Visual Overview
Abstract
Physiologically based pharmacokinetic (PBPK) modeling for itraconazole using a bottom-up approach is challenging, not only due to complex saturable pharmacokinetics (PK) and the presence of three metabolites exhibiting CYP3A4 inhibition, but also because of discrepancies in reported in vitro data. The overall objective of this study is to provide a comprehensive mechanistic PBPK model for itraconazole in order to increase the confidence in its drug-drug interaction (DDI) predictions. To achieve this, key in vitro and in vivo data for itraconazole and its major metabolites were generated. These data were crucial to developing a novel bottom-up PBPK model in Simcyp (Simcyp Ltd., Certara, Sheffield, United Kingdom) for itraconazole and two of its major metabolites: hydroxy-itraconazole (OH-ITZ) and keto-itraconazole (keto-ITZ). Performance of the model was validated using prespecified acceptance criteria against different dosing regimens, formulations for 29 PK, and DDI studies with midazolam and other CYP3A4 substrates. The main outcome is an accurate PBPK model that simultaneously predicts the PK profiles of itraconazole, OH-ITZ, and keto-ITZ. In addition, itraconazole DDIs with midazolam and other CYP3A4 substrates were successfully predicted within a 2-fold error. Prediction precision and bias of DDI expressed as geometric mean fold error were for the area under the concentration-time curve and peak concentration, 1.06 and 0.96, respectively. To conclude, in this paper a comprehensive data set for itraconazole and its metabolites is provided that enables bottom-up mechanism-based PBPK modeling. The presented model is applicable for studying the contribution from the metabolites and allows improved assessments of itraconazole DDI.
Introduction
After recommendations to stop the usage of ketoconazole by regulatory agencies (CDER, 2013; CHMP, 2013), itraconazole has emerged as one of the best candidates to use as a standard CYP3A4 inhibitor in drug-drug interaction (DDI) studies (CDER, 2017a). Nevertheless, relatively limited experience exists for itraconazole in clinical DDI studies and the optimal study design remains to be defined (Liu et al., 2016). Itraconazole is classified as a class II drug according to the biopharmaceutical drug disposition classification system; it is highly permeable and undergoes more than 90% metabolism (Benet et al., 2011). Itraconazole and its metabolites exhibit complex DDI mechanisms since all act both as CYP3A4 substrates and inhibitors affecting their own metabolic clearance (CL) and that of other drugs. Three major metabolites are sequentially formed by CYP3A4 metabolism: hydroxy-itraconazole (OH-ITZ); keto-itraconazole (keto-ITZ); and N-desalkyl itraconazole (ND-ITZ) (Supplemental Fig. 1) (Isoherranen et al., 2004).
A physiologically based pharmacokinetic (PBPK) model for itraconazole is clearly beneficial for both DDI risk assessment and optimization of clinical trial design (Rowland et al., 2011; Liu et al., 2016). An advantage with a PBPK model for itraconazole and its metabolites is the possibility of bringing all information into one model with respect to drug-specific factors (e.g., CYP3A4 affinity, CYP3A4 inhibition, distribution, etc.) in addition to physiological factors. This complexity cannot be intuitively captured by simpler models. Simcyp software (Simcyp Ltd., Certara, Sheffield, United Kingdom) is an established PBPK modeling and simulation tool and has been used in this publication to develop the itraconazole PBPK model presented herein (Rostami-Hodjegan et al., 2012). Three different approaches to develop PBPK models have previously been described and discussed: the bottom-up, top-down, and middle-out approaches (Jamei et al., 2009; Rostami-Hodjegan, 2012; Tsamandouras et al., 2015). Briefly, the bottom-up approach is based on in vitro-in vivo extrapolation techniques, which allow mechanistic extrapolation from in vitro and in silico data to human pharmacokinetics (PK). In the top-down approach parameters are optimized with respect to observed clinical data but the predictive utility of such models may be limited by issues such as parameter identifiability and model misspecification. The middle-out approach is a combination of the other two.
Several challenges hinder the development of PBPK models for itraconazole. First, the complex PK with saturable CYP3A4 metabolism and inhibition presents a challenge. All three metabolites are potent CYP3A4 inhibitors and their contribution to clinical DDIs has been considered relevant in a first assessment using static equations (Templeton et al., 2008). Second, there is a lack of robust in vitro data in the literature, which causes uncertainty when building a mechanistic PBPK model based on a bottom-up approach (Chen et al., 2016). Large variabilities in reported in vitro data are mainly due to solubility issues and high protein binding for itraconazole itself as well as its metabolites (Heykants et al., 1989; Isoherranen et al., 2004). The existing itraconazole PBPK library model in Simcyp includes only one metabolite (OH-ITZ) and tends to overpredict the DDI risk (Marsousi et al., 2018). Although Chen et al. (2016) recently published a PBPK model for itraconazole and OH-ITZ using a top-down approach to overcome the challenges of obtaining robust in vitro data from the literature, it is important to continue the efforts of producing robust in vitro data to further develop a more mechanistic PBPK model and this has been emphasized previously for itraconazole (Chen et al., 2016). We believe that further improvements to existing PBPK models are needed, ideally including all metabolites to assess in depth their contribution to the DDI. One limitation with the software selected for our PBPK model is that it is only possible to have two metabolites included. Given the sequential formation of the metabolites, OH-ITZ and keto-ITZ were included in our model and in vitro data for ND-ITZ was generated to be incorporated in future models.
