Drug–drug interaction prediction of ziritaxestat using a physiologically based enzyme and transporter pharmacokinetic network interaction model

Abstract

Ziritaxestat, an autotaxin inhibitor, was under development for the treatment of idiopathic pulmonary fibrosis. It is a substrate of cytochrome P450 3A4 (CYP3A4) and P‐glycoprotein and a weak inhibitor of the CYP3A4 and OATP1B1 pathways. We developed a physiologically based pharmacokinetic (PBPK) network interaction model for ziritaxestat that incorporated its metabolic and transporter pathways, enabling prediction of its potential as a victim or perpetrator of drug–drug interactions (DDIs). Concurrently, we evaluated CYP3A4 autoinhibition, including time‐dependent inhibition. In vitro information and clinical data from healthy volunteer studies were used for model building and validation. DDIs with rifampin, itraconazole, voriconazole, pravastatin, and rosuvastatin were predicted, followed by validation against a test dataset. DDIs of ziritaxestat as a victim or perpetrator were simulated using the final model. Predicted‐to‐observed DDI ratios for the maximum plasma concentration (C _max) and the area under the plasma concentration–time curve (AUC) were within a two‐fold ratio for both the metabolic and transporter‐mediated simulated DDIs. The predicted impact of autoinhibition/autoinduction or time‐dependent inhibition of CYP3A4 was a 12% decrease in exposure. Model‐based predictions for ziritaxestat as a victim of DDIs with a moderate CYP3A4 inhibitor (fluconazole) suggested a 2.6‐fold increase in the AUC of ziritaxestat, while multiple doses of a strong inhibitor (voriconazole) would increase the AUC by 15‐fold. Efavirenz would yield a three‐fold decrease in the AUC of ziritaxestat. As a perpetrator, ziritaxestat was predicted to increase the AUC of the CYP3A4 index substrate midazolam by 2.7‐fold. An overarching PBPK model was developed that could predict DDI liability of ziritaxestat for both CYP3A4 and the transporter pathways.

Study Highlights.

WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

The use of model‐informed drug development in addressing knowledge gaps in the product label is increasing. Specifically, physiologically based pharmacokinetic (PBPK) modeling, which is used to simulate untested clinical scenarios for metabolism‐based drug–drug interactions (DDIs), was applied in over 60% of submissions to the US Food and Drug Administration in a 10‐year time frame. In contrast, transporter‐based DDIs were only included in 7% of PBPK submissions in the same time period.

WHAT QUESTION DID THIS STUDY ADDRESS?

Application of in silico techniques to predict transporter‐based DDIs is challenging given the non‐specificity of transporter substrates. Such substrates commonly undergo enzymatic metabolism, and the metabolic or transporter liability usually cannot be disentangled from each other. Prediction of ziritaxestat's potential as either the victim or perpetrator of DDIs would have been instrumental in informing restrictions on concomitant medications during clinical use.

WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

The differentiation of the different pathways may be addressed by network interaction models that incorporate all major metabolic and transporter pathways involved in a compound's disposition, as well as pathways implicated in determining its DDI liability as a perpetrator with commonly prescribed co‐medications. The concept is illustrated with the new chemical entity (NCE) ziritaxestat, developed for the treatment of idiopathic pulmonary fibrosis (IPF), that is metabolized by CYP3A4 and is a substrate of P‐glycoprotein, and an inhibitor of CYP2C8 as well as transporters such as OATP1B1 and OATPB3. This yields a potential for DDIs with standard of care therapies for IPF such as nintedanib, as well as with statins that are routinely prescribed in this patient population.

HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

By addressing knowledge gaps in our understanding of the interplay between enzymatic and transporter pathways, such models could optimize DDI risk assessment for NCEs, and potentially enable interpolation of the findings to untested scenarios.

INTRODUCTION

Ziritaxestat is a small‐molecule, selective autotaxin inhibitor, that was, until recently, under development for the treatment of idiopathic pulmonary fibrosis (IPF) and systemic sclerosis. However, following the outcomes of the ISABELA phase III trials in IPF patients, the development was terminated after ziritaxestat 200 mg and 600 mg daily doses failed to demonstrate efficacy and were perceived to be associated with increased mortality compared with placebo. ¹

In vitro studies showed that ziritaxestat is primarily metabolized by cytochrome P450 (CYP)3A4, with 84.5% and 28.6% of [¹⁴C]‐ziritaxestat being metabolized at 10 and 100 μM, respectively, based on the percent compound remaining after 60 min in the presence of NADPH. The other CYP isoforms were minor contributors to the metabolism of ziritaxestat (maximum of 2.1%). In vitro data indicated that its metabolism could be affected by CYP3A4 autoinduction or autoinhibition. Ziritaxestat is a substrate for both P‐glycoprotein (P‐gp) and breast cancer resistance protein (BCRP) efflux transporters. In P‐gp‐ or BCRP‐transfected cell monolayers, the half‐maximal inhibitory concentration (IC₅₀) values for both P‐gp and BCRP inhibition were >40 μM, indicating a potential for inhibition. Thus, a transporter interaction at the intestinal level cannot be excluded. As a perpetrator of drug–drug interactions (DDIs), ziritaxestat demonstrated strong time‐dependent inhibition of CYP2C8, as well as weak inhibitory potential for the organic anion transporting polypeptide 1B1 (OATP1B1).

