In vitro inhibitory potency (Ki)-based predictions of P-glycoprotein (P-gp)-mediated drug-drug interactions (DDIs) are hampered by the substantial variability in inhibitory potency. In this study, in vivo-based [I]/Ki values were used to predict the DDI risks of a P-gp substrate dabigatran etexilate (DABE) using physiologically based pharmacokinetic (PBPK) modelling.
A baseline PBPK model was established with digoxin, a known P-gp substrate. The Km (P-gp transport) of digoxin in the baseline PBPK model was adjusted to Kmi to fit the change of digoxin pharmacokinetics in the presence of a P-gp inhibitor. Then ‘in vivo’ [I]/Ki of this P-gp inhibitor was calculated using Kmi/Km. Baseline PBPK model was developed for DABE, and the ‘in vivo’ [I]/Ki was incorporated into this model to simulate the static effect of P-gp inhibitor on DABE pharmacokinetics. This approach was verified by comparing the observed and the simulated DABE pharmacokinetics in the presence of five different P-gp inhibitors.
This approach accurately predicted the effects of five P-gp inhibitors on DABE pharmacokinetics (98–133% and 89–104% for the ratios of AUC and Cmax respectively). The effects of 16 other P-gp inhibitors on the pharmacokinetics of DABE were also confidently simulated.
Conclusions and implications
‘In vivo’ [I]/Ki and PBPK modelling, used in combination, can accurately predict P-gp-mediated DDIs. The described framework provides a mechanistic basis for the proper design of clinical DDI studies, as well as avoiding unnecessary clinical DDI studies.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
the area under concentration-time curve up to the last measured time point
the ratio of the area under the plasma concentration-time curve (AUC0→∞) in the presence and absence of the inhibitor
peak plasma concentration
the ratio of the peak concentration in the presence and absence of the inhibitor
the effective inhibitor concentration
organic anion-transporting polypeptide
physiologically based pharmacokinetic modelling
The transporter protein P-glycoprotein (P-gp) affects drug pharmacokinetics by limiting oral absorption, restricting CNS penetration and promoting excretion. Drug-drug interactions (DDIs) may occur when a P-gp substrate is co-administered with other drugs that inhibit P-gp. Accurate prediction of a potential risk of DDI is of great importance both within the pharmaceutical industry and in clinical practice in order to improve the safety and efficacy of new drugs.
Dabigatran etexilate (DABE) is a P-gp substrate and the first new oral anticoagulant to become available for the prevention of stroke and systemic embolism in patients with atrial fibrillation in over 50 years (Connolly et al., 2009). The orally administered prodrug DABE is absorbed in the intestine and then converted by carboxylesterase (CES) to the active moiety dabigatran (DAB) which is not a P-gp substrate) (Härtter et al., 2013; Hu et al., 2013). Drugs that inhibit P-gp, such as verapamil, amiodarone and quinidine, raise exposures to DAB (Härtter et al., 2013).
P-gp-mediated DDIs can be predicted by either a static approach or a dynamic (physiologically based pharmacokinetic, PBPK) approach. Although the static approach for the prediction of CYP450- and organic anion-transporting polypeptide (OATP)-mediated DDIs is quite successful (Hu et al., 2012; Hu, 2013), the same approach for the prediction of P-gp-mediated DDIs is only qualitative due to the high rates of false-negative (10–12%) and false-positive (40–51%) results (Fenner et al., 2009; Ellens et al., 2013). This is probably caused by the inaccurate estimation of [I] (the effective inhibitor concentration) and Ki (inhibition constant) in the prediction of P-gp-mediated DDIs. Studies using the dynamic approach are very limited. Recently, a PBPK approach for the prediction of DDI between the P-gp substrate digoxin and verapamil was introduced (Neuhoff et al., 2013b) and the predicted AUC ratio of digoxin following administration of verapamil was 1.20, while the observed value was 1.50. The verapamil Ki values considered in this study ranged from 0.1 to 16.3 (eight values from different references). The authors used the lowest Ki values for the worst-case scenario. Underestimation of the DDI will occur when other Ki values were considered. The success of this approach is dependent on the accuracy of the in vitro Ki used. Recently, a collaborative study was initiated between 23 pharmaceutical and contract research laboratories and one academic institution to compare in vitro IC50 data generated using four in vitro systems (Caco-2, MDCKII-MDR1, LLC-PK1-MDR1 cells and membrane vesicles) and six transport activity equations (Bentz et al., 2013). Unfortunately, the difference in IC50 values for each of the 16 inhibitors across all test systems ranged from a minimum of 20- to 24-fold between lowest and highest IC50 values for sertraline and isradipine, to a maximum of 407- and 796-fold for telmisartan and verapamil respectively. Further analysis showed that variability in IC50 values was mainly due to inter-laboratory variability, rather than the systematic difference between test systems. Such substantial variability will seriously decrease the accuracy of prediction of DDIs. For example, the in vitro IC50 value from some laboratories may predict significant DDI, whereas others may either over- or underestimate a DDI.
