIs. There were no FPs and only one FN, ambrisentan, which did not inhibit MDR1 in vitro (IC50 > 1000 µM, Table 6), supporting recently published data (Richards et al., 2011; Ellens et al., 2013). A different threshold would have therefore not helped to detect this FN. Most importantly, although a digoxin Cmax increase of 1.29 was observed in the clinical DDI between ambrisentan and digoxin, the reported AUC increase was only of 1.09 and there were no safety findings. The original publication categorized the interaction as unlikely to be clinically relevant (Richards et al., 2011). Overall, the model was applied to the test set with high success.

View this table:
TABLE 6

Contingency information for DDI training and testing data sets

Test set depicts clinical DDIs with digoxin as a victim drug; numbers in parentheses represent clinical DDI with fexofenadine or talinolol as victim drugs. All percentages represent TP/Positive, FN/Positive, FP/Negative, and TN/Negative.

To understand the clinical DDIs in which the victim drug was not digoxin but another MDR1 substrate, the test set was extended to 16 clinical DDIs in which either fexofenadine or talinolol were the victim drugs (Supplemental Table 4). Three clinical DDIs were identified for talinolol with verapamil, simvastatin, or erythromycin as perpetrators (Supplemental Table 4), and the associated IC50 data were generated using the same methods as before but with talinolol as the probe substrate. The static model predicted the simvastatin clinical DDI correctly, with an [I2]/IC50(ER) of 1.3, which is below the refined threshold (6.5) for a nonsignificant exposure increase. Verapamil and erythromycin both showed an [I2]/IC50(ER) above the established threshold (230 and 36, respectively), even though verapamil decreased rather than increased talinolol exposure (Table 6).

Thirteen clinical DDIs were identified for fexofenadine, involving eight different perpetrators (Supplemental Table 3). Fexofenadine ER could not be estimated in the assay due to low analytical sensitivity and limited permeability. Therefore, it was decided to use the IC50 that was estimated using digoxin as the probe substrate (Table 4). The static model correctly predicted omeprazole and cimetidine clinical DDIs, with an [I2]/IC50(ER) below the established threshold (4.2 and 3.2, respectively) for nonsignificant exposure increases. However, there was one FP (diltiazem and fexofenadine) and one FN with erythromycin dosed at 300 mg—[I2]/IC50(ER) = 5.4, maximal exposure increase = 2.09.

Discussion

MDR1 plays an important role in drug distribution to the brain and is one of the main determinants of digoxin absorption and renal clearance and associated DDIs. Health authorities have recommended thresholds for MDR1 substrate and inhibition assays to guide drug development, while a number of groups have reported improved in vitro to in vivo correlations. Here, we aimed at calibrating a widely used cellular MDR1 substrate and inhibition assay, with a view to development of safe and effective drugs in line with health authorities’ regulations.

Refinement of MDR1 Transport Exclusion Criteria for CNS Indications.

We tested 196 drugs for transcellular permeability and MDR1 substrate properties. Data are reported for 167 drugs (Supplemental Tables 1 and 2), of which 17 are newly published results. From the 150 previously tested drugs, 139 results correlated with published measurements, and a systematic 2-fold difference in mean passive permeability compared with published MDCKII-MDR1 results was observed (Mahar Doan et al., 2002). These discrepancies are likely due to differences in cell type and experimental parameters, such as substrate concentration, underlining the importance of experimental system calibration.

The present work aims at calibrating the assay in its current set-up and establishing a relevant threshold for transport by human MDR1, which is often used to deprioritize compounds from development that have a reduced probability to enter the human brain. Therefore, a low FP rate (high specificity) was favored to prevent non–brain penetrating compounds to be carried through the development process.

Given the absolute error measurement of around 0.5 (S.D. in Supplemental Tables 1 and 2) and the results from the ROC analysis, the human MDR1 ER cut-off value was set to 2, equivalent to the generic FDA recommendation.