Following a single oral (p.o.) administration of 600 mg [¹⁴C]‐ziritaxestat to healthy male subjects, fecal excretion was the main route of elimination, based on total radioactivity (parent+metabolite), and renal excretion was negligible (<7%). The absolute bioavailability of ziritaxestat following a 600 mg dose was estimated to be approximately 51% and the fraction absorbed at least 86%. ² Ziritaxestat is a low‐clearance and low‐extraction‐ratio drug, indicating that hepatic first‐pass metabolism is likely to be low (9%). ² CYP3A4 was the main enzyme identified in vitro for the metabolism of ziritaxestat. This was confirmed by data showing that only one inhibitor of CYP3A4, ketoconazole, had a significant effect on the intrinsic clearance of ziritaxestat in human liver microsomes (data on file). The disposition of ziritaxestat following oral administration in humans is depicted in Figure S1. ²

The DDI potential of ziritaxestat was investigated in several clinical pharmacology studies, presented in Table S1. As ziritaxestat was intended to be administered on top of standard of care (nintedanib, pirfenidone) for IPF, understanding its DDI liability with these compounds, prior to phase III trials, was essential. CYP3A4 is a minor contributor to nintedanib's metabolism (~7%), and it is also a substrate for P‐gp (data on file). A DDI of unknown etiology has been reported for the combination of pirfenidone and nintedanib in a phase II study. ³ Over 30% of the IPF population are estimated to be on statin therapy, most commonly on simvastatin, atorvastatin, rosuvastatin, and pravastatin. These compounds are metabolized by CYP3A4 and CYP2C8 and transported by OATP1B1/B3 and BCRP. ⁴

The regulatory guidelines highlight the challenges in evaluating transporter DDIs. Given the relative lack of specificity of transporter substrates, inhibitors may also inhibit CYP‐mediated metabolism. This has been documented for CYP3A4 metabolism and P‐gp transport, as both of these pathways are mediated by the overarching nuclear pregnane X receptor pathway. ⁵ Hence, extrapolation of transporter‐mediated DDIs to other drugs is difficult. The use of physiologically based pharmacokinetic (PBPK) modeling to simulate clinical DDIs is well established for metabolic DDIs; however, its role in characterizing transporter DDIs is not as well documented. ⁶ , ⁷ , ⁸

The aim of the current analysis was to develop a network interaction PBPK model for ziritaxestat to predict its DDI liability as a victim and perpetrator (Figure 1). Concurrently, the net effect of ziritaxestat on autoinhibition/autoinduction of CYP3A4, as well as time‐dependent inhibition by a strong CYP3A4 inhibitor, were tested.

FIGURE 1.

Network of the drug–drug interactions that were incorporated in the model.

The interaction network model incorporated:

• Ziritaxestat as a substrate with perpetrators of the CYP3A4/P‐gp pathway, namely itraconazole, voriconazole, rifampin, efavirenz, and fluconazole.⁹

• Ziritaxestat as a perpetrator on CYP3A4/CYP2C8/CYP2C9 and OATP1B1/B3 and BCRP pathways, with impact on pravastatin, rosuvastatin, and midazolam.⁷

METHODS

Development of a PBPK model for ziritaxestat

Absorption

The Advanced Compartmental Absorption and Transit (ACAT™) model in GastroPlus® was used to describe the local solubility, dissolution, precipitation, absorption, and metabolism of ziritaxestat in each region of the intestinal tract. ¹⁰ Formulation dissolution was described by accounting for its pH‐dependent solubility and bile salt solubility. For this purpose, the Johnson dissolution model and the first‐order precipitation time model were employed, ¹¹ assuming fixed particle diameter of 2 μm for the suspension, 2.6 μm for the capsule, and 14 μm for the tablet formulation. Stomach transit time was decreased to 0.15 h instead of 1 h for suspension formulations. This adjustment was considered reasonable for a liquid formulation.

There is an uncertainty in the intestinal first pass (Fg) of ziritaxestat. The molecule is metabolized by CYP3A4 and is a substrate of P‐gp based on in vitro data (data on file). Thus the integration of both CYP3A4 and P‐gp at the intestinal level was tested against clinical data.

Paracellular absorption was included, although its contribution to the overall absorption was negligible. The absorption rate constant (K _a) for each gastrointestinal (GI) compartment is the product of the effective permeability (P _eff) and the absorption scale factor (ASF) for the compartment, which account for changes in permeability as the drug travels along the GI tract. The P _eff value was fitted against a plasma concentration versus time profile after oral (p.o.) administration of 600 mg ziritaxestat. The default “Opt logD Model SA/V 6.1” ASF model was used in all simulations.

Systemic disposition

A whole‐body PBPK model, which incorporates a wide range of anatomical, physiological, and drug‐disposition parameters and characteristics, was developed.

The Population Estimates for Age‐Related Physiology® module, which includes algorithms that account for sex‐, age‐, and body weight‐dependent changes in the physiological and anatomical parameters (such as fasted and fed intestinal physiologies, blood flows, cardiac output, organ/tissue volumes, and compositions), was used to create whole‐body physiologies for all simulations. ¹⁰ To reproduce the available clinical studies, single‐dose simulations were performed with a virtual individual representing the mean tendency of the healthy volunteers as closely as possible in terms of sex, age, body weight or body mass index, and dosing regimen. The sex of the virtual subject was set to match the sex of the majority of the subjects in each studied population.

There was an uncertainty in the human V _ss prediction. Two models were tested during the development, one with perfusion‐limited tissues and one with permeability‐limited tissues.

The tissue distribution was simulated using a perfusion‐limited tissue model for the liver and a permeability‐limited tissue model for all other tissues. The tissue–plasma partition coefficients (K _ps) were predicted using default methods, namely Lukacova for perfusion‐limited and Poulin and Theil extracellular for permeability‐limited tissues. ¹² , ¹³ The permeability surface‐area products (PStc) for tissues described by a permeability‐limited model were calculated from specific PStc values (PStc per mL of tissue cell volume) and the individual tissue cell volumes. The specific PStc was fitted along with logD against the pharmacokinetic (PK) profile after intravenous (i.v.) administration in humans.