Therefore, the major challenge for the prediction of P-gp-mediated DDIs is how to identify or obtain the ‘true’ Ki value, which is a key input for both the static and dynamic approach. In order to overcome this difficulty, we have used the information from a typical P-gp-mediated DDI (between the P-gp substrate digoxin and a particular P-gp inhibitor) to calculate the ‘in vivo’ [I]/Ki for this P-gp inhibitor based on a PBPK approach. Then the ‘in vivo’ [I]/Ki was incorporated into the PBPK model of a test P-gp substrate (DABE) to simulate the DDI risk of DABE with this P-gp inhibitor. Our prediction method can be seen as a ‘static’ approach because the effective inhibitor concentration [I] in the inhibiting site is kept unchanged. This assumption is reasonable if the primary site of inhibition is in the intestine during oral absorption. We found that the elimination half-life and renal clearance of P-gp substrates such as digoxin and fexofenadine, were hardly altered by an inhibitor (Yasui-Furukori et al., 2005; Tateishi et al., 2008; Härtter et al., 2013), suggesting the DDIs occur mainly in the intestine during the absorption process. Indeed, there were only minimal DDI risks when intravenous digoxin was given together with P-gp inhibitors (Fenner et al., 2009).
The primary aim of this study was to develop an ‘in vivo’ [I]/Ki- and PBPK modelling-based approach for the quantitative prediction of P-gp-mediated DDIs with DABE. The approach was verified by the known effects of five different P-gp inhibitors on DABE pharmacokinetics. After that, the verified approach was used to predict the effects of 16 other P-gp inhibitors that had not been studied, on the pharmacokinetics of DABE.
Development of the PBPK model
A workflow describing the model building and validation processes that were applied to the development of the PBPK model is outlined in Figure 1. The details are described as follows:
(i) PBPK models for the orally administered P-gp substrates, digoxin and DABE (including its metabolite DAB), were constructed using in vitro and in silico data (Bottom-up), as well as the observations in a set of PK studies (Top-down). The model building strategy by Neuhoff et al. (2013a) was used here with some modifications.
(ii) The developed PBPK models were validated by a set of independent validation PK studies, including different doses, multiple doses, intravenous dosing or drug interaction data (P-gp-mediated induction). The major purpose of this step is to verify the contribution of passive and active efflux (P-gp) in the absorption of the studied P-gp substrates.
(iii) A reported DDI between the typical P-gp substrate (digoxin) and an inhibitor is used as the learning set. Assuming competitive inhibition, the Km value of digoxin in the baseline PBPK model was adjusted to Kmi to fit the change of digoxin AUC in the presence of a P-gp inhibitor. For competitive inhibition, intrinsic clearance of P-gp-mediated efflux (digoxin) in the presence of inhibition (CLinti) can be calculated using equation (1), which is under first order conditions (Rowland and Matin, 1973).
where Vmax is the maximum rate of reaction. Km, DI is the Michaelis–Menten constant for digoxin without inhibition. Km,DIi is the Michaelis–Menten constant for digoxin in the presence of inhibition. [I] is the effective inhibitor concentration. Ki is the inhibition constant. Equation (2) can be derived from equation (1).