The overall objective of this study is to provide a comprehensive mechanistic PBPK model for itraconazole to increase the confidence in its DDI predictions. To achieve this, key in vitro and in vivo data for itraconazole and its major metabolites were generated, providing improved PK knowledge. These new data were crucial to the development of a bottom-up PBPK model for itraconazole incorporating for first time two of its major metabolites: OH-ITZ and keto-ITZ. Thus, in this paper we provide a comprehensive data set for itraconazole and its three metabolites. This data set enables the development of a bottom-up mechanistic PBPK model that is applicable of assessing clinical itraconazole DDIs and studying the contribution from its metabolites to the DDI.
Materials and Methods
Clinical Data Collection
A total of 11 clinical studies, previously collected by Chen et al. (2016), were used for PK profile verification of itraconazole and OH-ITZ. Two studies included data for keto-ITZ and were used for model verification of this metabolite (Templeton et al., 2008; Liang et al., 2016). Plasma concentration-time profiles and variability were extracted from the figures in the publications. A total of 18 DDI studies using itraconazole as a CYP3A4 inhibitor were used for model validation. Ten clinical DDI studies including midazolam as the substrate have been reported and previously collected by Chen et al. (2016). Eight DDI studies involving other CYP3A4 substrates previously collected by Marsousi et al. (2018) were also used. Different dosing regimens, such as single or multiple dosing with oral solution or capsule formulation of itraconazole were given in these studies. Detailed clinical trial information including original references and the simulated trial designs are described in Tables 1 and 2.
Experimental In Vitro Determination
Materials.
Itraconazole stock solutions in dimethylsulfoxide (DMSO) (10 mM) were obtained from the AstraZeneca Compound Management. OH-ITZ, keto-ITZ, and ND-ITZ were purchased from Toronto Research Chemical (Toronto, Canada). Ultra-Pool human liver microsomes (HLM) (mixed gender, lot no. 38289) were purchased from BD Gentest (San Jose, CA). NADPH and phosphate-buffered saline (pH 7.4) were obtained from Sigma (St. Louis, MO). Human blood from three donors was purchased from Bioreclamation VT (New York City, NY). A rapid equilibrium dialysis device was procured from Thermo Fisher Scientific (Waltham, MA). Midazolam was purchased from International Laboratory (San Francisco, CA). Recombinant CYP3A4 enzymes were acquired from Cypex (Dundee, United Kingdom).
Human and Rat Plasma Protein Binding.
Equilibrium dialysis in human and rat plasma using the rapid equilibrium dialysis device was performed to measure the fraction of unbound drug in plasma (fu,p) in human and rat plasma. First, 300 μl plasma and 500 μl phosphate buffer (pH 7.4) were added to the two separate chambers of the device. A stock solution of compound in DMSO was added into the plasma compartment to yield a final drug concentration of 5 µM and final percent volume of DMSO of 0.5%. The compounds were dialyzed for 18 hours at 37°C while shaking at 500 rpm in an air incubator. After completion of dialysis, plasma and buffer samples were quenched with acetonitrile (ACN), diluted, and analyzed by liquid chromatography–tandem mass spectrometry (LC-MS/MS). Calculations of fu,p in plasma were performed as previously described (Wan and Rehngren, 2006).
Fraction Unbound in HLM Incubations.
Equilibrium dialysis was performed to determine the fraction unbound drug in microsomal incubation (fu,mic) in HLM (1 mg/ml in 0.1 M phosphate buffer, pH 7.4) as previously described (Chen et al., 2017). First, 150 µl of HLM and 150 µl phosphate buffer (0.1 M, pH 7.4) were added to the two separate chambers of the dialysis device. Following 60 minutes of preincubation at 37°C, the compound was added to the HLM chamber. The compounds were dialyzed for 4 hours (37°C, 300 rpm). After completion of dialysis, HLM and buffer samples were quenched with ACN, diluted and analyzed by LC-MS/MS. The DMSO concentration did not reach more than 0.5% of the total incubation volume. Calculations of fu,mic in HLM were calculated as previously described (Wan and Rehngren, 2006).
Substrate Depletion Experiment.
