FormalPara Key Points

A PBPK model was developed, accurately predicting the in vivo pharmacokinetics of both entrectinib and its active metabolite M5.

Dosing recommendations for co-administering entrectinib with CYP3A4 inhibitors derived from the model were 6-fold lower and 3-fold lower entrectinib doses when co-administered with a strong and a moderate CYP3A4 inhibitor, respectively, and the use of entrectinib with moderate or strong CYP3A4 inducers should be avoided.

This PBPK modeling approach provided key support for the filing of entrectinib and dosing recommendations in the drug label.

1 Introduction

The use of physiologically based pharmacokinetic (PBPK) modeling to predict drug concentrations in plasma and tissue has demonstrated utility for accelerating pharmaceutical development, and is now an integral part of many drug development programs [1,2,3,4,5,6,7,8]. Advancement in the discipline has been followed by increasing acceptance by regulatory authorities [5, 7,8,9,10,11,12,13], and there are now numerous examples of drug approvals supported by PBPK modeling in lieu of in vivo clinical studies [14, 15, 17, 18]. One area where PBPK modeling is now particularly widely used is the prediction of drug–drug interactions, in part because it allows quantitative predictions in complex scenarios, e.g., simultaneous induction and inhibition, multiple perpetrators, etc. [19,20,21,22]. Understanding the clinical consequences of such interactions has been further facilitated by the development of models which simultaneously predict effects on multiple pharmacologically-active species [23,24,25,26,27,28].

PBPK modeling has traditionally been regarded as a bottom–up approach whereby in vitro–in vivo extrapolation (IVIVE) techniques are used within a mechanistic framework to predict plasma and tissue concentrations from physicochemical and in vitro data. This contrasts with a top–down approach whereby empirical models are generated to describe observed in vivo data. However, both approaches are recognized to have limitations, and use of a middle–out strategy combining elements of both represents an alternative [7, 29,30,31,32]. Such hybrid multilevel models combine prior information on the system and drug with analysis of observed data, for example, by using clinical data to optimize IVIVE model parameters. Generation of a middle–out model offers potential advantages, but is not without challenges [30]. While there are now numerous examples where a middle–out modeling strategy has been used to support high-impact regulatory activities, (e.g., drug–drug interaction dosing recommendations without the need for a corresponding clinical study [12]), success of this approach ultimately requires acceptance and endorsement by regulatory authorities.

Entrectinib is a potent and selective inhibitor of pan-TRK, ROS1, and ALK receptor tyrosine kinases. These kinases are overexpressed or dysregulated in many types of cancer, such that cancer cell growth is dependent on abnormal kinase activity [33]. Recently, entrectinib was approved for treatment of adult and pediatric patients with tumors that harbor NTRK1/2/3 or ROS1 gene rearrangements.

Entrectinib is predominantly cleared by CYP3A4-mediated metabolism to a pharmacologically-active metabolite (M5), and both parent and metabolite are believed to contribute equally to the overall effect of entrectinib treatment [34]. Clinical drug–drug interaction studies with the potent CYP3A4 inhibitor, itraconazole, and CYP3A inducer, rifampicin, demonstrated that modulation of CYP3A enzyme activity profoundly alters the pharmacokinetics of both entrectinib and M5. However, the effects on entrectinib and M5 were quantitatively different, making it more difficult to extrapolate the observed itraconazole and rifampicin drug–drug interaction study data to other scenarios.

Here, we describe the development of a PBPK model for entrectinib and M5 where two independent methods [sensitivity analysis and nonlinear mixed effect (NLME) modeling] were used to refine estimates for key entrectinib and M5 clearance parameters. The final model was then used to define appropriate entrectinib dosing strategies with various different CYP3A4 inhibitors and inducers in order to support regulatory decision-making.