An important limitation of this approach is that in vitro predictors for CNS penetration (ER, permeability) were correlated with the therapeutic indication (CNS and non-CNS categories). In addition to the penetration into the CNS, the pharmacologic activity at the CNS target is equally important; also, cerebrospinal fluid to unbound plasma ratios could be used as endpoints, as has been done for the analysis of rodent data (Caruso et al., 2013). However, based on data availability, the therapeutic class has been used as the endpoint in this study.

Refinement of MDR1 Transport Exclusion Criteria for Clinically Relevant Drug-Drug Interactions.

Digoxin is between 60% to 85% intestinally absorbed (depending on the oral formulation) (Johnson et al., 1977) and mainly eliminated from the body by glomerular filtration and MDR1-mediated tubular secretion; its metabolism is negligible. Several MDR1 inhibitors have been reported to increase the levels of digoxin, with its narrow therapeutic index, above its upper limit of safety (Fenner et al., 2009). Cook et al. established a 25% cut-off criterion in digoxin AUC increase based on toxic effects (Cook et al., 2010). This was used here to categorize the clinical relevance of 68 clinical digoxin DDIs and to test the predictive power of nine [I]/IC50 predictors.

Table 4 contains 14 IC50 outliers (more than 2-fold difference) based on previously published results using L-MDR1 and Caco-2 cells (Cook et al., 2010; Sugimoto et al., 2011b). Differences in cell type, experimental setup, IC50 calculation methods, and interlaboratory variability can explain these discrepancies, as recently discussed (Bentz et al., 2013). This observation also underlines the importance of an individual assay calibration.

The predictive powers of IC50(ER), IC50(PappAB), and IC50(PappBA) were evaluated. The mathematical relationship between IC50(ER) and IC50(PappBA) has been described (Sugimoto et al., 2011b) and the variability of IC50 generated from different equations was assessed (Bentz et al., 2013). In our hands, IC50(PappAB) and IC50(PappBA) were almost identical, on average 3.5-fold higher than IC50(ER). For 17 compounds, IC50 values, based on ER, could be estimated, while it was impossible to fit an IC50 based on Papp (Table 4). Therefore, the ER appears to be a more sensitive parameter for estimating the degree of MDR1 inhibition than Papp. The AUCROC was only slightly better for [I2]/IC50(ER) versus [I2]/IC50(Papp), but [I2]/IC50(Papp) yielded a much lower specificity (54%) compared with [I2]/IC50(ER) (71%) using the optimized thresholds. To reduce the FP rate, the further digoxin/MDR1 risk assessment was based on IC50(ER) (Fig. 4).

Fig. 4.
Fig. 4.

Roche internal decision tree for MDR1 inhibition properties. In a first step, a single concentration of inhibitor is tested on digoxin MDR1-mediated transcellular transport (either 30 µM or 10-fold predicted intestinal concentration); if positive, the inhibitor IC50 based on digoxin efflux ratio—IC50(ER)—is estimated and the ratio [I2]/ IC50(ER) calculated. The threshold of 6.5 is used to advise on potential DDI clinical studies.

The superiority of using [I2] versus [I1] for the prediction of MDR1-mediated digoxin DDIs based on the ROC analysis indicates that inhibition of digoxin absorption in the gastrointestinal tract is a more significant site of interaction than the active renal secretion of digoxin in the kidney (via MDR1). Indeed, only modest changes in digoxin CLrenal have been reported in comparison with changes in plasma exposure (either Cmax or AUC) (Fenner et al., 2009). Another important difference of using [I2] versus [I1] is the finding that neither of the two FNs from the data set presented here could be categorized as TPs by the use of [I1], and out of the 11 FPs generated by the [I2]/IC50(ER) (cut-off of 6.5), only two, at most, would be categorized as TNs using [I1]. The refined [I1] cut-offs would generate more than 50% FPs compared with 29% with [I2]/IC50(ER) without any gain in accuracy. More and more groups focus on intestinal concentrations for the prediction of MDR1-mediated digoxin DDIs, and the approach is equally supported by regulators (Tachibana et al., 2009; Sugimoto et al., 2011b; Agarwal et al., 2013; Tweedie et al., 2013).

