Advertisement
Research ArticleMetabolism, Transport, and Pharmacogenomics

Physiologically Based Pharmacokinetic Model of All-trans-Retinoic Acid with Application to Cancer Populations and Drug Interactions

Jing Jing, Cara Nelson, Jisun Paik, Yoshiyuki Shirasaka, John K. Amory and Nina Isoherranen
Journal of Pharmacology and Experimental Therapeutics May 2017, 361 (2) 246-258; DOI: https://doi.org/10.1124/jpet.117.240523

Abstract

All-trans retinoic acid (atRA) is a front-line treatment of acute promyelocytic leukemia (APL). Due to its activity in regulating the cell cycle, it has also been evaluated for the treatment of other cancers. However, the efficacy of atRA has been limited by atRA inducing its own metabolism during therapy, resulting in a decrease of atRA exposure during continuous dosing. Frequent relapse occurs in patients receiving atRA monotherapy. In an attempt to combat therapy resistance, inhibitors of atRA metabolism have been developed. Of these, ketoconazole and liarozole have shown some benefits, but their usage is limited by side effects and low potency toward the cytochrome P450 26A1 isoform (CYP26A1), the main atRA hydroxylase. We determined the pharmacokinetic basis of therapy resistance to atRA and tested whether the complex disposition kinetics of atRA could be predicted in healthy subjects and in cancer patients in the presence and absence of inhibitors of atRA metabolism using physiologically based pharmacokinetic (PBPK) modeling. A PBPK model of atRA disposition was developed and verified in healthy individuals and in cancer patients. The population-based PBPK model of atRA disposition incorporated saturable metabolic clearance of atRA, induction of CYP26A1 by atRA, and the absorption and distribution kinetics of atRA. It accurately predicted the changes in atRA exposure after continuous dosing and when coadministered with ketoconazole and liarozole. The developed model will be useful in interpretation of atRA disposition and efficacy, design of novel dosing strategies, and development of next-generation atRA metabolism inhibitors.

Introduction

All-trans retinoic acid (atRA) is an active metabolite of vitamin A (retinol), circulating at endogenous plasma concentration of about 2 nM in healthy humans (Jing et al., 2016). One of the main biologic activities of atRA is regulating the cell cycle by binding to the nuclear retinoic acid receptors (RARs). Via RAR activation, atRA controls the expression of target genes involved in differentiation, cell cycle arrest, and apoptosis in a concentration-dependent manner.

Because of its activity in inducing apoptosis, atRA has been pursued for the treatment of various cancers including non-small-cell lung cancer, head and neck cancer, astrocytoma, and Kaposi’s sarcoma (Phuphanich et al., 1997; Saiag et al., 1998; Arrieta et al., 2010; Lim et al., 2012). In acute promyelocytic leukemia (APL), promyelocyte differentiation is blocked due to the formation of promyelocytic leukemia-RARα fusion protein caused by the chromosomal translocation between chromosomes 15 and 17. Therapeutic doses of atRA resulting in a peak plasma concentration of 1 μM (Adamson, 1996) can overcome the blockage and trigger RAR-mediated promyelocyte differentiation (Ablain and de The, 2011). As monotherapy, atRA induces clinical remissions in 70%–90% of APL patients, but relapses are observed if atRA is continued as a single agent, so combination therapies are required for long-term disease control (Coombs et al., 2015). The loss of atRA efficacy is believed to be largely due to the 42%–62% reduction of plasma concentrations of atRA upon chronic dosing caused by autoinduction of atRA clearance (Muindi et al., 1992a; Regazzi et al., 1997; Russo et al., 1998; Ozpolat et al., 2003). This induction is likely due to increased expression of the cytochrome P450 isoform 26A1 (CYP26A1), the main atRA hydroxylase in human liver (Tay et al., 2010; Topletz et al., 2015).

To overcome the autoinduction of atRA clearance and increase systemic atRA concentrations, inhibitors of atRA metabolism have been developed. The decline of atRA area under the plasma concentration–time curve (AUC) after continuous oral treatment was attenuated by ketoconazole, a pan-CYP inhibitor, in advanced lung cancer patients (Rigas et al., 1993). Liarozole, an inhibitor of atRA metabolism, was found to partially reverse the decline of atRA AUC after chronic administration to patients with solid tumors (Miller et al., 1994).

Despite these promising clinical results, the use of atRA metabolism inhibitors in cancer therapy has been limited due to the adverse effects and lack of satisfactory potency and selectivity for the primary atRA metabolizing enzymes (Njar et al., 2006; Nelson et al., 2013). To develop more potent and selective inhibitors and efficient dosing strategies, a better understanding of atRA disposition and kinetics of induction of atRA metabolizing enzymes is needed.

CYP26A1 is predicted to be responsible for more than 90% of hepatic atRA clearance, although CYP2C8, CYP3A4, and CYP3A5 also metabolize atRA (Thatcher et al., 2010). The induction of atRA metabolism upon multiple dosing is likely caused by increased CYP26A1 enzyme expression because atRA treatment significantly induces CYP26A1 mRNA and activity in HepG2 cells and human hepatocytes (Tay et al., 2010; Topletz et al., 2015). However, the quantitative findings of CYP26A1 activity, atRA concentration-dependent induction of atRA metabolism, and the magnitude of pharmacokinetic changes in different populations receiving atRA have not been integrated into a model of atRA disposition in humans that could be applied to the design of better inhibitors of atRA metabolism or improved dosing regimens in atRA therapy. We hypothesized that physiologically based pharmacokinetic (PBPK) modeling and simulation could be used as a mechanistic tool to comprehensively understand the dose- and time-dependent disposition of atRA in different populations in the presence and absence of inhibitors of atRA metabolism.