The CYP3A4 pathway was characterized by the kinetic parameters of maximal velocity (V _max) and the Michaelis constant (K _m), along with the default expression levels of CYP3A4 in the gut and liver. The kinetics of ziritaxestat's metabolism by CYP3A4 were parameterized with K _m from an in vitro experiment and V _max fitted against the observed p.o. data. The V _max values in the gut and liver were synchronized to account for differences in enzyme expression levels in these tissues. Autoinhibition of its own metabolism by CYP3A4 was considered in the model simulations. Although negligible, passive renal excretion of ziritaxestat was estimated as the unbound fraction in plasma multiplied by the glomerular filtration rate (Fup*GFR). The GFR was provided by GastroPlus for the standard population considered.

Similar to the CYP3A4 pathway, the P‐gp pathway was investigated by incorporating its kinetic parameters V _max and K _m and its tissue expression level in the liver and gut. The expression levels of P‐gp in relevant tissues were used as provided by GastroPlus. The K _m was estimated from an in vitro experiment (data on file) and V _max was fitted against the observed clinical data.

DDI modeling and validation

Following verification of ziritaxestat's PK profile prediction using the p.o. single‐ and multiple‐dose clinical data, the PBPK model was used to simulate DDIs between ziritaxestat as victim, and itraconazole, voriconazole, and rifampin as perpetrators. The itraconazole, voriconazole, and rifampin PBPK models were available from the GastroPlus library for input in the simulations. These PBPK models were previously validated by reproducing published results of clinical DDI studies. The in vitro interaction parameters used in the simulations are listed in Table 1. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ Simulations were set to match the clinical study designs with these perpetrator drugs (Table S2). The simulated individual had a normal CYP2C19 polymorphism, given this pathway is not implicated in the metabolism of ziritaxestat, the substrate drug. The contribution of CYP3A4 to the overall ziritaxestat elimination was verified using three DDI clinical studies (one induction and two inhibitions), acting on the CYP3A4 pathway with ziritaxestat as the victim.

TABLE 1.

Input interaction parameters used in the drug–drug interaction simulations with ziritaxestat.

Study	Perpetrator	Enzyme	Interaction type	Parameter	Parameter value	F u in vitro	Source
1	Itraconazole OH‐itraconazole Keto‐itraconazole ND‐itraconazole	CYP3A4	Competitive inhibition	K i, U	1.3 nM 14.4 nM 14 nM 0.38 nM		Isoherranen et al. 14
2	Voriconazole	CYP2C19	Time‐dependent inhibition	K I, k inact	8 μM 0.11 min−1		Jeong et al. 15 GastroPlus® user manual 10
		CYP3A4	Time‐dependent inhibition	K I, k inact	2.97 μM 0.006 min−1		Jeong et al. 15
			Competitive inhibition	IC50	9.3 μM		Center for Drug Evaluation and Research 16
3	Rifampin	CYP3A4	Induction	EC50, U E max	0.064 μM 15		Asaumi et al. 28 GastroPlus user manual 10
		CYP3A4	Competitive inhibiton	K i, U	18.5 μM		Kajosaari et al. 32
4	Fluconazole	CYP3A4	Competitive inhibition	K i, U	7.4 μM		Isoherranen et al. 33
5	Efavirenz	CYP3A4	Induction	EC50 E max	1.00 μM 9.9		GastroPlus user manual 10
	Efavirenz	CYP2B6	Induction	EC50 E max	0.82 μM 5.76
6	Ziritaxestat	CYP3A4	Inhibition	IC50, T	3.7 μM	0.24	Austin et al. 34 (measured a )
7	Ziritaxestat	CYP2C8	Inhibition	K i, k inact, T k inact	0.753 μM 0.0353 min−1	0.52	Measured
8	Ziritaxestat	CYP2C9	Inhibition	IC50, T	13.1 μM	0.24	Measured
9	Ziritaxestat	OATP1B1	Inhibition	IC50, U	0.1 μM	0.605	Austin et al. 34 (fitted a )
10	Ziritaxestat	OATP1B3	Inhibition	IC50, U	7.04 μM	0.605	Austin et al. 34 (measured a )
11	Ziritaxestat	BCRP	Inhibition	IC50, U	>40 μM	0.605	Austin et al. 34 (fitted a )

Abbreviations: CYP, cytochrome P450; DDI, drug–drug interaction; EC₅₀, half‐maximal effective concentration; E _max, maximal effect; F _u, fraction unbound; IC₅₀, half‐maximal inhibitory concentration; K _I, concentration at 50% maximal inactivation; k _inact, maximal inactivation; inhibition or induction parameters for voriconazole and rifampin refer to total concentrations; ND, N‐desalkyl; OH, hydroxy; T, total concentration; U, unbound concentration.

^a The method described by Austin et al. ³⁴ was applied to compute non‐specific microsomal binding.

Lastly, the PBPK model was employed to simulate DDIs with ziritaxestat as perpetrator and rosuvastatin and pravastatin as victims. The model for rosuvastatin was available from the GastroPlus library, while that for pravastatin was developed specifically for this purpose. The in vitro interaction parameters used in the simulations are listed in Table 1. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ The predictive performance of the model involving the inhibition of OATP1B1/B3 and the BCRP transporter was assessed by reproducing the results of a clinical DDI study with the probe drugs pravastatin and rosuvastatin following single oral dose simulations. Standard DDI clinical study designs were simulated wherein ziritaxestat was dosed in the absence of and in combination with the victim (rosuvastatin, pravastatin) (Table S2). The simulated DDIs were verified using data from two standard clinical DDI studies in healthy subjects, in which the victim drugs (rosuvastatin and pravastatin) were dosed orally in the absence of and in combination with the perpetrator ziritaxestat. The inhibition constants for OATP1B1 and BCRP were adjusted to provide the best fit to the observed data.