The digoxin/MDR1 risk assessment based on [I2], for digoxin interactions and oral administration of both victim and perpetrator, is summarized in a decision tree (Fig. 4). A refined cut-off of 6.5 gives the best discriminatory power of [I2]/IC50(ER). This new threshold is slightly more conservative than the one proposed by the FDA but is tailored to the in vitro experimental conditions and IC50 calculation method presented here, using orally administered digoxin as the victim drug. The refined cut-off was challenged with a test set including four relevant and seven nonrelevant clinical DDIs. The analysis revealed no FPs and only one FN (ambrisentan).

To further test the model beyond digoxin, other victim drugs were evaluated. Dabigatran etexilate, the double prodrug of dabigatran, was recently recommended by the EMA as an in vivo probe for MDR1-mediated intestinal DDIs (http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2012/07/WC500129606.pdf). Dabigatran etexilate is a substrate of MDR1, has very low oral bioavailability (≈6.5%), and exhibits altered pharmacokinetics due to MDR1 inhibition in the intestine (Ishiguro et al., 2014). Therefore, dabigatran etexilate would serve as good alternative victim drug to test DDI extrapolations. However, the lack of a suitable analytical method for this double-prodrug precluded its use in our assay setup. Instead, talinolol and fexofenadine were selected. The bioavailability of talinolol is quite low (about 55%), and its absorption from the gastrointestinal tract is affected by a dose-dependent, saturable process involving MDR1-dependent intestinal secretion (Gramatte et al., 1996; Gramatte and Oertel, 1999). Fexofenadine, an H1-receptor antagonist, has a low bioavailability (35%), undergoes minimal metabolism, and is frequently used as an in vivo probe substrate for MDR1 or organic anion transporting polypeptides (Lappin et al., 2010). Fexofenadine is mainly excreted into bile with ∼30% (of an intravenous dose) eliminated into urine. Talinolol and fexofenadine are therefore considered as suitable model compounds for intestinal MDR1-mediated DDIs; they are also well tolerated and characterized across a broad therapeutic dose range (Spahn-Langguth and Langguth, 2001; Kawashima et al., 2002; Lappin et al., 2010).

One debatable assumption in the present approach is that [I2] equals the oral dose divided by 250 ml. This may result in a considerable overestimation of the intestinal concentration, particularly for poorly soluble compounds. Other groups have used fraction absorbed, absorption rate, and enterocyte blood flow, as well as built-in functions in physiologically–based pharmacokinetic (PBPK) models to estimate gut concentrations (Agarwal et al., 2013; Reyner et al., 2013). Further exploring the caveats of the static model, three main assumptions should be mentioned: 1) the victim drug absorption, elimination or brain distribution is driven by a single pathway; 2) the neglect of the dynamic nature of the involved processes; and 3) the intestinal, plasma, and extracellular concentrations used ([I2], [I1], or IC50) do not correspond to the intracellular MDR1 binding site. To address all these weaknesses and to reduce the FP rate, the predictions of a commercially available integrated DDI PBPK model (SimCYP) were tested using physicochemical properties as input parameters (for both the victim and perpetrator drugs) simulating the respective clinical DDI. This approach unfortunately failed to improve predictions as compared with the static model (data not shown).

General Conclusion.

In conclusion, MDR1 inhibition assays were calibrated under specific experimental conditions. Based on MDR1 transport data from 166 CNS and non-CNS marketed drugs, a cut-off value of 2 for the ER in L-MDR1 cells would support the optimization of brain-penetrating molecules. For MDR1 substrate properties, 68 distinct digoxin clinical DDIs were assessed; a cut-off value of 6.5 for [I2]/IC50(ER) in L-MDR1 cells was established to best predict an interaction with orally administered digoxin. Further parameters, including human cerebrospinal fluid concentration data and dynamic estimation of [I2] using an integrated PBPK approach, along with multicompartmental analysis of in vitro kinetic data, should help to further improve the predictions.