2 Methods

The overall PBPK model development, qualification, and simulation strategy is depicted schematically in Fig. 1. The initial model building focused on entrectinib only; subsequently,, M5 was incorporated. The PBPK model based on IVIVE of physiochemical, in vitro, and in vivo metabolism data was refined using sensitivity analysis and NLME modeling to optimize parameter estimates and improve model fit to data from a clinical drug–drug interaction study with the strong CYP3A4 inhibitor, itraconazole. The refined model was subsequently qualified by comparing simulated entrectinib and M5 plasma concentrations with observed data from other clinical studies in which patients with solid tumors or healthy volunteers received entrectinib dosing, including a clinical drug–drug interaction study with the strong CYP3A4 inducer, rifampin. Thereafter, the final qualified PBPK model was used to simulate the effects of several moderate-to-strong CYP3A4 inhibitors and inducers on entrectinib and M5 pharmacokinetics. The Simcyp input parameters for the final PBPK model are detailed in Table 1.

Fig. 1
figure 1

Schematic of PBPK model development, qualification, and simulation strategy. CYP cytochrome P450, DDI drug-drug interaction, NLME non-linear mixed effect, PBPK physiologically based pharmacokinetic  

Full size image
Table 1 Simcyp input parameters for final PBPK model
Full size table

2.1 Model Development and Qualification

The PBPK model was established in Simcyp® Simulator (v.17.1; Certara, Princeton, USA). The model integrated available physiochemical, in vitro, and in vivo metabolism data for entrectinib and M5 (Table 1). The retrograde modeling tool was used to refine the intrinsic clearance (CLint) values obtained via in vitro to in vivo extrapolation, and a full PBPK distribution model was used with the Rodgers and Rowland method to predict tissue to plasma partition coefficients [35]. Based on insights derived from independent modeling of entrectinib absorption using the GastroPlus software tool [36], an advanced distribution, absorption, and metabolism model [37] was used to describe the kinetics in the gastrointestinal tract using the “solution formulation without precipitation” option. A built-in virtual healthy volunteer adult population was used for simulations. For simulation of dosing in the fed state, the Simcyp default gastric emptying time (1 h) was increased to 2 h to better reflect the observed timing of peak entrectinib concentrations.

2.1.1 Sensitivity Analyses

Sensitivity analyses were performed to optimize the fraction of entrectinib unbound in gut (fuGut), the fraction of entrectinib metabolized by CYP3A4 (entrectinib fmCYP3A4), the fraction of entrectinib metabolized by CYP3A4 to M5 (entrectinib fmCYP3A4[M5]), the metabolic clearance of M5, the fraction of M5 metabolized (M5 fm), and the fraction of M5 metabolized by CYP3A4 (M5 fmCYP3A4).

A matrix of 20 different values of fuGut and entrectinib fmCYP3A4 parameters were initially assessed during parent model development (fuGut: 0.5–1; entrectinib fmCYP3A4 0.737–0.831, based on the estimated 0.72 from in vitro, represented by CYP3A4 CLint of 4–7 μL/min/pmol). Predicted entrectinib exposures from each pair of parameter values were visually compared against observed data from the clinical drug–drug interaction study with itraconazole [38] (Study RXDX-101-12). Based on this initial assessment, it was determined to be most relevant to fix fuGut at 1, and a more intensive sensitivity analysis was conducted that focused on the entrectinib fmCYP3A4 over a range of 0.778–0.808 (CLint ranging between 5 and 6 μL/min/pmol).

Based on in vitro experiments [38], M5 was identified as the major metabolite, and that it was mainly formed via CYP3A4 (> 50%). In addition, CYP3A4 was the main isoform involved in M5 metabolism (70–99%). As such, during M5 model development, sensitivity analyses were performed that explored parameter value ranges for entrectinib (fmCYP3A4[M5] and M5 fm of 50–99%, and 70–99% for M5 fmCYP3A4). Metabolic clearance of M5 was explored over a 0.5- to 2-fold range relative to the metabolic clearance of entrectinib. In each case, the predicted M5 exposures from each parameter value were visually compared against observed data from a clinical drug–drug interaction study with itraconazole (Study RXDX-101-12). There was no hierarchy among the sensitivity analyses.