A substrate depletion method was used to identify the enzyme kinetic parameters: the maximum velocity of the metabolic reaction (Vmax) and the Michaelis constant (Km) of itraconazole, OH-ITZ, keto-ITZ, and ND-ITZ. Itraconazole and its metabolites were incubated at 37°C 600 rpm with recombinant CYP3A4 enzymes (15 nM). All experiments were optimized for protein and time to ensure linearity and run at different test occasions (n = 3). Substrates were added to incubations from a 50 nM stock solution in 90% ACN. Nominal substrate concentrations in the final incubation were 5, 10, 20, 40, 80, 160, 320, and 500 nM and the total ACN content in the incubation mixture did not exceed 0.1%. After preincubation for 5 minutes, the reaction was initiated with NADPH to a final concentration of 1 mM. At time points 1, 2, 3.5, 5, 10, and 15 minutes, 15 µl samples were quenched with 30 µl of ice-cold ACN. Then, samples were centrifuged (3220g at 4°C for 30 minutes) and supernatant was transferred and diluted for analysis by LC-MS/MS.
Calculation of Enzyme Kinetics Parameters.
The Km and Vmax values were determined according to the multiple depletions curve method (MDCM) described by Sjögren et al. (2009). The model including enzyme activity change (EAC) was selected (eq. 1). The variable EAC was described as monoexponential decay (MDCM + EAC constant), taking into account potential enzyme degradation and/or inhibition effects during the incubation period:(1)where vmax is the theoretical maximum depletion rate; Km is the substrate concentration at one-half the vmax; [C] is the substrate concentration; and ke is the EAC constant. The parameters, vmax and Km, were estimated by simultaneous fitting of the equation to all concentration-time profiles using the nonlinear mixed effect (NLME) in Phoenix NLME 15 version 6.0 (Pharsight Corporation, A Certara Company, Princenton, NJ). The Vmax value was obtained by dividing vmax by the protein concentration used in the incubation (Cp):
(2)CYP3A4 Inhibition Assay.
The inhibition parameter describing the inhibitor concentration that reduces the enzyme activity by 50% (the IC50 value) for itraconazole, OH-ITZ, keto-ITZ, and ND-ITZ in HLM was measured using 1′-hydroxylation of midazolam as a probe reaction for CYP3A4-mediated metabolism. The incubations were performed with 0.2 mg/ml pooled HLM in 100 mM phosphate buffer (pH 7.4) with 1 mM EDTA and a total concentration of 1 mM NADPH (n = 3). The midazolam concentration was 3 µM. The optimal substrate conditions were previously internally validated. Itraconazole, OH-ITZ, keto-ITZ, and ND-ITZ were added at nominal concentrations of 0, 3, 10, 30, 100, 300, and 1000 nM. The final concentration of DMSO:ACN was 0.3:0.7% v/v. A time-zero sample was collected after preincubation for 5 minutes at 37°C and the reaction was initiated by the addition of NADPH. A second sample was taken after a 5-minute incubation. All samples were quenched with ice-cold ACN (1:1). Then, the samples were centrifuged (3220g for 30 minutes) and supernatant was transferred and diluted for analysis by LC-MS/MS.
The IC50 value was determined by fitting the experimental data to an Imax model (eq. 3) using Phoenix NLME:(3)where E is the effect of inhibition; E0 is the baseline; and C is the concentration of substrate. The constant of inhibition (Ki) was calculated by a classic competitive inhibition model described in eq. 4:(4)where S is the concentration in the incubation and Km (mdz) is the Michalis constant of midazolam (2 µM) (Cer et al., 2009).
Pharmacokinetic Study of Intravenous Administration in Rat.
Male Han Wistar rats (n = 2, ∼300 g) (Charles River, Sulzfeld, Germany), were dosed with keto-ITZ solution (1 mg/kg, 1 ml/kg, bolus) intravenously into the tail vein. The formulation was 5% DMSO and 95% 2-hydroxypropyl-beta-cyclodextrin (30% w/v) in water adjusted to pH 4. Blood samples were collected into EDTA-coated tubes at 2, 7, 15, 30, 60, 120, 240, 360, 420, and 1440 minutes after dose. Urine samples were also collected during the following intervals 0–120, 120–360, and 360–1440 minutes after dose. All samples were centrifuged at 3220g for 5 minutes at 4°C. Blood samples were immediately stored in polypropylene tubes at −80°C and deproteinated by solvent precipitation prior to analysis. Samples were diluted for analysis by LC-MS/MS. The PK parameters were estimated by one-compartmental analyses using Phoenix NLME.
Before the study, the animals were acclimatized for a minimum of 3 days and allowed food and water ad libitum. All of the animal studies were conducted in accordance with the National Institutes of Health guidelines on animal welfare.