Acknowledgments

The authors thank Dr. Kuresh Youdim for support in DDI PBPK simulations and discussions and James Michael Zimmerman for proofreading and editing.

Authorship Contributions

Participated in research design: Poirier, Cascais, Portmann, Brun, Ullah, Funk.

Conducted experiments: Cascais, Bader, Portmann, Brun.

Contributed new reagents or analytic tools: Bader, Walter.

Performed data analysis: Poirier, Cascais, Portmann, Brun, Hillebrecht, Ullah, Funk.

Wrote or contributed to the writing of the manuscript: Poirier, Cascais, Walter, Hillebrecht, Ullah, Funk.

Footnotes

Abbreviations

AUC
area under the curve
AUCi
area under the curve in presence of inhibitor
Cmax,ss
maximal total plasma concentration at steady state
Cmax,i,ss
maximal total plasma concentration at steady state in presence of inhibitor
CNS
central nervous system
DDI
drug-drug interactions
EMA
European Medicines Agency
ER
efflux ratio
FDA
US Food and Drug Administration
FN
false negative
FP
false positive
fup
unbound fraction of drug in plasma
[I2]
intestinal concentration
[I1u]
unbound plasma concentration
[I1T] total plasma concentration
LLC-PK1
Lewis lung cancer porcine kidney 1 epithelial cells
L-MDR1
Lewis lung cancer porcine kidney 1 epithelial cells overexpressing MDR1
MDR1
multidrug resistance protein 1
Papp
apparent permeability
PappAB
permeability value in the apical-to-basolateral direction
PappBA
permeability value in the basolateral-to-apical direction
Pappi
passive permeability
PBPK
physiologically-based pharmacokinetic
P-gp
P-glycoprotein
ROC
receiver operating characteristic
TN
true negative
TP
true positive