2.1.2 NLME Modeling to Estimate Fg and fmCYP3A4

A novel data analysis approach was also used to estimate entrectinib fraction escaping intestinal metabolism (Fg) and fmCYP3A4 parameters. A combination of NLME and PBPK modeling was used to analyze data from the clinical drug–drug interaction study with itraconazole (Study RXDX-101-12). A description of the assumptions, methodology, results, and model verification is presented elsewhere [15].

2.2 Clinical Study Data

Model qualification employed plasma concentration data from three clinical studies in which patients or healthy volunteers received entrectinib dosing (Table 2). In each study, bioanalytical samples were collected according to an intensive sampling scheme; entrectinib and M5 plasma concentrations were measured using a validated LC-MS/MS method for simultaneous determination of entrectinib and M5 (Ignyta, San Diego, CA, USA; data on file). Two different oral immediate release capsule formulations (F2A and F06) were employed, but were not differentiated during model development because the two formulations were bioequivalent [16].

Table 2 Summary of clinical studies providing data for model qualifications
Full size table

2.3 Simulations with CYP3A4 Inhibitors and Inducers

The final qualified PBPK model was used to simulate the effects of moderate to strong CYP3A4 inhibitors and inducers on the pharmacokinetics of entrectinib and M5 in a virtual population of adult healthy volunteers. The perpetrator drugs and their simulated dosing regimens are detailed in Table 3. Simulations employed compound files from Simcyp (v.17.1). Simcyp parameter values for creating a virtual healthy volunteer population (e.g., physiological parameters including liver volume and blood flows, enzyme abundances) have been described previously [40]. While initial simulations employed a single 600-mg dose of entrectinib, subsequent simulations for selected perpetrators were generated using lower doses of entrectinib (100 mg and 200 mg) and dosing to steady state with a once-daily dosing regimen.

Table 3 Summary of simulated perpetrator dosing regimens
Full size table

3 Results

3.1 Model Qualification

Input parameters for the final PBPK model are detailed in Table 1. Refinements to initial parameter estimates based on sensitivity analysis and NLME modeling are indicated. The two independent methods used to derive estimates for key entrectinib and M5 clearance parameters gave similar results. Based on sensitivity analyses, an entrectinib fmCYP3A4 value of 0.78 (reflected by a CYP3A4 CLint of 5.17 μL/min/pmol) was selected. A fuGut of 1 resulted in a mean estimated Fg (i.e., fraction of entrectinib escaping intestinal metabolism) of 0.60 (geometric mean 0.58). Separately, the NLME estimated an fmCYP3A4 of 0.755 (95% CI 0.697–0.804), and Fg of 0.58 (95%CI 0.460–0.718) [15]. Overall, the model predicts that the majority of an absorbed entrectinib dose is cleared by CYP3A4-mediated metabolism to the M5 metabolite, while the M5 metabolite is itself almost exclusively cleared by CYP3A4-mediated metabolism (Fig. 2).

Fig. 2
figure 2

Schematic of entrectinib and M5 clearance pathways with key parameter values. CYP cytochrome P450, UGT uridine 5'-diphospho-glucuronosyltransferase

Full size image

During model development, it was noted that the observed entrectinib and M5 exposures in the clinical drug–drug interaction study with the strong CYP3A4 inducer rifampin (Part 2 of Study RXDX-101-12) were ~ 30% lower than other clinical studies employing the same entrectinib dose. As a consequence, the PBPK model initially over-estimated exposure parameters for this specific study. To improve the PBPK model fit, a study-specific lower bioavailability was incorporated by reducing the effective permeability in human value for entrectinib (Peff,man) from 1.34 × 10−4 to 0.33 × 10−4 cm/s for this study.