Model Acceptance Criterion for PK Verification and DDI Validation
The following criteria were predetermined to assess model performance. First, the performance of the model in describing the PK profiles of itraconazole, OH-ITZ, and keto-ITZ was verified if the observed concentration-time profiles were within the 90% prediction interval (5th to 95th percentile range of the virtual population). Following the verification of the PK model for itraconazole and its metabolites, simulations were performed to test the capability of the model to accurately describe DDI studies with itraconazole and midazolam. The geometric mean area under the plasma concentration- time curve (AUC) ratio for each DDI study was determined and the predicted and observed data were compared (see Table 2 for results). Precision and bias for the DDI predictions were evaluated using the geometric mean fold error (GMFE) described in eq. 5:(5)To assess the predictive performance and ensure no bias and good precision of DDI, the predicted/observed ratio for AUC and Cmax should be within a predefined criterion of a 2-fold range and the GMFE should be between 0.85 and 1.25. To further strengthen the final model predictive performance, an additional validation including other CYP3A4 substrates was also performed. The details of the studies used are listed in Table 2.
PBPK Parameter Input and Model Development
A full PBPK model was constructed for itraconazole and its metabolites (OH-ITZ and keto-ITZ) using mainly a bottom-up approach (Jamei et al., 2009) in Simcyp. Key parameters (i.e., fu,p, fu,mic), enzyme kinetics, and CYP3A4 inhibition were redefined using the in vitro data presented in this paper (Table 3). The overall model development, verification, and validation process is presented in Fig. 1.
The physicochemical properties for itraconazole and OH-ITZ were used as described in the compound library files. The fu,p value for itraconazole was set to 0.01 according to US Food and Drug Administration guidelines. For keto-ITZ, where no previous model was available, the physicochemical parameters were calculated from its chemical structure using the Biobyte software, version 4.3 (Pomona College and BioByte, Inc., Claremont, CA). The blood-to-plasma ratio and pKa were assumed to be the same as the parent. All values are listed in Table 3.
The first-order absorption model for itraconazole was used keeping the fraction absorbed and the absorption constant as in the software library. The values were adjusted when clinical studies with solutions or capsule formulations were simulated (Table 3). A nominal flow in the gut model was predicted using the Simcyp toolbox and the fraction unbound drug in the gut was assumed to be the same as fu,p in plasma.
For the volume of distribution (Vss) parameter, a full PBPK model was used to reach predefined values for itraconazole and OH-ITZ. Method 1, described originally by Poulin and Theil (2009), was selected for prediction of the tissue-to-plasma partition coefficients for individual tissues. The Vss value for itraconazole was taken from Mouton et al. (2006). The Vss values for OH-ITZ and keto-ITZ were scaled from rat (Yoo et al., 2000) to human using the Oie-Tozer method (Oie and Tozer, 1979). A minimal PBPK model was used for keto-ITZ since a full PBPK model is not currently available for the second metabolite in the software. The addition of a single compartment was used to describe the two-compartmental behavior of the plasma PK profiles of keto-ITZ. The single adjustable compartment (SAC) parameters, the SAC volume, first-order rate constant for the distribution to a SAC, and first-order rate constant for the distribution from a SAC, were determined by best fit using data set 1 on day 1. The parameters used in the model are listed in Table 3.
The hepatic CL for itraconazole and its metabolites was scaled in the model from the CYP3A4 enzyme kinetic parameters Km and Vmax that were determined by the MDCM method described previously. The renal CL (CLR) for keto-ITZ was scaled from rat using the Simcyp toolbox. In the case of keto-ITZ, CYP3A4 metabolism and CLR were not sufficient to describe the observed CL in humans. In addition, unspecific CL was included for this compound to capture a potential alternative route of elimination. The additional CL was added using the retrograde method and the CL determined in rats and scaled to humans was used as the starting value.
The fu,mic determined by dialysis in microsomal incubation mixtures was incorporated into each model to correct for protein binding in the in vitro experiments. The fu,mic value was corrected for the protein concentrations used in each assay (Table 3). Specifically, for enzyme kinetics parameters (fu,mic1), the method described by Wan and Rehngren (2006) was used. For inhibition parameters (fu,mic2), where protein concentration was similar to measured fu,mic conditions, no correction was deemed necessary based on the binding profile of OH-ITZ (Tran et al., 2002).