References

  1. ↵
    1. Agarwal S,
    2. Arya V, and
    3. Zhang L
    (2013) Review of P-gp inhibition data in recently approved new drug applications: utility of the proposed [I(1) ]/IC(50) and [I(2) ]/IC(50) criteria in the P-gp decision tree. J Clin Pharmacol 53:228–233.
    OpenUrlCrossRef
  2. ↵
    1. Bentz JE,
    2. O'Connor M,
    3. Bednarczyk D,
    4. Coleman J,
    5. Lee CA,
    6. Palm JE,
    7. Pak A,
    8. Perloff ES,
    9. Reyner EL,
    10. Balimane P,
    11. 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
  3. ↵
    1. Berginc K,
    2. Žakelj S, and
    3. Kristl A
    (2010) In vitro interactions between aged garlic extract and drugs used for the treatment of cardiovascular and diabetic patients. Eur J Nutr 49:373–384.
    OpenUrlCrossRefPubMed
  4. ↵
    1. Caruso A,
    2. Alvarez-Sánchez R,
    3. Hillebrecht A,
    4. Poirier A,
    5. Schuler F,
    6. Lavé T,
    7. Funk C, and
    8. Belli S
    (2013) PK/PD assessment in CNS drug discovery: Prediction of CSF concentration in rodents for P-glycoprotein substrates and application to in vivo potency estimation. Biochem Pharmacol 85:1684–1699.
    OpenUrlCrossRef
  5. ↵
    1. Chen C,
    2. Hanson E,
    3. Watson JW, and
    4. Lee JS
    (2003) P-glycoprotein limits the brain penetration of nonsedating but not sedating H1-antagonists. Drug Metab Dispos 31:312–318.
    OpenUrlAbstract/FREE Full Text
  6. ↵
    1. Cook JA,
    2. Feng B,
    3. Fenner KS,
    4. Kempshall S,
    5. Liu R,
    6. Rotter C,
    7. Smith DA,
    8. Troutman MD,
    9. Ullah M, and
    10. Lee CA
    (2010) Refining the in vitro and in vivo critical parameters for P-glycoprotein, [I]/IC50 and [I2]/IC50, that allow for the exclusion of drug candidates from clinical digoxin interaction studies. Mol Pharm 7:398–411.
    OpenUrlCrossRefPubMed
  7. ↵
    1. Davit B,
    2. Reynolds K,
    3. Yuan R,
    4. Ajayi F,
    5. Conner D,
    6. Fadiran E,
    7. Gillespie B,
    8. Sahajwalla C,
    9. Huang SM, and
    10. Lesko LJ
    (1999) FDA evaluations using in vitro metabolism to predict and interpret in vivo metabolic drug-drug interactions: impact on labeling. J Clin Pharmacol 39:899–910.
    OpenUrlCrossRefPubMed
  8. ↵
    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,
    11. 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
    1. Elsby R,
    2. Gillen M,
    3. Butters C,
    4. Imisson G,
    5. Sharma P,
    6. Smith V, and
    7. Surry DD
    (2011) The utility of in vitro data in making accurate predictions of human P-glycoprotein-mediated drug-drug interactions: a case study for AZD5672. Drug Metab Dispos 39:275–282.
    OpenUrlAbstract/FREE Full Text
  10. ↵
    Fawcett T (2004) ROC graphs: notes and practical considerations for data mining researchers. Technical Report HPL-2003-4. Hewlett-Packard Labs, Palo Alto, CA.
  11. ↵
    1. Fenner KS,
    2. Troutman MD,
    3. Kempshall S,
    4. Cook JA,
    5. Ware JA,
    6. Smith DA, and
    7. Lee CA
    (2009) Drug-drug interactions mediated through P-glycoprotein: clinical relevance and in vitro-in vivo correlation using digoxin as a probe drug. Clin Pharmacol Ther 85:173–181.
    OpenUrlCrossRefPubMed
  12. ↵
    1. Gertz M,
    2. Harrison A,
    3. Houston JB, and
    4. Galetin A
    (2010) Prediction of human intestinal first-pass metabolism of 25 CYP3A substrates from in vitro clearance and permeability data. Drug Metab Dispos 38:1147–1158.
    OpenUrlAbstract/FREE Full Text
  13. ↵
    1. Giacomini KM,
    2. Huang SM,