PBPK modeling, which integrates physiologic parameters such as blood flow, tissue sizes, and composition with drug-specific parameters, is widely applied to simulate drug disposition in plasma and tissues (Jones et al., 2015). Uniquely altered physiologic parameters in different populations and disease states can be integrated into PBPK models to investigate variable drug disposition in humans. PBPK modeling also provides the possibility to simulate population variability and time-varying phenomena, including induction and saturation of clearance pathways. Therefore, PBPK modeling is exceptionally well suited to study atRA disposition. Surprisingly, existing atRA PBPK models (Clewell et al., 1997; Louisse et al., 2015) have not incorporated the induction of atRA metabolism or multiple dosing scenarios, or been applied to any disease population. We developed a full atRA PBPK model incorporating human physiology and atRA specific pharmacokinetic parameters to predict atRA disposition in humans after single and multiple doses of atRA in the absence and presence of atRA metabolism inhibitors in both healthy subjects and in cancer patients.

Materials and Methods

Chemicals and Reagents.

Ketoconazole, atRA, and acitretin were purchased from Sigma-Aldrich (St. Louis, MO). atRA-d5 was purchased from Santa Cruz Biotechnology (Santa Cruz, CA). We purchased 4OH-atRA and 4oxo-atRA-d3 from Toronto Research Chemicals (North York, ON, Canada). Liarozole was purchased from Tocris Bioscience (Bristol, United Kingdom). CYP3A4 and CYP2C8 supersomes coexpressed with reductase and b5 were purchased from BD Gentest (BD Biosciences, San Jose, CA). NADPH and high-pressure liquid chromatography-grade ethyl acetate were purchased from EMD Millipore (Billerica, MA). CYP26A1 was expressed in Sf9 insect cells and microsomes containing CYP26A1 were prepared as described elsewhere (Lutz et al., 2009). CYP26A1 content was measured by CO spectrum. Rat cytochrome P450 oxidoreductase was expressed in Escherichia coli and purified as described elsewhere (Lutz et al., 2009). Liquid chromatography with tandem mass spectrometry (LC-MS/MS) mobile phase solvents (water, acetonitrile, and methanol) were Optima grade and purchased from Fisher Scientific (Pittsburgh, PA).

atRA and 4OH-atRA Quantification.

All analyses were done using AB Sciex 5500 qTrap Q-LIT mass spectrometer (Foster City, CA) equipped with an Agilent 1290 UHPLC (Santa Clara, CA) using validated analytic methods described elsewhere (Arnold et al., 2012, 2015; Jing et al., 2016). Sample preparation methods are described here for each type of sample. For the analysis of atRA-d5 and atRA concentrations in mouse serum and tissue samples, human blood and plasma samples, Caco-2 cell permeability assay, and atRA depletion in CYP26A1 inhibition assay, atRA-d5 and atRA were separated using an Ascentis Express RP-Amide column (2.7 μm, 15 cm × 2.1 mm; Sigma-Aldrich). A gradient elution with aqueous (A) and acetonitrile (B) with 0.1% formic acid and 40% methanol in both A and B was used. The gradient was from initial 60% A for 2 minutes to 45% A over 8 minutes, then to 10% A over 7 minutes, and finally held at 5% A for 3 minutes before re-equilibration. Mobile phase flow was 0.5 ml/min. Analytes were detected using positive ion atmospheric-pressure chemical ionization mode. MS/MS transitions for atRA, atRA-d5 (internal standard for atRA assay), and acitretin (internal standard for atRA-d5 assay) were m/z 301 > 205, m/z 306.1 > 127.2, and m/z 327.0 > 77, respectively. The declustering potential, collision energy, and collision exit potential were 80, 17, and 10 for atRA, 46, 23, and 8 for atRA-d5 and 46, 77, and 14 for acitretin. A minimum of six quality control samples prepared in the relevant matrix were included in each LC-MS/MS run.

The formation of atRA metabolite 4OH-atRA in CYP2C8 and CYP3A4 inhibition assays was measured as described elsewhere (Topletz et al., 2015). The 4OH-atRA was separated using an Agilent Zorbax Extend C18 column (3.5 μm, 2.1 × 100 mm; Agilent Technologies) with a gradient elution at a flow rate of 0.3 ml/min. The gradient was from initial 95% of aqueous with 0.1% formic acid (A) and 5% of acetonitrile (B) to 5% A over 5.5 minutes, then held for 1 minute before re-equilibration to initial conditions. Analytes were detected using positive ion electrospray ionization mode. The declustering potential, collision energy, and collision exit potential were set to 80, 30, and 13 for 4OH-atRA and 80, 35, and 2 for 4oxo-atRA-d3 (internal standard). The tandem mass spectrometry transitions for 4OH-atRA and 4oxo-atRA-d3 were m/z 299.2 > 197.2 and m/z 300.0 > 226.0, respectively.

atRA Blood to Plasma Ratio.

The blood to plasma ratio of atRA was measured in fresh human blood collected into EDTA-containing tubes to avoid clotting. We spiked atRA-d5 into 600-µl aliquots of blood in triplicate with 25 nM final concentration. Samples were mixed and incubated at 37°C for 2 hours. After incubation, 120 µl of whole blood was removed, and protein was precipitated with 120 µl acetonitrile. The remaining sample was centrifuged at 16,100g for 5 minutes at room temperature to pellet red blood cells and isolate plasma. Then 120 µl of acetonitrile was added to 120 µl of plasma to precipitate plasma proteins. We added 2 µl of 10 µM acitretin to all samples as the internal standard. Samples were centrifuged at 3000g for 15 minutes at 4°C, 100 µl of supernatant was transferred to glass vials, and samples were analyzed using LC-MS/MS as described earlier.

atRA Permeability in Caco-2 Cells.

Caco-2 cells were cultured at 37°C in a humidified atmosphere of 5% CO2 using Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum, 1% l-glutamine, 1% nonessential amino acids, and 5% antibiotic-antimycotic solution. Cells were routinely subcultured at 90% confluency with trypsin-EDTA. Caco-2 cells were plated onto cell culture inserts (3.0 μm pores, 0.9 cm2 growth area) at a density of 6.4 × 104 cells/insert. The culture medium (0.8 ml in the insert and 2.0 ml in the well) was replaced with fresh medium at day 5 after initiation of cell culture and every 48 hours thereafter.