Simulations Using PBPK Modeling
The Simcyp (version 15) population-based PBPK simulator (Simcyp Ltd., Certara) was used to simulate the PK of itraconazole and its metabolites (OH-ITZ and keto-ITZ), and relevant DDI in virtual healthy volunteers. Simulations were performed with randomly selected individuals from a simulated healthy volunteer population built in the Simcyp software (Sim-Healthy). A total of 100 individuals—10 trials with 10 individuals per trial—were simulated to assess variability across groups. The age, sex ratio, dose, formulation, and regimen used in the simulations were matched to the clinical studies described in each trial (Tables 1 and 2). The CYP3A4 substrate models available in the Simcyp compound library were directly used in the simulations. When the DDI simulations were performed, the itraconazole compound file including the two metabolites were put as the substrate file and output was exported to Phoenix. The AUCs from each simulation were calculated by noncompartmental analyses and the log/linear trapezoidal method. The relative contribution of itraconazole and its metabolites (OH-ITZ and keto-ITZ) to the total AUC ratio was calculated following the approach of Wang et al. (2004), where net inhibition is related to the sum of inhibitory contributions of all circulating inhibitors as described in Templeton et al. (2008). Simulated unbound liver and gut concentrations and the respective apparent Ki,u for itraconazole and its metabolites were used in the calculation. The simulated concentrations were taken from a dosing regimen of once daily administration for 4 days of 200 mg itraconazole (capsule and oral solution) given in the fasted state.
Model Parameter Sensitivity Analysis
Sensitivity analysis in this context is a description of how sensitive the model is to changes in the model parameters. Key parameters were evaluated in this sensitivity analysis to gain additional insight on their impact on the AUC of itraconazole and its metabolites. The sensitivity analysis was done using two different methods: The one-factor-at-a-time approach and the sensitivity index (SI) approach (Nestorov, 1999; Bonate, 2011). In the one-factor-at-a-time approach, the numerical value of a parameter is varied within a specified region around the estimated optimal parameter value while the change in AUC is observed. The SI for each parameter was calculated using eq. 6:(6)where AUCmax and AUCmin are the maximum and minimum AUCs, respectively, within the explored parameter space in the sensitivity analysis. The advantage of the SI compared with the one-factor-at-a-time methods is that a direct comparison of the estimated sensitivity of the model parameters is possible.
Postanalysis of the Simulated Plasma Profiles
To further assess the performance of the model on simulating PK profiles in a quantitative manner, a postanalysis was conducted according to previously established methods (Marston and Polli, 1997). The difference factor (f1), which is a model-independent parameter, was applied for the comparison of the plasma concentration-time profiles of itraconazole, OH-ITZ, and keto-ITZ according to eq. 7:(7)where n is the number of time points, and Rt and Tt are the plasma drug concentrations observed and simulated, respectively, at each time point t.
Results
Experimental In Vitro Determination
Human and Rat Plasma Protein Binding.
Human and rat plasma protein binding was assessed using equilibrium dialysis. All test compounds were highly bound to plasma proteins in human and rat resulting in low fu,p values (0.001–0.029), which are presented in Table 4.
Fraction Unbound in HLM Incubations.
Protein binding was also determined in HLM incubation mixtures by dialysis for itraconazole, OH-ITZ, keto-ITZ, and ND-ITZ. All of the test compounds were extensively bound and this resulted in overall low fu,mic values (0.01–0.076). The details are presented in Table 4.
Enzyme Kinetics Analysis.
The CYP3A4 enzyme kinetic parameters (Vmax and Km) for itraconazole, OH-ITZ, keto-ITZ, and ND-ITZ were estimated using the MDCM method. A summary of the resulting parameters is presented in Table 4. All concentration-time profiles and model fits for itraconazole and its three metabolites are presented in Fig. 2. Biphasic depletion curves were observed for all four substrates, and therefore MDCM with EAC was used to take into account potential enzyme depletion or inhibition. All four compounds had low unbound Km values, indicating high affinity to CYP3A4.
CYP3A4 Inhibition.
The potency of the CYP3A4 inhibition for itraconazole, OH-ITZ, keto-ITZ, and ND-ITZ was determined in HLM using midazolam as the substrate. A summary of the resulting parameters (IC50 and Ki) is presented in Table 4. The fitted IC50 curves versus a range of inhibitor concentrations are presented in Fig. 3A. In addition, the simulated IC50 curves with remaining midazolam (%) versus calculated free concentrations are presented in Fig. 3B.
Rat PK.
Rats were intravenously dosed with keto-ITZ with the purpose of obtaining Vss, CL, and CLR for this metabolite. The plasma and urine concentration-time curves for the individual animals (Supplemental Fig. 2) and the PK parameters (Vss, CL, and CLR) calculated by noncompartmental analysis are presented in Table 4.
PBPK Model Development and Verification
The PBPK model for itraconazole, OH-ITZ, and keto-ITZ (Table 3) was first developed and optimized using data set 1 (Table 1) as a training data set in a step-wise manner. The simulated plasma concentration profiles of the training data set met the predefined model acceptance criteria (observed plasma concentration falling within the 90% prediction interval) (Fig. 4, A, C, and E). Furthermore, the model was verified using a separate test data set of 10 human PK studies including observations at a wide range of time points varying from 1 day up to 4 weeks across a multiple dosings regimen, both for solution and capsule formulation at different doses, and under different fed and fasted conditions. Similar to the training data set, all of the simulated concentration profiles (Fig. 4, B, D, and F; Fig. 6) for the verification studies fell within the acceptance criteria. Accumulation over days was reasonablly well captured by the model for all three compounds under all of the different conditions (Figs. 4 and 5).