    3. Tweedie DJ,
    4. Benet LZ,
    5. Brouwer KL,
    6. Chu X,
    7. Dahlin A,
    8. Evers R,
    9. Fischer V,
    10. Hillgren KM,
    11. et al., and
    12. International Transporter Consortium
    (2010) Membrane transporters in drug development. Nat Rev Drug Discov 9:215–236.
    OpenUrlCrossRefPubMed
  14. ↵
    1. Gnoth MJ,
    2. Buetehorn U,
    3. Muenster U,
    4. Schwarz T, and
    5. Sandmann S
    (2011) In vitro and in vivo P-glycoprotein transport characteristics of rivaroxaban. J Pharmacol Exp Ther 338:372–380.
    OpenUrlAbstract/FREE Full Text
  15. ↵
    1. Gramatté T and
    2. Oertel R
    (1999) Intestinal secretion of intravenous talinolol is inhibited by luminal R-verapamil. Clin Pharmacol Ther 66:239–245.
    OpenUrlCrossRefPubMed
  16. ↵
    1. Gramatté T,
    2. Oertel R,
    3. Terhaag B, and
    4. Kirch W
    (1996) Direct demonstration of small intestinal secretion and site-dependent absorption of the beta-blocker talinolol in humans. Clin Pharmacol Ther 59:541–549.
    OpenUrlCrossRefPubMed
  17. ↵
    1. Hassan HE,
    2. Myers AL,
    3. Lee IJ,
    4. Coop A, and
    5. Eddington ND
    (2007) Oxycodone induces overexpression of P-glycoprotein (ABCB1) and affects paclitaxel’s tissue distribution in Sprague Dawley rats. J Pharm Sci 96:2494–2506.
    OpenUrlCrossRefPubMed
  18. ↵
    1. Hellinger E,
    2. Veszelka S,
    3. Tóth AE,
    4. Walter F,
    5. Kittel A,
    6. Bakk ML,
    7. Tihanyi K,
    8. Háda V,
    9. Nakagawa S,
    10. Duy TD,
    11. et al.
    (2012) Comparison of brain capillary endothelial cell-based and epithelial (MDCK-MDR1, Caco-2, and VB-Caco-2) cell-based surrogate blood-brain barrier penetration models. Eur J Pharm Biopharm 82:340–351.
    OpenUrlCrossRefPubMed
  19. ↵
    1. Ishiguro N,
    2. Kishimoto W,
    3. Volz A,
    4. Ludwig-Schwellinger E,
    5. Ebner T, and
    6. Schaefer O
    (2014) Impact of endogenous esterase activity on in vitro p-glycoprotein profiling of dabigatran etexilate in Caco-2 monolayers. Drug Metab Dispos 42:250–256.
    OpenUrlAbstract/FREE Full Text
  20. ↵
    1. Johnson BF,
    2. Smith G, and
    3. French J
    (1977) The comparability of dosage regimens of Lanoxin tablets and Lanoxicaps. Br J Clin Pharmacol 4:209–211.
    OpenUrlCrossRefPubMed
  21. ↵
    1. Kalvass JC,
    2. Polli JW,
    3. Bourdet DL,
    4. Feng B,
    5. Huang SM,
    6. Liu X,
    7. Smith QR,
    8. Zhang LK,
    9. Zamek-Gliszczynski MJ, and
    10. International Transporter Consortium
    (2013) Why clinical modulation of efflux transport at the human blood-brain barrier is unlikely: the ITC evidence-based position. Clin Pharmacol Ther 94:80–94.
    OpenUrlCrossRefPubMed
  22. ↵
    1. Kanaan M,
    2. Daali Y,
    3. Dayer P, and
    4. Desmeules J
    (2009) Uptake/efflux transport of tramadol enantiomers and O-desmethyl-tramadol: focus on P-glycoprotein. Basic Clin Pharmacol Toxicol 105:199–206.
    OpenUrlCrossRefPubMed
  23. ↵
    1. Kawashima M,
    2. Harada S, and
    3. Tango T
    (2002) Review of fexofenadine in the treatment of chronic idiopathic urticaria. Int J Dermatol 41:701–706.
    OpenUrlCrossRefPubMed
  24. ↵
    1. Lappin G,
    2. Shishikura Y,
    3. Jochemsen R,
    4. Weaver RJ,
    5. Gesson C,
    6. Houston B,
    7. Oosterhuis B,
    8. Bjerrum OJ,
    9. Rowland M, and
    10. Garner C
    (2010) Pharmacokinetics of fexofenadine: evaluation of a microdose and assessment of absolute oral bioavailability. Eur J Pharm Sci 40:125–131.
    OpenUrlCrossRefPubMed