After 21 days in culture, the Caco-2 monolayer was used for the permeability experiments. Cell monolayers were preincubated in transport medium (Hank’s balanced salt solution; 0.952 mM CaCl2, 5.36 mM KCl, 0.441 mM KH2PO4, 0.812 mM MgSO4, 136.7 mM NaCl, 0.385 mM Na2HPO4, 25 mM d-glucose, and 10 mM HEPES, pH 7.4) for 30 minutes at 37°C. After preincubation, the transepithelial electrical resistance (TEER) was measured routinely with a Millicell-ERS system (Millipore, Bedford, MA) to ensure cell monolayer integrity. The cell monolayers that exhibited TEER values higher than 500 Ω·cm2 were used for experiments.

Permeability measurement was initiated by adding test compounds (10 μM) to the donor side and transport medium to the receiver side. Transfer of test compounds was observed in two directions, apical to basolateral and basolateral to apical. Samples were obtained from the donor side at 5 minutes for measurement of initial concentration and from the receiver side at 30, 60, 90, 120, and 180 minutes. Permeability experiments were performed under no pH gradient condition (apical pH = basal pH = 7.4) at 37°C. After all experiments, TEER was measured to ensure cell monolayer integrity, and data were generated in cell monolayers in which viability had not been adversely affected by the experimental conditions were accepted.

The apparent permeability (Papp, cm/s) of test compounds across cell monolayers was calculated using eq. 1:Embedded Image(1)where Q is the amount of compound transported over time t (therefore, dQ/dt is the amount of compound transported within a given time period [μmol/s]). CD is the initial concentration of compound in the donor compartment (μM), and A is the membrane surface area (cm2). The concentration of atRA was analyzed by LC-MS/MS as described earlier. For all transport assays, atenolol and metoprolol were used as controls, and the LC-MS/MS methods for these compounds are described in supplemental data.

Tissue Distribution of atRA-d5 in C57BL/6J Mice.

Disposition of atRA was characterized and tissue to plasma partition coefficients (Kp) were determined in C57BL/6J mice. Animal experiments were approved by the University of Washington’s Institutional Animal Care and Use Committee. For the study, 12 male C57BL/6J mice were dosed with 1 mg/kg atRA-d5 i.p., and blood, liver, kidney, and pancreas were collected at 0.5, 1, 2, and 4 hours after dosing under yellow light. Three mice were sacrificed at each time point. In addition, three mice were dosed with vehicle and sacrificed immediately after vehicle administration. At each time point serum was separated from blood by centrifugation, samples were protected from light, and each tissue sample was snap frozen in liquid nitrogen and stored in −80°C until analysis.

Tissue homogenization and atRA-d5 extraction were performed as described elsewhere (Arnold et al., 2015). In brief, 100 mg of tissue per sample was homogenized in a 1:1 volume of 0.9% NaCl, and atRA-d5 was extracted with 10 ml of hexanes after addition of 10 µM acitretin as an internal standard. After evaporation of the organic layer under nitrogen, the sample was reconstituted with 65 μl of 60:40 acetonitrile/H2O, and the concentration of atRA-d5 was measured using LC-MS/MS as described earlier.

For the pharmacokinetic analysis, the area under the tissue or serum concentration versus time curve from time 0 to infinity (AUC0–∞) was calculated by standard noncompartmental analysis with Phoenix (St. Louis, MO) using a linear trapezoidal method. Kp was calculated using eq. 2 with conversion of serum AUC from h⋅pmol/ml to h⋅pmol/g, assuming 1 ml of serum = 1 g.

Embedded Image(2)

Development of atRA PBPK Model in Healthy Humans.

A PBPK model of atRA was constructed using Simcyp population-based ADME simulator v.14 (Certara, Sheffield, UK) using a full PBPK model (Fig. 1A), and the model was developed and verified according to the workflow shown in Fig. 1B. The absorption kinetics of atRA was simulated with first-order absorption model using the Caco-2 cell permeability data and the absorption kinetics of atRA observed in healthy volunteers. In brief, the fraction absorbed (Fa) and a nominal flow in gut model (Qgut) were predicted using atRA Caco-2 cell permeability data within Simcyp. The ka and lag time were estimated using the observed concentration–time profiles from three studies in healthy volunteers administered with a single oral dose of atRA (Ozpolat et al., 2003; Thudi et al., 2011; Peng et al., 2014). The reported concentration–time profiles were digitized using Plot Digitizer software (http://plotdigitizer.sourceforge.net), and one compartment pharmacokinetic model with first-order absorption was fitted to the data using Phoenix WinNonlin 6.4 (Pharsight Corporation, Cary, NC).

Fig. 1.
Fig. 1.

Development of the atRA PBPK model. (A) The atRA PBPK model structure. (B) Workflow of model development and verification.

For atRA distribution, a full PBPK model was used. Because intravenous atRA studies are not available in humans, steady-state volume of distribution (Vss) in humans was first predicted using allometric scaling from studies in adult Wistar rats and Macaca fascicularis monkeys after administration of atRA intravenously (Sandberg et al., 1994; Saadeddin et al., 2004). The Vss of atRA in adult humans was estimated using the allometric equation:Embedded Image(3)where Y is Vss (l), W is body weight (kg), and a and b are the allometric coefficient and exponent, respectively. The Kp values were independently predicted in Simcyp using the method described by Rodgers and Rowland (2006). The Kp values for liver, kidney, and pancreas were then refined based on data from the mouse pharmacokinetic study described earlier. For the lung and brain, the Kp values were refined based on reported literature values (Wang et al., 1980). To achieve concordance between the allometric scaling Vss and in silico predicted Vss, a Kp value of 0.2 was assigned to all other tissues except those mentioned previously.

The hepatic clearance of atRA was predicted using previously reported enzyme kinetic data for CYP2C8, CYP3A4, CYP3A5, CYP3A7, and CYP26A1 (Thatcher et al., 2010). Intersystem extrapolation factor (ISEF) of CYP2C8, CYP3A4, CYP3A5, and CYP3A7 used Simcyp default values for BD supersomes as 0.43, 0.24, 0.24, and 0.24, respectively. For CYP26A1, the ISEF value of 0.14 was calculated based on the reported ratio between the predicted atRA clearance (171.0 ± 99.5 µl/min/mg) from recombinant insect cell microsome data and the observed clearance (23.2 ± 20.4 µl/min/mg) obtained in human liver microsomes (Thatcher et al., 2010).