Model Parameter Sensitivity Analysis
To gain additional insights into this complex model of a parent compound and two sequential metabolites in which all are substrates and inhibitors of CYP3A4, a sensitivity analysis was conducted to assess the relative importance of the key parameters toward changes in itraconazole, OH-ITZ, and keto-ITZ AUC. In general, the AUCs for all three compounds are more sensitive to elimination kinetics parameters and plasma protein binding than to inhibition constants (Fig. 6). The parameters in Fig. 6, A–C, which have an index above 0.75, have the greatest effect on the AUC of itraconazole and its metabolites. Few of the tested parameters were below SI 0.25, indicating that all have considerable impact on the AUC of itraconazole and its metabolites.
Postanalysis of the Simulated Plasma Profiles
To further quantify the accuracy of the model in predicting the PK profiles for itraconazole and its metabolites, the f1 (deviation observed vs. predicted at identical time points) was calculated. Overall, the PK profiles were predicted adequately, exhibiting f1 of 43%, 30%, and 52% for itraconazole, OH-ITZ, and keto-ITZ, respectively. In Fig. 7 the predicted versus observed concentrations at the same time points are presented.
Model Validation: Prediction of Itraconazole DDI
The results of the simulated clinical trials including midazolam and itraconazole are summarized in Table 2 and Fig. 8. Overall, the model meets the prespecified criteria predicting 100% of observed midazolam AUC and Cmax ratios within 2-fold (Fig. 8A). In addition, 70% of the simulated midazolam AUC ratios were within 1.5-fold of the observed data. To further strengthen the DDI validation, other CYP3A4 substrates were also evaluated, confirming the good overall prediction within 2-fold (Fig. 8B; Table 2). The GMFE values were determined to be 1.06 and 0.96 for the AUC and Cmax ratios, respectively. The GMFE values indicate good precision and no bias when predicting the observed DDI.
Model Application: Relative Contribution to the AUC Ratio
The predicted time course of CYP3A4 inhibition after itraconazole dosing using different formulations was simulated (Supplemental Figs. 3 and 4). The figures illustrate the relationship between simulated unbound liver and gut concentrations and the respective unbound apparent Ki for itraconazole and its metabolites. The corresponding liver average unbound concentration (Cuss)/Ki,u values (over 96 hours) for itraconazole, OH-ITZ, and keto-ITZ for this dose are 85, 31, and 0.7 for capsule formulation and 200, 62, and 0.85 for oral solution formulation. To gain further insights, the relative contributions of itraconazole and its metabolites to the AUC ratio were calculated and are presented in Fig. 9. In the first hours after dose the main contributor is itraconazole (>80%), but as time passes its contribution equals out to a similar level as OH-ITZ (∼55% vs. 40%). The relative role of OH-ITZ increases with time after itraconazole dosing. The contribution of keto-ITZ is minor throughout the studied 8 day period compared with itraconazole and OH-ITZ (<5%).
Discussion
In the present work, an accurate PBPK model was developed that simultaneously predicts the PK profiles for itraconazole, OH-ITZ, and keto-ITZ. The performance of the model was successfully verified against several PK (n = 11) and DDI (n = 18) studies, which included different dosing regimens, formulations, and CYP3A4 substrates (Tables 1 and 2). This study is the first to generate in vivo PK data for keto-ITZ as well as in vitro PK data for itraconazole and its three major metabolites, enabling the necessary scaling PK parameters for building a PBPK model. The strengths of this itraconazole PBPK model compared with previous published are the following: 1) the model includes the second major metabolite of itraconazole, keto-ITZ; 2) the model is built primarily using a bottom-up approach including robust data for inhibition, elimination, and binding parameters; 3) the model includes a sensitivity analysis, which aims to highlight the impact and importance of the parameters included; 4) the model contains a quantitative assessment of model accuracy to predict PK profiles; 5) the model is validated against several CYP3A4 substrates; and 6) the model provides insights into the relative contribution of itraconazole metabolites to the clinical DDI.