  25. ↵
    1. Mahar Doan KM,
    2. Humphreys JE,
    3. Webster LO,
    4. Wring SA,
    5. Shampine LJ,
    6. Serabjit-Singh CJ,
    7. Adkison KK, and
    8. Polli JW
    (2002) Passive permeability and P-glycoprotein-mediated efflux differentiate central nervous system (CNS) and non-CNS marketed drugs. J Pharmacol Exp Ther 303:1029–1037.
    OpenUrlAbstract/FREE Full Text
    1. Melchior DL,
    2. Sharom FJ,
    3. Evers R,
    4. Wright GE,
    5. Chu JW,
    6. Wright SE,
    7. Chu X, and
    8. Yabut J
    (2012) Determining P-glycoprotein-drug interactions: evaluation of reconstituted P-glycoprotein in a liposomal system and LLC-MDR1 polarized cell monolayers. J Pharmacol Toxicol Methods 65:64–74.
    OpenUrlCrossRefPubMed
  27. ↵
    1. Moons T,
    2. de Roo M,
    3. Claes S, and
    4. Dom G
    (2011) Relationship between P-glycoprotein and second-generation antipsychotics. Pharmacogenomics 12:1193–1211.
    OpenUrlCrossRefPubMed
  28. ↵
    1. Polli JW,
    2. Wring SA,
    3. Humphreys JE,
    4. Huang L,
    5. Morgan JB,
    6. Webster LO, and
    7. Serabjit-Singh CS
    (2001) Rational use of in vitro P-glycoprotein assays in drug discovery. J Pharmacol Exp Ther 299:620–628.
    OpenUrlAbstract/FREE Full Text
  29. ↵
    1. RCoreTeam
    (2012) R: A language and environment for statistical computing, R Foundation for Statistical Computing, Vienna, Austria.
  30. ↵
    1. Reyner EL,
    2. Sevidal S,
    3. West MA,
    4. Clouser-Roche A,
    5. Freiwald S,
    6. Fenner K,
    7. Ullah M,
    8. Lee CA, and
    9. Smith BJ
    (2013) In vitro characterization of axitinib interactions with human efflux and hepatic uptake transporters: implications for disposition and drug interactions. Drug Metab Dispos 41:1575–1583.
    OpenUrlAbstract/FREE Full Text
  31. ↵
    1. Richards DB,
    2. Spence R,
    3. Mandagere A,
    4. Henderson LS, and
    5. Magee MH
    (2011) Effects of multiple doses of ambrisentan on the pharmacokinetics of a single dose of digoxin in healthy volunteers. J Clin Pharmacol 51:102–106.
    OpenUrlCrossRefPubMed
  32. ↵
    1. Robin X,
    2. Turck N,
    3. Hainard A,
    4. Tiberti N,
    5. Lisacek F,
    6. Sanchez JC, and
    7. Müller M
    (2011) pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 12:77–84.
    OpenUrlCrossRefPubMed
  33. ↵
    1. Schwab D,
    2. Fischer H,
    3. Tabatabaei A,
    4. Poli S, and
    5. Huwyler J
    (2003) Comparison of in vitro P-glycoprotein screening assays: recommendations for their use in drug discovery. J Med Chem 46:1716–1725.
    OpenUrlCrossRefPubMed
  34. ↵
    1. Seelig A
    (1998) A general pattern for substrate recognition by P-glycoprotein. Eur J Biochem 251:252–261.
    OpenUrlPubMed
  35. ↵
    1. Singh SK,
    2. Shawer M,
    3. Andrus PK,
    4. Raub TJ, and
    5. Vrbanac JJ
    (2003) Ocular delivery screening scheme for systemically administered ophthalmic drugs (Abstract). Invest Ophthalmol Vis Sci 44:4426.
    OpenUrl
  36. ↵
    1. Spahn-Langguth H and
    2. Langguth P
    (2001) Grapefruit juice enhances intestinal absorption of the P-glycoprotein substrate talinolol. Eur J Pharm Sci 12:361–367.
    OpenUrlCrossRefPubMed
  37. ↵
    1. Sugimoto H,
    2. Hirabayashi H,
    3. Amano N, and
    4. Moriwaki T
    (2013) Retrospective analysis of P-glycoprotein-mediated drug-drug interactions at the blood-brain barrier in humans. Drug Metab Dispos 41:683–688.
    OpenUrlAbstract/FREE Full Text
  38. ↵
    1. Sugimoto H,
    2. Hirabayashi H,
    3. Kimura Y,