Therefore, [I]/Ki of this particular P-gp inhibitor can be calculated using equation (2). It should be noted that [I]/Ki is purely derived from in vivo information and can been seen as the in vivo [I]/Ki (compared with the in vitro measured [I]/Ki).
A DDI between DABE and the same inhibitor is used as the test set. Kmi of DABE in the presence of the same P-gp inhibitor is calculated using equation (3). As shown in this equation, we need not know the actual Ki of an inhibitor to calculate the Kmi of DABE.
where k is the correction factor to compensate for the dose differences of this P-gp inhibitor between the learning set and test set (k = 1, if the same dose is used). Then the calculated Kmi was incorporated into the PBPK model of DABE to simulate the PK of DABE and DAB.
The approach was verified by the comparison of the observed and simulated PK in the presence of five different P-gp inhibitors. The verified approach was finally used to predict the effects of various P-gp inhibitors on the PK of DABE and DAB.
Model structure and parameterization
The PBPK models for digoxin, DABE (incorporating its metabolite DAB) were built using the commercial software tool, PK-Sim® (version 5.1; Bayer Technology Services GmbH, Leverkusen, Germany), which implements a whole-body PBPK model consisting of 15 organs. The PBPK models of DABE and DAB were exported to MoBi (version 3.1; Bayer Technology Services GmbH), where they were linked to build the coupled parent-metabolite model.
All PBPK models described in the following assume an average 30-year-old white man with a body weight of 78 kg and a height of 177 cm. Physiological parameters describing basic model structure such as organ volumes, organ weights, tissue composition, blood flow rates or gastrointestinal physiology are provided within the software. The key measured and predicted input data for digoxin, DABE and DAB are listed in Table 1.
Table 1. In silico, in vitro and in vivo data for parameterization of the physiologically based pharmacokinetic (PBPK) models for digoxin, dabigatran etexilate (DABE) and its active metabolite dabigatran (DAB)
If not indicated, the input data were calculated by PK-Sim or was the default value in PK-Sim. The refined values were obtained by fitting the simulated plasma concentration-time profiles with the observed ones. Vmax, maximum rate of reaction; Km, Michaelis–Menten constant; Koff, dissociation rate constant; Kd, dissociation constant; NA, not applicable. The unit for the specific Vmax used in the simulation is μmol min−1 L−1 (referring to the cellular volume).
bAccording to the study by Neuhoff et al. (2013a), the formulations currently on the market sufficiently increase the solubility.
cUnchanged dabigatran in urine accounted for about 80% of the i.v. dose of dabigatran (Blech et al., 2008). Renal clearance was calculated by multiplying the total intravenous clearance (155 mL min−1) by 80%. The remaining 20% of total clearance was attributed to hepatic clearance.
dFrom DABE enzyme kinetic studies in human liver S9 (Laizure et al., 2013).
A detailed model for GI transit and absorption was used (Thelen et al., 2011; 2012). In this model, the alimentary canal from the stomach to the rectum is divided into 12 compartments, each representing a definite segment of the GI tract. Intestinal permeability was calculated from the drug's lipophilicity and effective molecular weight.
The PK-Sim standard distribution model was used, which assumes four subcompartments per organ, for example, compartments for blood cells, plasma, interstitial space and cellular space. This model type considers a permeation barrier between blood and organ tissue. The rates of permeation across the cell membranes (interstitial-cell barrier) depend on the drug-specific cellular membrane permeability and the surface area available of each organ. The cellular membrane permeability values were calculated using the ‘PK-Sim Standard’ method. Tissue/plasma partition coefficients were generated using the model of Rodgers and Rowland (Rodgers et al., 2005; Rodgers and Rowland, 2006; 2007).