As the standard Simcyp population file does not have CYP26A1, CYP26A1 was incorporated into the population profile as CYP2J2 in the liver and gastrointestinal tract. In the liver, the protein abundance of 1.6 pmol/mg protein (63% CV) was assigned for CYP26A1 enzyme abundance according to the literature (Thatcher et al., 2010). The CYP26A1 turnover half-life was estimated between 24 and 48 hours with the turnover rate constant (kdeg) of 0.029–0.014 hour−1 based on the in vitro data (Topletz et al., 2015). A sensitivity analysis of kdeg in the range of 0.029–0.014 hour−1 was conducted with less than 20% difference on the predicted atRA AUC. Based on this, a turnover rate constant for CYP26A1 was estimated at 0.014 hour−1 and used in this model. Because both the CYP26A1 mRNA and protein levels were low in human intestine (Topletz et al., 2012), 0 was assigned to CYP26A1 enzyme abundance in the gastrointestinal tract.

All-trans retinoic acid is known to induce the mRNA and protein expression of CYP26A1. The induction parameters determined previously in HepG2 cells were incorporated into the model. The maximum fold induction (Emax) of CYP26A1 by atRA was estimated between 22-fold and 44-fold based on the observed increase in formation of RA metabolites from atRA-d5 in HepG2 cells after treatment with 1 µM atRA (Topletz et al., 2015), and an Emax of 33 was used for the model. A sensitivity analysis of Emax in the range of 22–44 was conducted with less than 15% difference in the predicted atRA AUC. Therefore, Emax of 33 was considered acceptable. An atRA concentration (0.09 µM) that supports half maximal induction (EC50) was taken from the dose–response analysis of CYP26A1 induction by atRA after 24 hours of treatment in HepG2 cells (Tay et al., 2010). The fraction of unbound drug in the in vitro incubation (fu,inc) of 0.4 was used in the model based on the literature (Topletz et al., 2015).

Simulation of atRA Disposition in Healthy Humans.

To compare the simulated atRA disposition with the reported studies in healthy humans, PubMed searches were conducted using the search terms “retinoic acid pharmacokinetics” and “retinoic acid” in the title or abstract of the article. Three studies were found on atRA disposition in healthy volunteers (Ozpolat et al., 2003; Thudi et al., 2011; Peng et al., 2014). The observed concentration–time profiles from these publications were digitized using Plot Digitizer software (http://plotdigitizer.sourceforge.net). atRA pharmacokinetics were simulated using the Simcyp healthy volunteer profile with the modifications described earlier.

For one study (Peng et al., 2014) that was reported in Chinese, the Simcyp Chinese healthy volunteer profile was used with the modifications of the population as described earlier. The population profile was also modified to match the clinical study population for the age range and proportion of females.

For each study, 10 trials were simulated with the number (n) of subjects per trial matching that reported for that clinical study using random seed. The mean and standard deviation of the trials are reported. The dosing regimen was set identical to observed clinical studies. In the healthy volunteer studies, atRA was dosed either in mg basis or in mg/m2 basis. All the studies were simulated using the same dosing units as reported in the clinical studies, and the data are reported with the matching dosing regimen as was originally reported. The validity of the model for healthy volunteers was evaluated by comparing the observed and simulated data using unpaired t test and by calculating the fold difference between the observed and predicted data.

Development of atRA PBPK model in Cancer Patients.

The model of atRA disposition in cancer population was built based on the healthy population model with modifications on absorption kinetics and population parameters. To optimize the atRA PBPK model in adult cancer patients, Fa was estimated using bioavailability (F) calculated from published clinical studies in adult healthy subjects and cancer patients using eq. 4:Embedded Image(4)in which F1 is the bioavailability in healthy subjects. Dose1 (reported in mg/m2) and AUC1 are from a published clinical study in healthy subjects (Ozpolat et al., 2003). Dose2 (reported in mg/m2) and AUC2 are the mean values from published clinical studies in cancer patients with the same dosage (Muindi et al., 1992a,b, 1994; Rigas et al., 1993, 1996; Miller et al., 1994; Adamson et al., 1995; Lee et al., 1995; Russo et al., 1998). Clearance of atRA was assumed to be identical in patients because the observed elimination half-life and tmax of atRA in healthy subjects matched the half-life and tmax observed in cancer patients with same single oral dose, but the maximum plasma concentration (Cmax) and AUC were both altered in cancer patients. Therefore, Fa of 0.14 was used in the cancer population model. Due to the limited information on atRA disposition in healthy children and children with cancers, the Fa of 0.14 calculated from adult population was also used for the pediatric cancer population model.

The adult cancer population file was built based on the healthy adult population file with modifications of demographic parameters including age, sex, height, body weight, and blood composition including albumin, alpha-1-acid glycoprotein, and hematocrit to match the population parameters described elsewhere (Cheeti et al., 2013). For the pediatric population, the demographic parameters were kept the same as healthy children. CYP26A1 expression was incorporated in the cancer populations identically to that described earlier for the healthy population.

To compare the simulated atRA disposition with the reported studies in adult and pediatric cancer patients, PubMed searches were conducted using the search terms “retinoic acid pharmacokinetics” and “retinoic acid” in the title or abstract of the article, and studies that were conducted in cancer patients with sufficient pharmacokinetic data for evaluation were selected. Design of simulation trials was the same as described above for healthy subjects. In all of the reported atRA studies in patients the dose was reported as mg/m2 basis, so the atRA dose in all the simulations was also set in mg/m2 units. All the data are also reported for mg/m2 dosing.

Model Verification.