    4. Furuta A,
    5. Amano N, and
    6. Moriwaki T
    (2011a) Quantitative investigation of the impact of P-glycoprotein inhibition on drug transport across blood-brain barrier in rats. Drug Metab Dispos 39:8–14.
    OpenUrlAbstract/FREE Full Text
  39. ↵
    1. Sugimoto H,
    2. Matsumoto S,
    3. Tachibana M,
    4. Niwa S,
    5. Hirabayashi H,
    6. Amano N, and
    7. Moriwaki T
    (2011b) Establishment of in vitro P-glycoprotein inhibition assay and its exclusion criteria to assess the risk of drug-drug interaction at the drug discovery stage. J Pharm Sci 100:4013–4023.
    OpenUrlCrossRefPubMed
  40. ↵
    1. Summerfield SG,
    2. Read K,
    3. Begley DJ,
    4. Obradovic T,
    5. Hidalgo IJ,
    6. Coggon S,
    7. Lewis AV,
    8. Porter RA, and
    9. Jeffrey P
    (2007) Central nervous system drug disposition: the relationship between in situ brain permeability and brain free fraction. J Pharmacol Exp Ther 322:205–213.
    OpenUrlAbstract/FREE Full Text
  41. ↵
    1. Tachibana T,
    2. Kato M,
    3. Watanabe T,
    4. Mitsui T, and
    5. Sugiyama Y
    (2009) Method for predicting the risk of drug-drug interactions involving inhibition of intestinal CYP3A4 and P-glycoprotein. Xenobiotica 39:430–443.
    OpenUrlCrossRefPubMed
  42. ↵
    1. Thiebaut F,
    2. Tsuruo T,
    3. Hamada H,
    4. Gottesman MM,
    5. Pastan I, and
    6. Willingham MC
    (1987) Cellular localization of the multidrug-resistance gene product P-glycoprotein in normal human tissues. Proc Natl Acad Sci USA 84:7735–7738.
    OpenUrlAbstract/FREE Full Text
  43. ↵
    1. Tweedie D,
    2. Polli JW,
    3. Berglund EG,
    4. Huang SM,
    5. Zhang L,
    6. Poirier A,
    7. Chu X,
    8. Feng B, and
    9. International Transporter Consortium
    (2013) Transporter studies in drug development: experience to date and follow-up on decision trees from the International Transporter Consortium. Clin Pharmacol Ther 94:113–125.
    OpenUrlCrossRefPubMed
  44. ↵
    1. Wandel C,
    2. Kim R,
    3. Wood M, and
    4. Wood A
    (2002) Interaction of morphine, fentanyl, sufentanil, alfentanil, and loperamide with the efflux drug transporter P-glycoprotein. Anesthesiology 96:913–920.
    OpenUrlCrossRefPubMed
  45. ↵
    1. Wittgen HG,
    2. Greupink R,
    3. van den Heuvel JJ,
    4. van den Broek PH,
    5. Dinter-Heidorn H,
    6. Koenderink JB, and
    7. Russel FG
    (2012) Exploiting transport activity of p-glycoprotein at the blood-brain barrier for the development of peripheral cannabinoid type 1 receptor antagonists. Mol Pharm 9:1351–1360.
    OpenUrlPubMed
  46. ↵
    1. Zhang L,
    2. Zhang YD,
    3. Strong JM,
    4. Reynolds KS, and
    5. Huang SM
    (2008) A regulatory viewpoint on transporter-based drug interactions. Xenobiotica 38:709–724.
    OpenUrlCrossRefPubMed
  47. ↵
    1. Zhang Y and
    2. Benet LZ
    (2001) The gut as a barrier to drug absorption: combined role of cytochrome P450 3A and P-glycoprotein. Clin Pharmacokinet 40:159–168.
    OpenUrlCrossRefPubMed
Back to top

In this issue

Download PDF
Article Alerts
Email Article
Citation Tools

Research ArticleArticle

In Vitro MDR1 Calibration

Agnès Poirier, Anne-Christine Cascais, Urs Bader, Renée Portmann, Marie-Elise Brun, Isabelle Walter, Alexander Hillebrecht, Mohammed Ullah and Christoph Funk
Drug Metabolism and Disposition September 1, 2014, 42 (9) 1411-1422; DOI: https://doi.org/10.1124/dmd.114.057943
Reddit logo Twitter logo Facebook logo Mendeley logo

Jump to section

Advertisement