The final PBPK model showed good predictive performance for both entrectinib and M5. Predicted plasma exposures were similar to observed exposures when entrectinib was administered alone, or with the strong CYP3A4 inhibitor, itraconazole (Study RXDX-101-12 Part 1; Table 4; Fig. 3), or with the strong CYP3A inducer, rifampicin (Study RXDX-101-12 Part 2; Table 5 and Fig. 4). The 5th and 95th percentiles of the model-predicted concentrations encompassed most observed concentrations (Figs. 3, 4), while the magnitude of the drug–drug interaction effects predicted by the model were comparable with the observed results from NCA analyses. The ratios between predicted and observed changes in drug exposure (Ratiopredicted/Ratioobserved, see [11]) for co-administration of itraconazole were 1.14 (Cmax) and 0.76 (AUC) for entrectinib, and 0.33 (Cmax) and 0.52 (AUC) for M5 (Table 4). Corresponding values for co-administration of rifampicin were 0.82 (Cmax) and 0.87 (AUC) for entrectinib, and 1.00 (Cmax) and 1.36 (AUC) for M5 (Table 5). Predictive performance was also good when simulating exposures in healthy volunteers under fed and fasted conditions (Study RXDX-101-04; presented in Fig. 5), and in patients with solid tumors dosed to steady state (Study RXDX-101-02) (Figs. 6 and 7 for entrectinib and M5, respectively).

Table 4 Observed and simulated entrectinib and M5 exposure with and without co-administration of the potent CYP3A inhibitor itraconazole
Full size table
Fig. 3
figure 3

Simulated and observed entrectinib (a, b) and M5 (c, d) plasma concentrations following a single 100-mg dose of entrectinib alone (a, c) or co-administered with the CYP3A4 inhibitor, itraconazole (b, d). Black and gray lines median model-predicted concentrations with 5th and 95th percentiles; circles observed individual concentrations from Study RXDX-101-12 Part 1. Main panel linear Y-axis; inset log scale Y-axis. CYP cytochrome P450

Full size image
Table 5 Observed and simulated entrectinib and M5 exposure with and without co-administration of the potent CYP3A inhibitor rifampicin
Full size table
Fig. 4
figure 4

Simulated and observed Entrectinib (a, b) and M5 (c, d) plasma concentrations following a single 600-mg dose of entrectinib alone (a, c) or co-administered with the CYP3A inducer, ifampin (b, d). Black and gray lines median model-predicted concentrations with 5th and 95th percentiles; circles observed individual concentrations from Study RXDX-101-12 Part 2. Main panel linear Y-axis; inset log scale Y-axis. CYP cytochrome P450

Full size image
Fig. 5
figure 5

Simulated and observed entrectinib plasma concentrations following administration of entrectinib under fasted 400 mg (a) and 600 mg (b) or fed 600 mg (c) conditions. Black and gray lines median model-predicted concentrations with 5th and 95th percentiles; circles observed individual concentrations from Study RXDX-101-04. Main panel linear Y-axis; inset log scale Y-axis

Full size image
Fig. 6
figure 6

Simulated and observed entrectinib plasma concentrations in adults with NTRK1/2/3+ or ROS1+ solid tumors after a single dose of entrectinib (a) and at steady state (b). Black and gray lines median model-predicted concentrations with 5th and 95th percentiles; circles observed individual concentrations from Study RXDX-101-03. Main panel linear Y-axis; inset log scale Y-axis

Full size image
Fig. 7
figure 7

Simulated and observed M5 plasma concentrations in adults with NTRK1/2/3+ or ROS1+ solid tumors after a single dose of entrectinib (a) and at steady state (b). Black and gray lines median model-predicted concentrations with 5th and 95th percentiles; circles observed individual concentrations from Study RXDX-101-02. Main panel linear Y-axis; inset log scale Y-axis