AUC was the primary variable used for model verification. For all populations for which only a single study existed at a given dosage level of atRA, an unpaired t test was used to compare whether the observed and predicted AUCs were significantly different from one another. For studies that were conducted at the same dosage level of atRA, acceptance criteria with the consideration of sample size and variance of the parameter of interest in reported studies were used. A prediction was considered acceptable if the predicted AUC was within the calculated upper and lower verification range of the observed AUC. The acceptance range of the mean simulated AUC was calculated as described elsewhere (Abduljalil et al., 2014) according to eqs. 5, 6, and 7,Embedded Image(5)Embedded Image(6)Embedded Image(7)where the calculated values A and B are the boundary values for fold difference between the predicted and observed AUC in a given study, Embedded Image is the mean of the AUCs of atRA in all clinical studies giving the same dosing regimen, N is the sample size, and σ is the standard deviation of the AUC on the natural log scale. Due to the dose difference in reported clinical studies and dose- and time-dependent kinetics of atRA, the acceptance criteria (acceptable fold difference between observed and predicted) was calculated separately for each dose strength.

Determination of CYP Inhibition by Ketoconazole and Liarozole.

To determine the inhibition potency of ketoconazole and liarozole toward atRA metabolism, the inhibition of atRA hydroxylation by ketoconazole and liarozole was evaluated using recombinant CYP3A4 and CYP2C8 as described elsewhere (Thatcher et al., 2011). Inhibition of CYP26A1 was tested by measuring the inhibition of atRA depletion. In brief, 5 pmol CYP3A4 and CYP2C8 supersomes were preincubated with ketoconazole or liarozole (≥7 concentrations of 0–100 μM) and 1 μM atRA for 5 minutes at 37°C in 1 ml 100 mM potassium phosphate buffer (KPi; pH 7.4). The reactions were initiated with addition of 1 mM NADPH and allowed to proceed for 10 minutes. Reactions were terminated by addition of 3 ml of ethyl acetate containing 100 pmol internal standard (4oxo-atRA-d3). After centrifugation for 2 minutes to separate the organic and aqueous layers, the organic layer was collected then dried under nitrogen gas, and the sample was reconstituted with 100 μl of acetonitrile and analyzed by LC-MS/MS as described earlier.

To determine the inhibition of CYP26A1 by ketoconazole and liarozole, 0.1 pmol CYP26A1 was preincubated with 0.2 pmol reductase at room temperature for 5 minutes to allow the incorporation of reductase into the membrane. CYP26A1 with reductase was then preincubated with ketoconazole or liarozole (≥6 concentrations of 0–100 μM) and 2.5 nM atRA for 5 minutes at 37°C. Incubations (1 ml) were initiated by addition of 1 mM NADPH and quenched after 5 minutes with 5 ml of ethyl acetate containing. 15 pmol atRA-d5 as an internal standard. The samples were centrifuged to separate the organic layer, dried under nitrogen, and reconstituted with 100 μl of acetonitrile before analysis with LC-MS/MS as described earlier. Depletion of atRA was determined in comparison with samples incubated in the absence of NADPH.

All experiments were performed in triplicate. The IC50 values were estimated by fitting eq. 8 to the data using nonlinear regression in GraphPad Prism (Graphpad Software, San Diego, CA).Embedded Image(8)Total activity is the percentage of activity in the absence of inhibitor, [I] is the concentration of ketoconazole or liarozole, and the IC50 is the concentration of inhibitor that causes 50% of the total measured inhibition.

Determination of Unbound Fraction of Ketoconazole and Liarozole in Supersomes and Plasma.

The protein binding of the inhibitors was determined using ultracentrifugation as described elsewhere (Nakai et al., 2004; Shirasaka et al., 2013). To determine the unbound fraction in the in vitro incubations and in plasma, 100 μM of ketoconazole or liarozole was added to CYP3A4, CYP2C8, or CYP26A1 (5 pmol/ml) microsomes in KPi buffer or to plasma and the samples split into two aliquots. One aliquot was incubated at 37°C for 90 minutes; the other aliquot was centrifuged in a Sorvall Discovery M150 SE Ultracentrifuge at 100,000 rpm (435,630g) for 90 minutes at 37°C to precipitate protein. After centrifugation, 100 μl aliquots were transferred to a 96-well plate containing 100 μl acetonitrile.

A standard curve of each inhibitor was prepared in 50:50 KPi: acetonitrile (v: v) and analyzed in parallel. The samples were centrifuged using a Beckman Coulter Allegra 25R centrifuge at 3000g for 15 minutes at 4°C. After centrifugation, 100 μl from each well was transferred to a clean 96-well plate for LC-MS/MS analysis.

Samples were separated using a Shimadzu Prominence UFLC (Columbia, MD) equipped with a Thermo Hypersil Gold C18 column (2.1 × 100 mm, 1.9 μm) using gradient elution with (A) H2O (0.1% formic acid) and (B) acetonitrile at a flow rate of 0.5 ml/min. The gradient was from an initial 95% A to 10% A over 3 minutes, stayed at 10% A for 1.5 minutes, and then returned to 95% A for 2 minutes. Analytes were detected by an AB Sciex QTRAP 3200 mass spectrometer operated in electrospray positive mode with detection of m/z multiple reaction monitoring transition of 531.1/82.2 for ketoconazole and 309.14/241.1 for liarozole. The declustering potential, entrance potential, cell entrance potential, collision energy, and collision exit potential were set to 86, 5.5, 20, 73, and 6, respectively, for ketoconazole and 26, 4.5, 16, 17, and 4, respectively, for liarozole.

Development of Minimal PBPK Models for Ketoconazole and Liarozole.

For the ketoconazole minimal PBPK model, ketoconazole-400 mg QD compound file in Simcyp was first verified with clinical observations. However, oral clearance (CLpo) (7.4 l/h) assigned in the compound file was generally lower than the reported average CLpo value (9.5 l/h) (Daneshmend et al., 1981, 1984; Baxter et al., 1986). Therefore, CLpo was optimized to 9.5 l/h with 30% CV. Because CLpo (14.4 l/h) in Simcyp ketoconazole-200 mg QD file was similar to the average CLpo (13.5 l/h) reported in the literature (Daneshmend et al., 1981, 1984; Huang et al., 1986), this file was used without modification on CLpo. For both models, the inhibition potency of ketoconazole toward CYP3A4, CYP2C8, and CYP26A1 as determined in the in vitro experiments was incorporated into the compound file.