The DDI model in GastroPlus accounts for the effect of a time‐dependent inhibitor, competitive inhibitor, and inducer on the victim drug's metabolism or carrier‐mediated transport by a specific enzyme or transporter by changing the protein (enzyme or transporter) activity (see Equation 1) and by adjusting the victim drug's K _m to account for the competitive aspect of the interaction (see Equation 2), where ProtAct_t is the protein (enzyme or transporter) activity expressed in a given tissue at a given time; $K_{\mathrm{i}}^{\mathrm{rev}}$, $K_{\mathrm{i}}^{\text{irrev}}$, and EC₅₀ are the reversible (competitive) inhibition, irreversible inhibition, and induction constants, respectively, of a perpetrator drug against individual enzymes or transporters involved in the victim drug's metabolism or carrier‐mediated transport; k _inact and E _max are the inactivation rate constant and maximum induction potential, respectively, of a time‐dependent inhibitor against individual enzymes or transporters involved in the victim drug's metabolism or carrier‐mediated transport; and k _deg is the turnover rate of an enzyme or transporter.

$$\begin{array}{c}\frac{\mathrm{d}{\text{ProtAct}}_{\mathrm{t}}}{\mathrm{d}t}=-\left(\frac{k_{\text{inact}}\times {\left[I\right]}_{\mathrm{u}}}{K_{\mathrm{i}}^{\text{irrev}}+{\left[I\right]}_{\mathrm{u}}}\right)\times \mathrm{P}{\text{rotAct}}_{\mathrm{t}}+k_{\mathrm{deg}}\hfill \\ \times \left({\text{ProtAct}}_0-{\text{ProtAct}}_{\mathrm{t}}\right)+k_{\mathrm{deg}}\hfill \\ \times {\text{ProtAct}}_0\times \frac{E_{\max }\times {\left[I\right]}_{\mathrm{u}}}{{\mathrm{EC}}_{50}+{\left[I\right]}_{\mathrm{u}}}\hfill \end{array}.$$ (1)

$$v=\frac{{\text{ProtAct}}_{\mathrm{t}}\times V_{\max }\times {\left[S\right]}_{\mathrm{u}}}{K_{\mathrm{m}}\left(1+\frac{{\left[I\right]}_{\mathrm{u}}}{K_{\mathrm{i}}^{\mathrm{rev}}}\right)+{\left[S\right]}_{\mathrm{u}}}.$$ (2)

Prospective DDI simulations

The ziritaxestat PBPK model was used along with default GastroPlus models for probe substrates and perpetrators of several enzymes and transporters to predict DDIs under different scenarios.

• Ziritaxestat as a victim of

  • A moderate CYP3A4 inducer (efavirenz) and a moderate CYP3A4 inhibitor (fluconazole).
• Ziritaxestat as a perpetrator on

  • CYP3A4 as a potential moderate inhibitor and inducer of midazolam (sensitive index substrate) metabolism.

Prior to the respective simulations, the performance of the probe compounds (efavirenz, fluconazole, and midazolam) was verified by reproducing a published DDI. The results are presented in Tables S5–S7.

Summary of key assumptions

The Johnson's dissolution model was applied, that assumed a fixed particle size and diameter. This is a mechanistic model, wherein the dissolution is a function of API solubility and diffusion coefficient, particle size, and density. Fixed particle radius assumed that the dissolution rate will not increase due to the shrinkage of particles. ¹⁸

There was uncertainty as to the gastrointestinal first pass of ziritaxestat. Thus the integration of both CYP3A4 and P‐gp at the intestinal level was tested against clinical data. As these effects are difficult to disentangle, only the identifiable pathway was retained in the model. This implies that the model may overestimate the impact of one pathway.

Ziritaxestat exhibits high permeability, hence intestinal absorption was modeled as passive diffusion, with no carrier‐mediated uptake or efflux. The high permeability would also likely limit the impact of P‐gp on the absorption, further supporting the simplified model without intestinal P‐gp.

In the clinical DDI study between ziritaxestat and voriconazole, the CYP2C19 genotype for the subjects was not determined, and a normal/average CYP2C19 metabolizer was assumed in the simulation. Possible impact of CYP2C19 metabolizer status was not further explored.

Physiologies for typical healthy individuals were generated by PEAR Physiology™ module in GastroPlus. Simulated individuals expressed CYP3A4.

No carrier‐mediated hepatic uptake was assumed.

Renal excretion was considered as a passive process, and no influence of transporters was assumed, although P‐gp is expressed in the kidney also.

Sensitivity analysis of the ziritaxestat model parameters

Sensitivity analyses were performed on single parameters in the final ziritaxestat model. The parameters selected were the ones that were fitted. Changes in the parameter values were measured for their impact on maximum plasma concentration (C _max) and area under the plasma concentration–time curve (AUC) of a single 600 mg dose administered as tablet in the fasted state. Conditions for each parameter are described in Table S12.

Sensitivity analysis of the interaction parameters

Uncertainty in the fitted inhibition parameter values for OATP1B1 and BCRP was investigated by lowering the IC₅₀ value by 10‐fold to 100‐fold. The objective was to quantitatively assess the impact of uncertainty associated with the inhibition parameter on an outcome DDI. This investigation informed the gap between in vitro and in vivo results. Simulations were performed with the in vitro value first, then with a 10‐fold and 100‐fold decrease (Tables S10 and S11). For CYP3A4, the sensitivity analysis is presented in Table S14.