Full size image

3.2 Simulations with CYP3A4 Perpetrators

The final qualified PBPK model was used to simulate the effects of various moderate to strong CYP3A4 inhibitors and inducers on entrectinib and M5 pharmacokinetics in a virtual population of adult healthy volunteers. Ratios of simulated entrectinib AUCinf values in the presence and absence of the perpetrator, and corresponding 95% confidence intervals, are summarized in Fig. 8. The model predicted that co-administration of various moderate CYP3A4 inhibitors (verapamil, erythromycin, clarithromycin, fluconazole, and diltiazem) would result in average increases in entrectinib exposure between 2.2- and 3.1-fold (Fig. 8a). Corresponding average increases for M5 were predicted to be approximately 2-fold (Fig. 8b). The model predicted that co-administration of various moderate CYP3A4 inducers (efavirenz, carbamazepine, phenytoin) would result in average decrease in entrectinib exposure between 45 and 79% (Fig. 8a), while corresponding average decreases for M5 were predicted to be approximately 50% (Fig. 5b). Simulations of repeat dosing with entrectinib produced predicted interactions of similar magnitudes. For example, median AUC interaction ratios after a single dose of entrectinib co-administered with itraconazole were 4.58 and 1.40 for entrectinib and M5, respectively, while the corresponding values from repeat dosing to steady state were 5.06 and 1.86, respectively (data not shown).

Fig. 8
figure 8

Proportional changes in simulated entrectinib (a) and M5 (b) AUCinf exposure parameters from co-administration of various CYP3A4 inhibitors and inducers. Symbols geometric mean ratios, error bars upper and lower 95% confidence intervals. AUCinf area under the concentration-time curve from time zero to infinity, CYP cytochrome P450, sim simulated, obs observed, PK pharmacokinetic

Full size image

Based on the magnitude of the simulated interactions, 3-fold and 6-fold lower entrectinib doses (i.e., 200 mg and 100 mg) are required to mitigate the effects of moderate and strong CYP3A4 inhibitors, respectively. To confirm the appropriateness of these dose adjustments, 100 mg and 200 mg entrectinib co-administered with strong and moderate CYP3A4 inhibitors, respectively, were also simulated (Table 6). These confirmed that simulated entrectinib and M5 exposures using the recommended dose adjustments were comparable to those from dosing with 600 mg entrectinib alone.

Table 6 Predicted mean entrectinib and M5 parameters from co-administration of moderate and strong CYP3A4 inhibitors with 100 mg or 200 mg entrectinib
Full size table

4 Discussion

A PBPK model of entrectinib and its active metabolite M5 was developed by integrating in vitro, non-clinical, and clinical data. The PBPK model based on IVIVE was refined using a sensitivity analysis and NLME modeling (described in detail elsewhere [15]) to optimize parameter estimates of the fraction metabolized by CYP3A4 and the fraction escaping gut metabolism. The two separate approaches were used in parallel, and both gave very similar parameter estimates (Fg: 0.6 vs. 0.58; fmCYP3A4: 0.78 vs. 0.75). As well as demonstrating the utility of a NLME modeling approach as a tool to refine key parameter estimates, concordance increased confidence in the two key determinant parameters of the pharmacokinetics and drug–drug interaction liability of entrectinib. Parameter estimates were further corroborated by independent data from a human ADME study in which entrectinib disposition in humans in vivo was investigated by the administration of a single dose of radiolabeled entrectinib to healthy volunteers (unpublished data). Based on the radiolabel recovered in excreta, it was estimated that, on average, up to 73% of the administered entrectinib dose was cleared by metabolism to M5, while the corresponding parameter in the final PBPK model was 70% (Fig. 2). The consistency with a completely independent clinical data source thereby provides additional confidence in the robustness of the PBPK model.

The final PBPK model showed good predictive performance for both entrectinib and M5, and met commonly-used predictive performance acceptance criteria when compared with observed clinical data [11, 12, 39]. Considering the drug–drug interactions with itraconazole and rifampicin, the ratios of predicted AUCs were all within 2-fold of the observed ratio (i.e. calculated Ratiopredicted/Ratioobserved > 0.5 and < 2.0), while, in many cases, the ratios of AUC and Cmax were within 25% of the observed ratio (i.e., calculated Ratiopredicted/Ratioobserved > 0.8 and < 1.25). It is notable that predictions of the effect of itraconazole on M5 were less accurate, underpredicting the magnitude of the effect on AUC while overpredicting the effect on Cmax. While this suggests that there is still potential to improve this aspect of the model, it was not considered to compromise the value of the model for supporting dosing recommendations.