The liarozole minimal PBPK model was built using information from the literature (Bryson and Wagstaff, 1996; De Buck et al., 2007) and in vitro experiments. The absorption of liarozole was described with first-order absorption model with Simcyp predicted absorption parameters. Elimination of liarozole was characterized using the CLpo reported in humans (Bryson and Wagstaff, 1996). Liarozole distribution was described using a minimal PBPK model with Vss estimated from reported clinical data as 1.08 l/kg (Bryson and Wagstaff, 1996; Denis et al., 1998), and the elimination rate constant was calculated from eq. 9:Embedded Image(9)in which Css, max and Css, min are the peak and trough concentrations at steady state, k is elimination rate constant, and τ is the dosing interval. The Kp value of liver was predicted using method 2 (Rodgers and Rowland, 2006) in Simcyp. In vitro measurements of fu,mic and Ki of liarozole for CYP2C8, CYP26A1, and CYP3A4 were incorporated into the interaction module. Because inhibition experiments were done at S < < Km, the IC50 can be assumed to be approximately equal to Ki, and IC50 values determined in the inhibition assay were used as Ki (Lutz and Isoherranen, 2012).

Results

atRA PBPK Model in Healthy Humans.

The atRA PBPK model in Simcyp was first developed in healthy humans based on obtained experimental and literature data. To develop the PBPK model, the blood to plasma ratio of atRA was measured. The blood to plasma ratio of atRA measured in human blood was 2.3, indicating significant partitioning into blood cells. To develop the PBPK model and establish tissue Kp values, we determined the atRA disposition in mice (Fig. 2). The elimination of atRA from serum after i.p. dosing was efficient with elimination half-life of 0.5 hours. Similar half-lives of 0.4 to 0.6 hours were observed in the liver, kidney, and pancreas, indicating fast metabolism of atRA in tissues and a rapid distribution equilibrium. The AUCs in the liver, kidney, and pancreas were higher than the AUC in serum (Table 1), indicating efficient partitioning of atRA into tissues. The measured Kp values of 3.79 and 2.06 in the liver and kidney, respectively, were similar to the Kp values calculated from the reported concentration–time data in mice after i.v. dosing with 10 mg/kg atRA (Wang et al., 1980). The Kp values predicted using Rodgers and Rowland method were 16.78, 10.14, 17.34, 2.28, and 26.90 for the liver, kidney, pancreas, lung, and brain, respectively, with a predicted Vss of 28.10 l/kg. These values greatly exceeded the Kp values obtained from mouse studies. Similarly, the predicted Vss significantly exceeded the Vss of 0.46 l/kg predicted via allometric scaling. Therefore, the experimental Kp values were used to refine the distribution model (Table 2).

Fig. 2.
Fig. 2.

Concentration–time profiles of atRA in C57BL/6J mouse (n = 12) serum, liver, kidney, and pancreas after administration 1 mg/kg atRA-d5 i.p. Data are shown as mean ± S.D.

View this table:
TABLE 1

Pharmacokinetic parameters of atRA obtained in male C57BL/6J mice (n = 12) dosed with 1 mg/kg atRA-d5 i.p.

View this table:
TABLE 2

Summary of parameter input values for atRA PBPK model

To predict the Fa and Qgut in the first-order atRA absorption model, we measured the permeability of atRA in Caco-2 cells. In Caco-2 cells, a permeability of 3.58 × 10−6 cm/s was measured for atRA. The atRA permeability ratio (apical to basolateral/basolateral to apical) was 1.1, indicating that no active transporters were involved (Supplemental Fig. 1, Supplemental Table 1). The Fa of 0.48 and Qgut of 3.5 l/h, were predicted based on Caco-2 cell permeability data indicating a low permeability for atRA in the gut. A ka of 0.90 1/h with a 0.84-hour lag time for absorption was obtained from fitting the observed clinical data into one compartment model.

Due to the lack of dosage form, systemic atRA clearance has never been evaluated in humans. Clinical studies in humans after oral administration of atRA have reported that urinary excretion of unchanged atRA is negligible, and atRA is not found in human urine after β-glucuronidase incubation (Muindi et al., 1992b; Conley et al., 1997). Therefore, renal clearance and UDP-glucuronosyltransferase-mediated elimination were not incorporated into the model, and the clearance of atRA was all assigned to the hepatic P450 enzymes (Table 2). The complete model parameters are summarized in Table 2.

To verify the model, atRA disposition was first simulated in healthy subjects after administration of a single oral dose of atRA. Three different studies with atRA dosed at 10 mg, 20 mg, and 22.5 mg/m2 by mouth (PO) were simulated. The observed mean concentration–time curves were within the 5th and 95th percentile of the simulated, and the mean simulated AUCs were 1.5-fold, 1.2-fold, and 0.9-fold of the observed ones (Fig. 3). The predicted AUCs for two of the three studies (70%) met the predefined acceptance criteria (simulated AUC not significantly different from observed, P = 0.4 and 0.7) and predicted Cmax and tmax were similar to observed values (Table 3). The developed PBPK model successfully simulated the dose dependent kinetics of atRA after a single-dose administration with the oral clearance decreasing from 140 ± 136 l/h after a 10-mg dose to 87 ± 72 l/h after a 20-mg dose and to 65 ± 63 l/h after a 22.5-mg/m2 dose of atRA.

Fig. 3.
Fig. 3.

Predicted and observed mean plasma concentration–time profiles of atRA in healthy volunteers. (A) Administration of 10-mg atRA PO (as described in Thudi et al., 2011). (B) Administration of 20 mg atRA PO (as described in Peng et al., 2014). (C) Administration of 22.5 mg/m2 atRA PO BID (as described in Ozpolat et al., 2003). The solid line depicts the mean of the simulated data with 5th and 95th percentile shown with light gray line. Solid circles show the observed data in the study.