The first and most critical step when using PBPK simulations is to accurately predict the PK of the inhibitor and substrate drugs. The model presented meets the prespecified acceptance criteria, with the majority of observed plasma drug concentrations being within the 90% prediction interval (Figs. 4 and 5). Here, the criteria were chosen on the basis of PK variability of itraconazole and its metabolites derived from its complex PK and bioavailability (Poirier and Cheymol, 1998). Being a drug with a broad therapeutic window (Buchkowsky et al., 2005), more flexible criteria are considered acceptable (Jones et al., 2015). There is a lack of good standardization of model acceptance criteria in PBPK modeling (Sager et al., 2015). Herein, one of the methods highlighted in this review has been used for retrospective analysis together with goodness-of-fit plots following best-practice examples (Wagner et al., 2012; Gertz et al., 2013). The calculation of f1 considers all of the observed data points and is a direct comparison of the predicted value at the specific time point (Fig. 7). The f1 calculations confirm the accuracy in PK predictions for itraconazole (43%) and OH-ITZ (30%) and indicate that the prediction of the PK profile of keto-ITZ is less accurate (53%), following the criteria set by Sjögren et al. (2013). As shown in Fig. 7, precision could be improved for the lower concentrations. On the other hand, high model performance was observed for Cmax and steady-state concentrations of itraconazole (31%) and OH-ITZ (21%).
The second step in the present work was to use the validated PK model to assess how well the model could predict AUC and Cmax ratios for CYP3A4 substrates with and without the presence of an inhibitor. In general, the predicted versus observed ratios were within 2-fold (Fig. 8B) with good precision and no bias in the DDI prediction showing GMFE values close to 1. There are some specific scenarios where predictions could be further improved, which could also be observed in the Chen et al. (2016) model when the same studies were simulated. For example, when the itraconazole oral dose reached 400 mg an inhibition plateau was observed on day 1 (Templeton et al., 2010). The model overpredicted the AUC ratio and did not capture the plateau (data set 21). To date, the mechanism behind this plateau is not understood and more clinical studies using the 400 mg itraconazole dose would help to clarify this finding. Another example is the general trend to underpredict the DDI in scenarios where the substrate is given more than 12 hours after itraconazole dose (data sets 18, 26, and 27); this could be due to the unrecognized contribution of the last metabolite, ND-ITZ. Nevertheless, in the case of itraconazole and clinical study designs for DDI, the most critical data are the steady-state prediction, where our model showed more accurate performance with 89% of these studies within the more strict criteria of 1.5-fold error (i.e., trials 12 to 13, 15–17, 22 to 23, and 28 to 29 in Table 2).
The role of metabolites in DDI is a developing area of research. The presented in vitro results confirm that all three metabolites are potent inhibitors of CYP3A4 with unbound IC50 values in the nanomolar range, comparable to itraconazole IC50 (Fig. 3; Table 4). When the ratio between circulating metabolite concentrations and Ki is higher than 0.1, PBPK modeling is recommended (Callegari et al., 2013). All itraconazole metabolites were predicted to have values of the the ratio between circulating metabolite concentrations and Ki above 0.1 and to significantly contribute to the observed clinical DDIs by Templeton et al. (2008). Therefore, it can be considered highly relevant to build a PBPK model that includes all metabolites, and in this paper we have presented the first step that includes the two first metabolites that are sequentially formed: OH-ITZ and keto-ITZ. The current model enables simulations for hypothesis testing to gain a deeper understanding of the contribution of the metabolites to the observed DDI (Fig. 9). The relative contribution depends on the time window observed after dosing; therefore, this model also enables simulations of the metabolite profiles over time with different trial designs. The relative role of OH-ITZ is increased with time after itraconazole dosing, being similarly important to itraconazole after 12 hours and having even higher contribution after day 3. However, the contribution of keto-ITZ is minor (<5%) over time. Hence, this is the first study by PBPK modeling that assesses in detail the contribution of keto-ITZ and establishes the low impact to the overall inhibition. Nevertheless, the inclusion of this metabolite in the model is indispensable to allow further development including ND-ITZ, which is expected to play a more significant role in the inhibition 12 hours or more following the last dose of itraconazole. Given the long half-life, lower protein binding, and potency of ND-ITZ, it can be predicted that its contribution to the observed DDI is increasing with time (similar to OH-ITZ) and this will be crucial when the CYP3A4 substrate is given more than 12 hours or days after the last dose of itraconazole. Thus, the next step will be to also include the third metabolite (ND-ITZ) to further improve the DDI prediction in those scenarios.