PBPK model evaluation

The predictive performance of the model was evaluated by visual inspection of overlaid predicted and observed PK profiles. The fold error for the predicted parameters t _max, C _max, and AUC was computed as follows$:$

$$\text{fold error}=\frac{\text{predicted}\mathrm{PK}\text{parameter}}{\text{observed}\mathrm{PK}\text{parameter}}.$$ (3)

All AUC values were calculated from time zero to infinity (AUC_0–∞) or time of the last concentration measurement (AUC_0–tau) depending on whether simulations were single dose or up to steady state. Predictions were considered accurate if they fell within the 0.80–1.25 error range, acceptable if within 0.50–2.00, and inaccurate if outside of the two‐fold error range. ⁸ , ¹⁹

Model validation entailed comparing predicted versus observed plasma concentration–time profiles of the victim drug when administered alone and during co‐administration, along with the comparison of the AUC and C _max ratios (fold error – see previous equations).

Additionally, the acceptance limits reported by Guest et al. were applied. The limits tighten when the observed ratio is close to 1 and gradually tend to the traditional two‐fold limits as the observed ratio becomes larger. ²⁰

Software

GastroPlus software version 9.7 (Simulations Plus) was used to develop the PBPK model for ziritaxestat. The ADMET Predictor® version 9.0 (Simulations Plus) was used to obtain in silico estimates of key physicochemical and biopharmaceutical parameters from the molecular structure when experimentally determined values were not available, or to provide an objective alternative to experimental data. The PBPKPlus™ module (Simulations Plus) was used to simulate the systemic PK profiles and the DDI module was used to simulate the clinical DDIs. Phoenix WinNonLin 8.1 (Certara USA, Inc.) was used to determine PK parameters (AUC and C _max) using non‐compartmental analysis. Validation of all probes used in the simulations is described in the supplemental material.

Clinical studies

All subjects gave their written informed consent to participate in the studies. All studies were conducted following approval by the institutional review board and in accordance with the Declaration of Helsinki.

For the development and calibration of the ziritaxestat PBPK model, data from a number of oral studies and one i.v. study in healthy volunteers were available. The oral dose data included data from the first‐in‐human (FIH) single and multiple‐ascending dose studies with a liquid suspension.

Model validation was performed using a subset of data from the FIH study, as well as a relative bioavailability study with a capsule and tablet formulation. The various clinical studies used for either model development or validation are presented in Table S1 and S3. ² , ²¹

Clinical data with CYP3A4/P‐gp inhibitors (itraconazole, voriconazole) and an inducer (rifampin) were used to validate the CYP3A4 model for ziritaxestat as the victim drug, and with pravastatin and rosuvastatin to validate the metabolic enzyme/transporter model for ziritaxestat as the perpetrator.

RESULTS

Development of a PBPK model for ziritaxestat

The performance of the two tissue distribution models was comparable with regards to the DDI predictions. As the model with mostly permeability‐limited tissue distribution resulted in better prediction of the shape of the plasma profile and C _max after i.v. infusion, it was selected as the final model, except for the liver, which was described with the perfusion‐limited tissue model.

For the metabolic and transport processes, two models were evaluated during model development: one including only CYP3A4 in the gut and liver, and one including both CYP3A4 and P‐gp. Addition of P‐gp did not improve model performance. The intestinal absorption was modeled as passive diffusion only, with no carrier‐mediated process included. The available in vitro and in vivo data were not sufficient to disentangle the effect of CYP3A4 and P‐gp at the gut level. Thus, the simplest model with only CYP3A4 expressed at the gut and liver was retained. The logD was decreased to 3.0 (as opposed to a measured value of 3.81) to recalculate the tissues‐to‐plasma partition coefficients (K _ps) for best fit of the i.v. profile. Similarly, the effective permeability was identified as 1.2 (10⁻⁴ cm/s) while the measured apparent permeability in Caco‐2 cells was 0.47 (10⁻⁴ cm/s). The final drug‐dependent input parameters are presented in Table 2 (data on file). ² , ¹⁷

TABLE 2.

Summary of input parameters in the physiologically based pharmacokinetic (PBPK) model of ziritaxestat.

Parameter	Description	Value	Source
Chemical structure	–	C30H33FN8O2S
Molecular weight	Molecular weight (free base) (g/mol)	588.71
SolRef	Reference solubility in buffer at pH 7.5 (mg/mL)	0.066	Measured
Solubility	FaSSIF/FeSSIF (mg/mL)	0.54/3.49	Measured
pK a	–	pK a1 (base): 4.99/pK a2 (base): 5.89	Measured
LogD	Log of octanol/water distribution coefficient	3 at pH 8.25 (measured = 3.81)	Fitted against i.v. data
P eff	Effective permeability (10−4 cm/s)	1.2 (measured Papp value in Caco‐2 cells = 0.47)	Fitted against p.o. data for 600 mg dose
F u p	Plasma unbound (%)	0.9	Measured
B:P	Blood to plasma ratio	0.6	Measured
Metabolism	CYP3A4 liver/gut	K m = 3.03 mg/L V max a  = 0.002096 mg/s/mg of enzyme (liver) V max b  = 0.9264 mg/s (gut)	V max fitted against clinical data, K m values were estimated from in vitro data (measured c )
CLr	Renal clearance (mL/h)	F u p*GFR	–
Particle size	Mean radius (μm)	Fixed	Defined according to formulation used in clinical data (1–28 μm)
Density	Apparent density (g/mL)	1.2	Default value
Mean precipitation time	(s)	10,000	Increased to fit clinical data
D w	Aqueous diffusion coefficient (105 cm2/s)	0.52	Measured (ADMET Predictor®)

Abbreviations: CYP3A4, cytochrome P450 3A4; FaSSIF, fasted‐state simulated intestinal fluid; FeSSIF, fed‐state simulated intestinal fluid; F _up, unbound fraction in plasma; GFR, glomerular filtration rate; i.v., intravenous; K _m, Michaelis constant; PBPK, physiologically based pharmacokinetic; pK _a, acid dissociation constant; p.o., oral; V _max, maximal velocity.