View this table:
TABLE 3

Observed and predicted pharmacokinetic parameters of atRA in healthy subjects (mean ± S.D.) after single dose (sd) and multiple doses (dosing regimen and duration specified)

To test whether the induction parameters for CYP26A1 would accurately predict the autoinduction of atRA clearance, the disposition of atRA after 22.5-mg/m2 oral doses twice a day (BID) for 15 days was simulated (Fig. 3C). The predicted AUCs after multiple doses were not significantly different from the observed AUCs (P = 0.1 and 0.2) (Table 3). The simulated mean AUCs on days 9 and 15 were 0.6-fold and 0.5-fold, respectively, of the observed ones. The predicted 64% and 65% decreases in atRA AUC after multiple dosing were in good agreement with the observed 42% and 37% reductions of atRA AUC on days 9 and 15, respectively. Hence the developed model successfully simulated the saturation kinetics and autoinduction of atRA clearance, and based on these results the model was considered verified.

atRA PBPK Model in Adult and Pediatric Cancer Patients.

The atRA disposition in cancer patients was first simulated using the PBPK model developed in healthy population. Using the atRA drug file developed in healthy subjects overpredicted the AUC and Cmax by about 5-fold and 3-fold whereas the tmax and half-life were accurately predicted. This suggested that Fa is decreased in cancer patients without changes in ka, CL, or V. Therefore, Fa was optimized in the PBPK model from 0.48 to 0.14, and the population profile was modified according to published changes in cancer patients (Cheeti et al., 2013).

The atRA disposition was then simulated in 11 clinical studies conducted in cancer patients with atRA dosed at 40 mg/m2 (two studies) and 45 mg/m2 (nine studies). The predicted AUC for nine of the studies (82%) was within the predefined acceptance range, with mean simulated AUCs within 0.5-fold and 1.1-fold of the observed ones (Table 4). The observed mean AUC values in the two studies that were not accurately predicted were more than 2-fold different from the mean AUC of the other observed studies in cancer patients (Table 4). The observed AUC in the study conducted in large cell carcinoma and squamous cell carcinoma patients was <50% of the AUC values observed in the other studies with the same dosing regimen (Table 4). None of the other studies in cancer patients had a mean AUC that would have met the calculated acceptance range for the study in large cell carcinoma and squamous cell carcinoma patients. The observed AUC in one of the studies conducted in solid tumor patients had an observed AUC that was 2.7-fold to 3-fold higher than the mean atRA AUC observed in other studies in solid tumor patients, and none of the other studies in solid tumor patients had a mean AUC that would have met the calculated acceptance range for this study (Table 4). As such, neither the model nor the other clinical studies predicted the atRA AUCs in 2 of the 11 studies within the acceptance criteria. Unfortunately, the reported information on study populations was not sufficient to evaluate potential mechanisms of why these two studies showed discrepant disposition compared with other cancer studies.

View this table:
TABLE 4

Observed and predicted pharmacokinetic parameters of atRA in adult cancer patients (mean ± S.D.) after single and multiple doses

The PBPK model in a cancer population also successfully simulated the dose-dependent kinetics of atRA with the simulated mean oral clearance decreasing from 335 l/h (two studies) to 276 l/h (nine studies) as the dose increased from 40 mg/m2 to 45 mg/m2 atRA. Similarly, the predicted AUCs of atRA were within the acceptance range on both day 1 and day 7 of dosing when compared with the observed data of multiple dosing study in chronic myeloid leukemia patients dosed with 40 mg/m2 atRA BID (Table 4). The model predicted a 57% decrease in atRA AUC over the 7-day dosing which is in good agreement with the observed 62% reduction in atRA AUC on day 7.

Because atRA is frequently used in pediatric populations, the PBPK model was applied to a pediatric cancer population to simulate atRA disposition in pediatric patients. The atRA disposition after a single oral dose was simulated for three different studies with 30 and 40 mg/m2 doses of atRA to pediatric cancer patients. The simulated mean AUC was 1.0-fold, 0.4-fold, and 0.5-fold of the observed values. The predicted AUC for two of the studies was within the predefined acceptance range (Table 5). For the third study with a higher dosing regimen, the predicted AUC was not significantly different from the observed (P = 0.1) and was within 2-fold of the observed. Similar to what was observed in adult clinical studies, the large variability in atRA kinetics was also observed in pediatric patients. The simulation of atRA disposition in adult and pediatric populations successfully captured this large variability (Tables 4 and 5).

View this table:
TABLE 5

Observed and predicted pharmacokinetic parameters of atRA in pediatric cancer patients (mean ± S.D.) after single dose

PBPK Modeling of Effects of Ketoconazole and Liarozole on atRA Disposition.

To predict the drug-drug interactions (DDIs) between atRA and ketoconazole and liarozole, the inhibition of atRA metabolism by ketoconazole and liarozole was first characterized in vitro (Fig. 4), and the results were incorporated into ketoconazole and liarozole models (Supplemental Tables 3 and 4).

Fig. 4.
Fig. 4.

In vitro IC50 curves for CYP26A1, CYP3A4, and CYP2C8 with atRA as substrate and ketoconazole and liarozole as inhibitors. Data are shown as mean ± S.D. (A) Ketoconazole CYP3A4 IC50: 0.015µM; CYP26A1 IC50: 0.47 µM; CYP2C8 IC50: 6.65 µM. (B) Liarozole CYP3A4 IC50: 1.22 µM; CYP26A1 IC50: 3.26 µM; CYP2C8 IC50:1.33µM.

Ketoconazole had the highest inhibition potency toward CYP3A4 (IC50 0.015 µM; 95% confidence interval [CI], 0.010–0.023 µM), followed by CYP26A1 (IC50 0.47 µM; 95% CI, 0.25–0.91 µM) and CYP2C8 (IC50: 6.7 µM; 95% CI, 1.4–31 µM) (Fig. 4A). In contrast, liarozole had similar affinity to all three CYP isoforms with an IC50 of 3.3 µM (95% CI, 2.1–5.0 µM) for CYP26A1; 1.2 µM (95% CI, 0.92–1.6 µM) for CYP3A4; and 1.3 µM (95% CI, 0.77–2.3 µM) for CYP2C8 (Fig. 4B).