Disposition characteristics of digoxin and DABE in humans
Digoxin is a P-gp substrate without the confounding effects of metabolism. P-gp plays a predominant role in the absorption and elimination of digoxin in vivo (Fenner et al., 2009). Currently, there is no published evidence showing that digoxin is transported by any other drug transporter. The transportation of digoxin by P-gp was modelled based on Michaelis–Menten kinetics. The relative gene expression data of P-gp was integrated in the PBPK models (Meyer et al., 2012). According to this expression data, P-gp was expressed extensively in the body including intestine, kidney and liver. Therefore, the involvement of P-gp in the hepatic and renal clearance was also considered. Renal excretion (renal clearance) was considered as the only excretion route for the elimination of digoxin in humans (Neuhoff et al., 2013a). As with methotrexate (Bischoff et al., 1971) and docetaxel (Bradshaw-Pierce et al., 2007), specific binding to tissue components was included. This was justified by the finding that receptor (Na+/K+-ATPase) binding was the rate-limiting step for the distribution of digoxin (Weiss, 2007). Plasma levels of digoxin were influenced by the changes in capacity and/or affinity of digoxin binding to sodium pumps in vivo (Weiss, 2007). Two parameters, Koff (dissociation rate constant, 1 min−1) and Kd (dissociation constant, μmol L−1) were introduced to express the nonlinear binding within tissue. The addition of a non-linear binding term not only makes the model more relevant to the biochemical process in the body, but also resulted in better prediction.
The oral prodrug DABE, which is a P-gp substrate, was converted by CES enzymes to the active moiety DAB, which is not a P-gp substrate (Hu et al., 2013). DAB was formed primarily in the liver by CES1 (Laizure et al., 2013). Therefore, only hepatic CES1 was assumed to mediate the formation of DAB during first-pass metabolism. The relative gene expression data of CES1 is used in the PBPK models. The conversion of DABE to DAB by CES1 and the efflux of DABE by P-gp were both modelled using Michaelis–Menten kinetics. Renal excretion (renal clearance) and hepatic metabolism (hepatic clearance) were considered as the elimination pathways of DAB from the body (Blech et al., 2008).
PK data analysis
PK parameters were determined by a non-compartmental method using the Kinetica 5.0 (Thermo Fisher Scientific; Philadelphia, PA, USA). The area under the concentration-time curve up to the last measured time point (AUC0→t) was calculated using the trapezoidal method. The AUC from time zero to infinity (AUC0→∞) was generated by extrapolating the AUC0→t to infinity using the elimination rate constant and the last measured concentration. All the human in vivo data were obtained from published data by digitalization with data mining software (GetData Graph Digitizer version 2.25, http://www.getdata-graph-digitizer.com).
Simulation of DABE and DAB pharmacokinetics in humans
The predicted or assumed input data (Table 1) were refined to match the simulated plasma concentration-time profiles of DABE and DAB with the observed data (Stangier, 2008; Stangier et al., 2008; Härtter et al., 2012; 2013) (Figure 2A, single oral dosing of 150 mg DABE). As shown in Figure 2B and D, the model was verified by the comparison of the simulated and observed data of multiple dosing (twice daily of 150 mg DABE) and dose-exposure relationships (Stangier et al., 2007; 2012). For the dose-exposure relationships, it should be noted that the observed exposures to DAB was higher than the simulated data. This is possibly due to the different formulations that were utilized for the observed data (solution) and simulated data (capsule). In addition, there was a good agreement between the simulated [increased baseline Vmax by 3.5 times based on the induction of P-gp expression level by rifampicin (Greiner et al., 1999)] and observed exposures to DAB in the presence of the P-gp inducer rifampicin (Figure 2C). The observed ratio of DAB AUC0→∞ (AUCR) and Cmax (Cmax, R) in the presence and absence of the inducer was 0.33 and 0.34 respectively (Härtter et al., 2012). The predicted values were 0.29 and 0.28 respectively.