^a V _max values in the gut and liver were synchronized to account for differences in CYP3A4 expression levels.

^b Predicted from structure using ADMET Predictor® version 9.0.0.

^c Galapagos. Data on file: ZIRITAXESTAT‐PK‐040; January 2019.

The mean observed and simulated ziritaxestat plasma concentration–time profiles after single (i.v. and oral) and multiple (oral) doses are presented in Figures S2 and S3, respectively. Predictive performance and performance verification of the model are depicted in Figures 2 and 3, respectively. The observed data were predicted with fold errors 0.78–1.38 for C _max and 0.69–1.46 for AUC.

FIGURE 2.

Predictive performance of the final model depicted as the correlation between the predicted and observed maximum plasma concentration (C _max) (top panel) and area under the plasma concentration–time curve (AUC) (bottom panel) in all included clinical studies. The middle black line marks the line of identity. Gray lines and outside black lines indicate the 0.80–1.25‐fold and 0.50–2.0‐fold acceptance limits, respectively.

FIGURE 3.

Ziritaxestat physiologically based pharmacokinetic (PBPK) model verification following a single 1500 mg dose and multiple 600 mg doses in healthy Caucasian subjects.

The PKs of ziritaxestat could be well described following single or multiple doses, administered i.v. (0.1 mg fasted) or p.o. and ranging from 20 to 1500 mg (single dose) and from 150 to 1000 mg (once or twice daily, multiple doses). Oral doses were administered under fed conditions as a suspension, capsule, or tablet. Single‐dose PKs of ziritaxestat administered as a tablet in a fasted state were also well described (Table S3). Overall, the estimated PK parameters (C _max and AUC) were all within two‐fold (100%) of the observed values, with 77.8% falling within 1.25‐fold (Figure 2). Sensitivity analysis suggested that the C _max was sensitive to blood: plasma ratio, intestinal permeability, stomach transit time, logD, and V _max of CYP3A4. The T _max was sensitive to stomach transit time while AUC was sensitive to CYP3A4 V _max (Figure S5A–C).

Retrospective DDI simulations

CYP3A4 interaction model with ziritaxestat as the victim drug

The predictive performance of the model involving CYP3A4 inhibition pathway was assessed by reproducing the results of clinical DDI studies with the probe drugs itraconazole, voriconazole, and rifampin.

Following multiple‐dose simulations, the simulated‐to‐observed (single dose) ratios for itraconazole C _max and AUC from time zero to infinity (AUC_0–∞) were 1.5/1.4 and 1.9/3.1, respectively, and for voriconazole these were 1.7/1.4 and 4.2/4.0, respectively (Figure 4). In the latter case due to time‐dependent inhibition (Figure S4). For rifampin, the simulated‐to‐observed ratios for C _max and AUC_0–24 were 0.32/0.17 and 0.12/0.11, respectively, following a single‐dose simulation with ziritaxestat. The comparison of predicted‐to‐observed C _max and AUC ratios for a typical subject is presented in Table S4. ²⁰ The ratios were within acceptable limits except in the case of rifampin, where C _max ratio was overpredicted (Table S4).

FIGURE 4.

Predictive performance of the drug–drug interaction (DDI) simulations. Correlation between the predicted and observed DDI (maximum plasma concentration, C _max) and DDI (area under the plasma concentration–time curve, AUC) in all included studies. The middle black line marks the line of identity. Gray lines and outside black lines indicate the 0.80–1.25‐fold and 0.50–2.0‐fold acceptance limits. The dashed lines show the prediction success limits suggested by Guest et al. ²⁰

Autoinduction and autoinhibition of CYP3A4 by ziritaxestat

The impact of ziritaxestat on its own metabolism was assessed after multiple 1000 mg doses, which were intended to maximize the effect. The net impact of the aforesaid processes was predicted to be up to 12% decrease in the AUC, with C _max remaining practically unaltered (Figure S3, Table S13).

Ziritaxestat as a perpetrator of metabolic enzymes/transporters

The performance of the GastroPlus probe pravastatin could be verified as its interaction with gemfibrozil (OATP1B1 inhibitor) could be reproduced. The observed‐to‐predicted C _max and AUC ratios were 1.10 and 1.03, respectively.

The predicted ratios for C _max or AUC_0–∞ (ziritaxestat plus pravastatin/rosuvastatin or pravastatin/rosuvastatin alone) are presented in Table S4. The simulated‐to‐observed ratios for pravastatin C _max and AUC were 1.94/2.40 and 2.12/2.52, respectively, while for rosuvastatin these were 3.59/2.40 and 2.48/1.92, respectively.

Prospective DDI simulations with ziritaxestat

As victim with a moderate CYP3A4 inducer (efavirenz)

The performance verification of efavirenz with ketoconazole is displayed in Table S5. The observed DDI could be reproduced with the observed‐to‐predicted C _max and AUC ratios of 0.70 and 0.90, respectively.

The prospective DDI simulation with the induction of CYP3A4 was assessed with efavirenz as the perpetrator. Based on the model simulations, administration of a single 600 mg dose of ziritaxestat after multiple 400 mg doses of efavirenz QD would result in ratios of 0.53 and 0.33 for C _max and AUC_0–∞ for ziritaxestat, respectively.