Previous reports (Chen et al., 2016; Liu et al., 2016) have emphasized discrepancies in reported in vitro data. For example, a 30-fold range of values has been reported for fu,p in plasma for the parent compound (Heykants et al., 1989; Arredondo et al., 1995, 1999; Ishigam et al., 2001; Templeton et al., 2008). A second example is the reported Ki or IC50 values that vary between 150- and 170-fold for itraconazole and OH-ITZ, respectively (Back and Tjia, 1991; von Moltke et al., 1996; Wang et al., 1999; Ishigam et al., 2001; Tran et al., 2002; Isoherranen et al., 2004). As mentioned previously, high-quality in vitro data are critical when building a mechanistic bottom-up PBPK model (Jamei et al., 2009), and this is emphasized as well by our sensitivity analysis results (Fig. 6). Therefore, key parameters were experimentally determined in the same laboratory at the same occasion for the parent and the metabolites. In this report, data are generally presented with good precision on different experimental days (CV % < 30), increasing the confidence in our data. At the same time, the vitro-in vivo extrapolation and modeling done validate the in vitro data herein generated since the observed clinical PK and DDI were recovered for itraconazole and its metabolites. In the new draft US Food and Drug Administration in vitro DDI guideline it is recommended to use a fu,p value of 0.01 for mechanistic modeling even though lower values have been experimentally determined (CDER, 2017b). The reason is the high uncertainty with lower measured values for highly bound compounds. Given the uncertainties of the in vitro measured plasma protein binding, the value recommended by US Food and Drug Administration was used.
The scope for this study was to mechanistically describe the most critical processes for assessing the CYP3A4 inhibition of itraconazole and its metabolites by vitro-in vivo extrapolation, but it goes without saying that other questions remain to be studied. The contribution of the third metabolite, ND-ITZ, to DDI has been suggested as being clinically relevant (Templeton et al., 2008), but our model does not include it due to limitations of the software. However, in vitro data for ND-ITZ were generated during this study to enable future model development. There are conflicting data on hepatic uptake for itraconazole (Yamano et al., 1999; Higgins et al., 2014). In the current model hepatic uptake was set to 1 following the most recent publication, which showed lack of hepatic uptake in in vitro human primary hepatocytes and knockout mice. However, it could be beneficial to further investigate this to clarify the possibility that carrier-mediated transport might be involved. It is also important to remember that itraconazole and its metabolites are known inhibitors of P-gp and other transporters (Vermeer et al., 2016). The absorption of itraconazole is currently described by first-order kinetics. This limits the simulations on the potential regional differences in the inhibition of intestinal CYP3A4 and transporters. Expansion to a multicompartment gut model is something that could be evaluated and possibly included in future model versions to further improve the mechanistic behavior of the model.
In this paper, a model is presented in which we have successfully included the metabolite keto-ITZ into a PBPK model for itraconazole PK that enables DDI simulations. We believe that this model provides improved mechanistic understanding of the PK and DDI of ITZ and its metabolites. The results presented and sensitivity analyses highlight the importance of having robust in vitro and in vivo data to enable complex model building. The predictive DDI risk capability of this model is improved compared with the Simcyp itraconazole library model (100% vs. 80% predicted within 2-fold), showing no bias and good precision. Therefore, our observations suggest that this novel PBPK model built for itraconazole and two of its main metabolites can be successfully used to both evaluate DDI involving new victim compounds and to facilitate optimal study design.
Acknowledgments
We thank Dr. Martin Hayes at AstraZeneca for reviewing the manuscript. We also thank Danxi Li and Hongwen Du at Pharmaron and the Wave1 Drug Metabolism and Pharmacokinetics Department at AstraZeneca for providing experimental assistance.
Authorship Contributions
Participated in research design: Prieto Garcia, Kanebratt, Ericsson, Lennernäs, Lundahl.
Conducted experiments: Prieto Garcia.
Performed data analysis: Prieto Garcia, Janzén, Lundahl.
Wrote or contributed to the writing of the manuscript: Prieto Garcia, Janzén, Kanebratt, Ericsson, Lennernäs, Lundahl.
Footnotes
- Received March 14, 2018.
- Accepted July 27, 2018.
↵This article has supplemental material available at dmd.aspetjournals.org.
Abbreviations
- ACN
- acetonitrile
- AUC
- area under the plasma concentration-time curve
- CL
- clearance
- CLR
- renal clearance
- DDI
- drug-drug interaction
- DMSO
- dimethylsulfoxide
- EAC
- enzyme activity change
- f1
- difference factor
- fu,mic
- fraction of unbound drug in microsomal incubation
- fu,p
- fraction of unbound drug in plasma
- GMFE
- geometric mean fold error
- HLM
- human liver microsomes
- keto-ITZ
- keto-itraconazole
- Ki
- constant of inhibition
- Km
- Michaelis constant
- LC-MS/MS
- liquid chromatography–tandem mass spectrometry
- MDCM
- multiple depletions curve method
- ND-ITZ
- N-desalkyl itraconazole
- NLME
- nonlinear mixed effect
- OH-ITZ
- hydroxy-itraconazole
- PBPK
- physiologically based pharmacokinetic
- PK
- pharmacokinetics
- SAC
- single adjustable compartment
- SI
- sensitivity index
- vmax
- theoretical maximum depletion rate
- Vss
- volume of distribution
- Copyright © 2018 by The American Society for Pharmacology and Experimental Therapeutics