Simulation of digoxin pharmacokinetics in humans
The predicted or assumed input data (Table 1) were refined to match the simulated plasma concentration-time profile of digoxin with the observed data (Ragueneau et al., 1999; Westphal et al., 2000; Becquemont et al., 2001; Verstuyft et al., 2003) (Figure 3A, single oral dosing of 0.5 mg digoxin). Then the developed PBPK model was validated by the comparison of the simulated and observed data of other dose levels (single oral dosing of 0.25 mg and 1.0 mg) (Oosterhuis et al., 1991; Greiner et al., 1999; Eckermann et al., 2012) (Figure 3B and C). Good agreement was also observed between simulated and observed data from multiple dosing regimens (oral 0.25 mg once daily and twice daily) (Rodin et al., 1988; Vaidyanathan et al., 2008) (Figure 3D and E). With regard to the intravenous dosing, the simulated data and the observed data are very close (Kramer et al., 1979; Ding et al., 2004) (Figure 3F and G).
Calculation of the ‘in vivo’ [I]/Ki
The calculated ‘in vivo’ [I]/Ki (calculated using equation (3)) of various P-gp inhibitors are shown in Supporting Information Table S1. The values of these ‘in vivo’ [I]/Ki were dependent on the doses of the inhibitors. Kmi values of DABE (Km,DABEi, calculated using equation (3)) in the presence of various P-gp inhibitors are also shown in Supporting Information Table S1.
Comparison of the observed and simulated exposure to DAB in the presence of five P-gp inhibitors
As shown in Figure 4, there was a good agreement between the predicted and observed plasma concentration-time profile of DAB in the presence of verapamil. Positive DDIs between five P-gp inhibitors (amiodarone, clarithromycin, dronedarone, quinidine, verapamil) and DABE were reported. Our new ‘in vivo’ Ki-based PBPK approach successfully predicted these DDI risks. The predictive accuracy for AUCR and Cmax, R ranged from 98–133% to 89–104% respectively (Table 2).
Table 2. Observed and simulated exposures to active metabolite dabigatran (DAB) in humans after co-administration of dabigatran etexilate (DABE, single oral dosing of 150 mg) and various P-gp inhibitors
Inhibitor dosing schemes
Observed Cmax, R
Simulated Cmax, R
Accuracy (Cmax, R)
AUCR, the ratio of the area under the plasma concentration-time curve (AUC0→∞) in the presence and absence of the inhibitor; Cmax, R, the ratio of the peak concentration in the presence and absence of the inhibitor. qd, once daily; bid, twice daily, tid, three times a day; single, single dosing; multiple, multiple dosing; NA, not applicable.
aData for the female subjects;b data for the male subjects.
Predicting the DDI risks between various P-gp inhibitors and DABE
The potential DDI risks between all the other reported P-gp inhibitors and DABE were predicted (Table 2). AUCR and Cmax, R of DAB ranged from 1.19–1.96 to 1.13–1.57 respectively (Table 2).
The major finding of the present study is that ‘in vivo’ [I]/Ki can be used for the quantitative prediction of P-gp-mediated DDIs. Accurate prediction of P-gp-mediated DDIs (involving DABE) was achieved when the ‘in vivo’ [I]/Ki was used in combination with the PBPK modelling. The other finding is that exposure to DAB in humans is sensitive to P-gp inhibition. Identification of this class of DDIs is essential for the improvement in benefit-risk data for DABE, given the post-marketing reports of bleeding with the use of DABE (Southworth et al., 2013).
Currently, only one PBPK approach for the prediction of DDI between P-gp substrate (digoxin) and inhibitor (verapamil) has been reported (Neuhoff et al., 2013b). Our approach is simpler than the approach by Neuhoff et al. (2013b), which involved complex, dynamic DDI simulations, and metabolite inhibition potential was dynamically considered. Our prediction method can be seen as a special ‘static’ approach because the effective inhibitor concentration [I] is kept unchanged. However, unlike the usual static approach, our special ‘static’ method is still able to simulate the complete plasma concentration-time profile of the substrate drug in the presence of an inhibitor.