As victim with a moderate CYP3A4 inhibitor (fluconazole)

The performance verification of fluconazole with midazolam is displayed in Table S6. A single dose of fluconazole (unbound Ki), co‐administered with ziritaxestat, resulted in a 1.58‐fold increase in C _max and a 3.45‐fold increase in AUC_0–∞. Administration of ziritaxestat after multiple doses of fluconazole resulted in a 1.77‐fold increase in C _max and a 6.55‐fold increase in AUC_0–∞.

As perpetrator of a sensitive CYP3A4 substrate (midazolam)

The performance verification of ketoconazole with midazolam is displayed in Table S7. The predicted DDI results are in line with observed data, thus verifying the midazolam probe.

The predicted DDI between ziritaxestat and midazolam is depicted in Table 3. In addition, since CYP3A4 is located in the gut and liver, the impact of CYP3A4 inhibition on the amount of midazolam metabolized at each site is illustrated in Figure S6. After a single dose administered under fed conditions, at baseline, almost 40% of the dose of midazolam is metabolized in the liver and 60% in the gut. After CYP3A4 inhibition by ziritaxestat, the amount of midazolam metabolized in the gut is decreased to about 10%. Therefore, much more of the drug is available for hepatic metabolism (90%). Overall, the inhibition is more marked in the gut than in the liver and the net effect is an increase in exposure to midazolam by a factor of 2 to 3.

TABLE 3.

Summary of predicted area under the plasma concentration–time curve (AUC) and maximum plasma concentration (C _max) ratios from prospective drug–drug interaction simulations with index drugs and ziritaxestat (as CYP3A4 substrate or perpetrator).

Type of interaction	Perpetrator	Perpetrator dosing	Victim dosing	Simulated C max ratio (victim)	Simulated AUC ratio (victim)
Inhibition of CYP3A4	Ziritaxestat	Ziritaxestat p.o. 600 mg Single dose	Midazolam p.o. 7.5 mg Single dose	2.38	2.98
	Ziritaxestat	Ziritaxestat p.o. 600 mg Once daily	Midazolam p.o. 7.5 mg Single dose on Day 7	2.31	2.99
Inhibition + induction of CYP3A4	Ziritaxestat	Ziritaxestat p.o. 600 mg Single dose	Midazolam p.o. 7.5 mg Single dose	2.31	2.90
	Ziritaxestat	Ziritaxestat p.o. 600 mg Once daily	Midazolam p.o. 7.5 mg Single dose on Day 7	2.25	2.76
Moderate inhibition of CYP3A4	Fluconazole	Fluconazole p.o. 400 mg Single dose	Ziritaxestat p.o. 600 mg Single dose	1.58	3.45
	Fluconazole	Fluconazole p.o. 400 mg Once daily	Ziritaxestat p.o. 600 mg Single dose on Day 7	1.77	6.55
Moderate induction of CYP3A4	Efavirenz	Efavirenz p.o. 600 mg Once daily	Ziritaxestat p.o. 600 mg Single dose on Day 7	0.53	0.33

Abbreviations: AUC, area under the plasma concentration–time curve; C _max, maximum plasma concentration; CYP3A4, cytochrome P450 3A4; p.o., oral.

DISCUSSION

Clinical DDI studies, along with in vitro assays and in silico models, are standard practice in the evaluation of the risk of DDIs with new chemical entities (NCEs). PBPK modeling is key, especially in the evaluation of complex DDIs involving multiple pathways, including metabolic enzymes as well as transporters. ²² Drugs that are metabolized by the CYP3A4 pathway are likely P‐gp substrates. ²³ PBPK models incorporate not only human physiology, such as organ blood flows and transporter and CYP enzyme expression, but also compound physicochemical properties. ²⁴ Network interaction models, by incorporating all relevant pathways, provide a holistic perspective of the DDI liability of a given compound. ²⁵ , ²⁶ They may be deployed to investigate potential clinical scenarios in cases where dedicated trials are not always possible, or help to optimize the design of clinical DDI studies ²²

CYP3A4 was the major enzyme identified in vitro for the metabolism of ziritaxestat. In the PBPK model, this pathway was characterized by its kinetic parameters V _max and K _m and the enzyme expression level in relevant tissues, as incorporated in GastroPlus. V _max/K _m values were obtained from an in vitro experiment. However, further adjustment of V _max against clinical data were necessary, indicating the limitations of in vitro assays ²⁷ and underlining the need for adequate data for model calibration.

Ziritaxestat's tissue distribution was described as permeability‐limited, except in the liver, where it was described as perfusion‐limited. This accounted for sufficient metabolism in the liver while obviating the need for additional processes such as hepatic uptake that, though relevant, may have been difficult to identify. This distribution model was also verified in four preclinical species following i.v. administration (data not shown) and allowed a good fit of the observed profiles. The estimated apparent volume of distribution at steady state, estimated based on a simulated profile, was about 38 L, similar to that calculated from observed data (mean [SD] of 23 [11] L). ² The final model could predict the disposition of ziritaxestat across a range of dosing conditions. There was also a good agreement between estimated (65%) and observed (51%) bioavailability.