The PBPK model, which describes both entrectinib parent and M5 metabolite pharmacokinetics, has particular utility since M5 is pharmacologically active, and consequently both parent and metabolite are believed to contribute to the overall efficacy of entrectinib treatment. Therefore, the model provides a useful quantitative tool with which to evaluate alternative dosing strategies under circumstances where the pharmacokinetics of both entrectinib and M5 are altered. However, the dosing recommendations for co-administering entrectinib with CYP3A4 inhibitors or inducers focus principally on entrectinib exposure, primarily because entrectinib is the principal circulating species in vivo (M5 plasma exposures are typically ≤½ those of entrectinib under normal dosing conditions). Consequently, as metabolite exposures are well below those of the parent, it is expected that M5 makes a smaller contribution than entrectinib to the pharmacological effects of entrectinib treatment. This is supported by analyses of the exposure versus response relationships, which showed that using parameters representing the sum of entrectinib and M5 exposures together yielded no additional insight over use of entrectinib exposure alone [41]. Furthermore, the concurrent use of CYP3A inhibitors and inducers with entrectinib both lead to a decrease in the metabolite:parent ratio. While itraconazole use increases both observed entrectinib and M5 exposure, the proportional change is greater for entrectinib (approximately 5.8-fold) than M5 (approximately 2.6-fold), and the average metabolite:parent ratio decreases. Conversely, rifampicin use decreases both entrectinib and M5 exposure, the proportional change is smaller for entrectinib (approximately 66%) than M5 (approximately 92%), and the average metabolite:parent ratio decreases. Therefore, in each scenario, the contribution of the M5 metabolite to the pharmacological effect of treatment will be decreased rather than increased. As a consequence, it is appropriate to place most importance on entrectinib exposures when making dosing recommendations.

The final PBPK model has been used to derive dosing recommendations for co-administering entrectinib with CYP3A4 inhibitors or inducers. Based on the magnitude of the simulated interactions, 3-fold and 6-fold lower entrectinib doses (i.e., 200 mg and 100 mg) are required to mitigate the effects of moderate and strong CYP3A4 inhibitors, respectively. The appropriateness of the recommended dose adjustments was confirmed by further simulations of 100 mg and 200 mg entrectinib co-administered with CYP3A4 inhibitors. When considering the concomitant use of moderate and strong CYP3A inducers, the magnitude of the simulated interactions suggests that 2-fold and 4-fold higher entrectinib doses (i.e., 1200 mg and 2400 mg) would be required to mitigate the effects of enzyme induction. However, clinical use of entrectinib doses > 600 mg is not considered appropriate given the safety profile of entrectinib. While the recommended dose of 600 mg is well tolerated, doses above 600 mg produced dose-limiting toxicities in dose-finding studies [33, 42, 43]. Modeling of the exposure versus response relationship demonstrated that the likelihood of a patient experiencing a ≥ Grade 3 adverse event was markedly higher at exposures above those typically produced by 600 mg dosing [41]. Use of high doses of entrectinib would therefore carry potential safety risks for individuals, and in this context it is more prudent to recommend that use of entrectinib with moderate or strong CYP3A4 inducers be avoided rather than attempt a dose adjustment.

5 Conclusions

A PBPK model of entrectinib and its active metabolite M5 was developed, and has been shown to accurately predict the pharmacokinetics of both entrectinib and M5 in vivo. This model has been used to derive dosing recommendations for co-administering entrectinib with CYP3A4 inhibitors or inducers. A 6-fold lower entrectinib dose (i.e., 100 mg) is recommended when co-administered with a strong CYP3A4 inhibitor, and a 3-fold lower entrectinib dose (i.e., 200 mg) is recommended when co-administered with a moderate CYP3A4 inhibitor, but use of entrectinib with moderate or strong CYP3A4 inducers should be avoided. The PBPK modeling has been used in lieu of clinical studies to enable regulatory decision-making.