Simplicity has been the biggest challenge for the PBPK community to face: can such predictions be made without going through complex model building? In this study, we took a hybrid approach using substrate PBPK model and in vivo DDI data to define the [I]/Ki term for each inhibitor. There is, however, a simplified form of our approach. We can simply use the equation AUCR = 1 + [I]/Ki to calculate [I]/Ki, and then adjust [I]/Ki to compensate for the dose differences of this P-gp inhibitor before predicting the AUCR for DAB with the same equation. However, there is an important assumption underlying this simplified approach. For example, the relative contribution of P-gp-mediated clearance to total clearance is the same for the substrates in the learning set (digoxin) and the test drug (DABE). Therefore, our simplified approach can be used only if this assumption (about the contribution of P-gp to total clearance) is valid. If the assumption is known not to be valid or if we do not know if it is valid or not, then our hybrid approach can be used.
Although this approach accurately predicts the influence of five P-gp inhibitors on DABE pharmacokinetics, it will be of interest to extend the approach to the prediction of DDIs between other P-gp substrates and inhibitors. Fexofenadine is a typical P-gp substrate (Draft Guidance/Guidance for Industry 2012). There are four P-gp inhibitors (from 21 inhibitors in Table 2) that were reported to be able to interact with fexofenadine, for example, itraconazole, verapamil, ritonavir and lopinavir/ritonavir (Yasui-Furukori et al., 2005; Shimizu et al., 2006; van Heeswijk et al., 2006). However, fexofenadine is also an OATP substrate. The P-gp inhibitors verapamil, ritonavir and lopinavir/ritonavir are also able to inhibit OATP transporters (Cvetkovic et al., 1999). Therefore, only the DDI involving fexofenadine and itraconazole can be used to test our approach. The reported Cmax, R and AUCR of fexofenadine were 1.8 and 1.7 respectively (Shimizu et al., 2006). The predicted values were 1.4 and 2.2 respectively. This result suggests that our approach can be extended to the prediction of the DDIs between other P-gp substrates and inhibitors.
Regulatory authorities are increasingly focusing on the use of PBPK modelling during regulatory review (Zhao et al., 2011). Our newly developed approach was utilized to predict, with confidence, a large number of DDIs between DABE and various P-gp inhibitors that have not been studied. Based on these prediction results (Table 2), the potentially unfavourable exposure of real patients can be minimized, unnecessary clinical studies can be avoided and the study design of the clinical study can be optimized. Although [I]/Ki derived from this work may be platform-specific and would require re-parameterization when other platforms were to be used, the data of ‘in vivo’ [I]/Ki (Table 2) can serve as an important database for other researchers. According to a very recent study, the ‘in vivo’ inhibition potency of probenecid on renal secretion was derived using PBPK modelling (Hsu et al., 2013). The authors suggested that this in vivo data can be used to provide an initial estimate of AUCR values in the presence of probenecid.
In order to improve decision making in drug development and discovery, our approach can be used during the course of new drug development and discovery. If a new molecular entity is a substrate for P-gp and the PBPK model of this drug has already been established, the DDI between this drug and various P-gp inhibitors can be predicted. Our approach can reasonably exclude clinical DDI studies that later turn out to be unnecessary.
However there are limitations to the current study that need to be considered. The major challenge for the application of our approach is the accurate definition of the contribution of the P-gp transporter (to total PK) when a PBPK model for new molecular entity is being constructed. As in vivo clinical data were used to estimate the [I]/Ki value which was subsequently used for DAB, extrapolation outside the studied population and the conditions of the learning data set could sometimes not be justifiable. Care should be taken as the estimated parameter value might not hold in another population, with different ethnicities, P-gp genotypes or disease conditions.
In conclusion, ‘in vivo’ [I]/Ki and PBPK modelling can be used in combination for the quantitative prediction of P-gp-mediated DDIs. To our knowledge, this is the first proof-of-concept study demonstrating that ‘in vivo’ [I]/Ki is a useful parameter for the prediction of DDIs associated with P-gp substrates. The described general framework can be used in the course of new drug development to provide a mechanistic basis for proper design of clinical DDI studies, as well as avoiding unnecessary clinical DDI studies.
This study was not supported by any grant.
Conflict of interest
The authors declare no actual or potential conflicts of interest.