Validation of the DDI model (ziritaxestat as the victim) against clinical data established its usefulness in predicting DDIs with moderate inducers or inhibitors of the CYP3A4 pathway. Incorporation of P‐gp transport did not improve the predictions. Hence, it was not retained in the model. Our simulations also suggested that ziritaxestat was unlikely to influence its own metabolism in vivo. This was confirmed in clinical studies as well. ²¹ At the same time, time‐dependent inhibition by voriconazole would increase ziritaxestat's exposure by nearly four‐fold, following multiple doses. DDI ratios were within acceptable limits except in the case of rifampin, where the C _max ratio was slightly outside of the Guest acceptance limits (Table S4). From the aforesaid, a few limitations of the model can be inferred. First, P‐gp transport could not be disentangled from CYP3A4 metabolism, and was unidentifiable in our model. This may also explain why we overpredicted C _max, when co‐administered with rifampin, that is a perpetrator for several enzymes and transporters. ²⁸ Second, the distribution model accounted for sufficient metabolism in the liver while avoiding the need for active uptake transport processes that could not be reliably parametrized with the available data. For drugs undergoing significant hepatic uptake or elimination, OATP1B liability should be evaluated. ²⁹ The revised FDA guidelines on the evaluation of DDI liability ³⁰ recommend eight clinically important transporters, including P‐gp, BCRP, and OATP1B3. However, several challenges exist in the characterization of such a liability. Transporters do not function in isolation and substrates demonstrate overlapping specificity with metabolic enzymes. ²⁹ Understanding this interplay is key to adequate DDI design. ²⁶ This challenge is exacerbated since the FDA does not provide index drug recommendations for in vitro transporter studies. ⁹

Conversely, Chiney et al. reported the use of PBPK modeling to inform labeling for dosing recommendations for the DDI potential of elagolix (CYP3A4 inducer, P‐gp inhibitor) for doses not evaluated in phase I clinical studies. ³¹ The International Transporter Consortium recommends pravastatin and rosuvastatin as probe substrates with a relatively high specificity and sensitivity for BCRP and OATPB1/OATPB3, respectively. ⁷ , ²⁹

This work was intended to inform on the need to conduct a clinical DDI study or form the basis of product labeling for some drug–drug interactions to complement DDI studies conducted. Assuming 600 mg daily as the therapeutic dose, some recommendations may be made. Dose reduction of ziritaxestat would be needed when co‐administered with sensitive CYP3A4 substrates or use of a lower dose of the latter if applicable. For less sensitive substrates, a lower impact could be expected (<2‐fold). For statins, given the multiple enzymes and transporters involved in their ADME ("absorption, distribution, metabolism, and excretion") profiles and their narrow therapeutic index, a clinical DDI study with different statins (simvastatin, atorvastatin, rosuvastatin, and pravastatin) was necessary (data on file).

Based on clinical studies, ziritaxestat was classified as a moderately sensitive CYP3A4 substrate, given an increase in the AUC of ≥2 to <5 following co‐administration with index CYP3A4 inhibitors, itraconazole and voriconazole. Based on our work, co‐administration of moderate CYP3A4 inhibitors is not recommended or the dose of ziritaxestat requires adjustment to avoid exceeding the safety exposure margin. ²¹

Finally, the potential impact of ziritaxestat on the pharmacokinetics of the standard of care (pirfenidone, nintedanib) was evaluated. In vitro ziritaxestat did interact with CYP1A2, that is involved in pirfenidone metabolism. This was confirmed in a clinical DDI study, while an increase by approximately two‐fold in nintedanib exposure was noted (data on file). In the ISABELA phase III studies in IPF patients, a decrease in ziritaxestat exposure (1.7–2.6‐fold) was observed at 200 mg and 600 mg ziritaxestat doses when co‐administered with pirfenidone. ¹ Similarly, pirfenidone has previously been shown to reduce the exposure of nintedanib (1.5‐fold). ³ As both, ziritaxestat and nintedanib, are P‐gp substrates, these interactions may have been mediated by P‐gp transporters. In the event of continued development, the impact of interactions between ziritaxestat and standard of care (pirfenidone and nintedanib) and efficacy and/or safety responses would have mandated further exploration.

Overall, we were able to characterize the perpetrator liability of ziritaxestat in conformance with the standard criteria as the prediction error ranged from 0.69 to 1.46. Though clinical DDI studies with ziritaxestat as the perpetrator were necessary, untested scenarios would have been adequately informed by our model.

CONCLUSIONS

We developed a validated and comprehensive PBPK model for the CYP3A4 and P‐gp substrate ziritaxestat, which encompassed its perpetrator DDI liability on key transporter pathways while capturing the interplay between relevant metabolic enzymes and implicated transporters.

Such a model could assist decision making and hence inform labeling recommendations for relevant DDIs, potentially obviating the need for some clinical DDIs. Finally, we demonstrated an example of the recommended integrated approach ²⁴ combining in vitro, in vivo, and in silico evaluations for a comprehensive assessment of the DDI potential of a NCE.

AUTHOR CONTRIBUTIONS

J.P. and V.G. performed the research. E.H., F.N., and A.T. designed the research. V.L. designed and performed the research. J.P. and A.T. wrote the manuscript.

FUNDING INFORMATION

This analysis was funded by Galapagos.

CONFLICT OF INTEREST STATEMENT

J.P. is an employee of PhinC Development. He declared no competing interests for this work. V.G. is an associate director and co‐founder of PhinC Development. She declared no competing interests for this work. E.H. was an employee of Galapagos, owns subscription rights in the company, and now works for another company. He declared no competing interests for this work. F.N. is an employee of Galapagos. She declared no competing interests for this work. V.L. is an employee and holds stock of Simulations Plus, Inc. She declared no competing interests for this work. A.T. is a former employee of Galapagos and owns subscription rights in the company, and is now employed by Simulations Plus, Inc. He declared no competing interests for this work.

Supporting information

Data S1

Perrier J, Gualano V, Helmer E, Namour F, Lukacova V, Taneja A. Drug–drug interaction prediction of ziritaxestat using a physiologically based enzyme and transporter pharmacokinetic network interaction model. Clin Transl Sci. 2023;16:2222‐2235. doi: 10.1111/cts.13622