The ketoconazole PBPK models for different ketoconazole dosages were verified with clinical ketoconazole pharmacokinetic studies (Supplemental Fig. 2 and Supplemental Table 2), and the ketoconazole 200 mg PO dosing model and 400 mg PO dosing model were considered acceptable for DDI simulation. A liarozole PBPK model was also developed based on existing in vivo data (Supplemental Table 4), and liarozole disposition was simulated and compared with reported clinical studies (Supplemental Table 5). Due to the lack of information on the variability in liarozole disposition in the reported study, the liarozole model could not be verified using the acceptance criteria described. However, the predicted pharmacokinetic parameters were within the 2-fold range of the observed values and thus were considered to be similar to the reported parameters and acceptable for DDI simulation.

After model verification, a multiphase DDI study between ketoconazole and atRA was simulated. In the clinical study, cancer patients were treated with 45 mg/m2 atRA on days 1, 2, 28, and 29 of treatment and with 45 mg/m2 atRA BID for the remaining days 3 to 27. A single dose of 200 mg or 400 mg ketoconazole was given on days 2 and 29, one hour before atRA (Rigas et al., 1993). The study was simulated using the verified ketoconazole PBPK model in the cancer population, and the simulation successfully predicted the DDI between atRA and ketoconazole in adult cancer patients (Table 6 and Fig. 5). The predicted AUC of atRA on each kinetic day in the presence and absence of ketoconazole met the acceptance criteria and was also within 2-fold of the observed (Table 6). The observed AUC on the first day was not available for the individual ketoconazole dosing groups and hence a single control group for atRA disposition was simulated.

View this table:
TABLE 6

Observed and predicted AUC of atRA in adult cancer patients in the presence of ketoconazole and liarozole (mean ± S.D.)

The ketoconazole study data are from Rigas et al. (1993), and liarozole data are from Miller et al. (1994). In the studies atRA was dosed for 29 days and ketoconazole or liarozole administered on days 2 and 29 while the control session pharmacokinetics studies were done on days 1 and 28.

Fig. 5.
Fig. 5.

Predicted (black bars) and observed (white bars) atRA AUC in cancer patients dosed with a single dose of 45 mg/m2 of atRA on days 1, 2, 28, and 29, and twice a day for the rest of the days. A single dose of ketoconazole or liarozole was given on days 2 and 29 1 hour before atRA. Ketoconazole was dosed at (A) 200 mg and (B) 400 mg, and liarozole at (C) 75 mg, (D) 150 mg, and (E) 300 mg. The observed data are from Miller et al. (1994) and Rigas et al. (1993).

The predicted decrease in atRA AUC between day 1 and day 28 of dosing was about 61% and in excellent agreement with the observed 61% decrease in atRA AUC. The inhibition of atRA clearance by ketoconazole was also well predicted. The simulation predicted a 1.4-fold increase in atRA AUC after ketoconazole administration on day 2, a fold increase similar to the observed average 1.1-fold increase in atRA AUC. The simulation also predicted a 2.0-fold and 2.7-fold increase in atRA AUC after 200 mg and 400 mg ketoconazole administration on day 29, a fold increase in excellent agreement with the observed 1.5-fold and 2.2-fold increases in atRA AUC between days 28 and 29. In context of predicting the efficacy of ketoconazole in combating therapy resistance, the model accurately predicted that 400-mg ketoconazole would restore the atRA exposure on day 29 to the levels observed on day 1.

Similar to the DDI study between ketoconazole and atRA, a DDI study between liarozole and atRA was simulated according to a reported study (Miller et al., 1994). In the reported study, cancer patients were treated with 45 mg/m2 atRA on days 1, 2, 28, and 29 and with 45 mg/m2 BID for the remaining days 3 to 27. A single dose of 75 mg, 150 mg, or 300 mg of liarozole was given on days 2 and 29, 1 hour before atRA. The simulation successfully predicted the DDI between atRA and liarozole in adult cancer patients with the exception of 150-mg liarozole dosing where the observed studies reported atRA AUCs that had acceptance ranges that excluded all other clinical observations (Table 6 and Fig. 5). The other predicted AUCs of atRA on each reported study day in the presence or absence of liarozole were within the predefined model acceptance criteria.

The simulation predicted about 61% decrease in atRA AUC between day 1 and day 28, a prediction in good agreement with the observed 58%–84% decrease in atRA AUC. The simulation also correctly predicted that on day 2 despite the liarozole administration, the AUC of atRA would be lower or not different when compared with that on day 1 of treatment. Similarly, the simulation predicted a 1.2-fold to 1.5-fold increase in atRA AUC by liarozole dosing between day 28 and day 29, which is in excellent agreement with the 1.7-fold to 2.2-fold observed increase in atRA AUC (Table 6). The simulation also accurately predicted the efficacy of liarozole in combatting therapy resistance to atRA. The simulated data predicted that 300-mg liarozole would restore atRA AUC to 58% of the initial AUC after a single dose. This is in excellent agreement with the observed efficacy of liarozole restoring atRA AUC to 61% of that observed after single dose (Table 6).

Discussion

All-trans retinoic acid is widely used as a therapeutic agent in the treatment of APL and has shown promise in the treatment of other malignancies. However, the therapeutic use of atRA is limited due to its dose- and time-dependent kinetics. Although the dose- and time-dependent kinetics of atRA have been observed in animals and humans, the mechanisms and kinetics of saturation and induction of atRA metabolism have not been well characterized in vivo. The PBPK model of atRA disposition presented here provides a mechanistic framework for understanding the time- and dose-dependent kinetics of therapeutically administered atRA.

Complex atRA kinetics were incorporated in the PBPK model and successfully simulated clinical studies after single and multiple doses of atRA both in healthy volunteers and in cancer patients. The simulations strongly support the hypothesis that the dose-dependent, nonlinear kinetics of atRA are due to saturation of CYP26A1-mediated clearance of atRA, and that the time dependent increase in atRA clearance is due to induction of hepatic CYP26A1 by atRA. It is possible that CYP26A1 is induced in other metabolic sites as well, but at present CYP26A1 expression appears to be restricted predominantly to the liver (Topletz et al., 2012). Taken together, the developed atRA PBPK model demonstrates a unique application of PBPK modeling to a complex clinical scenario of drug disposition including time- and concentration-dependent changes in clearance. The presented data show that in vitro data can successfully be used to develop PBPK models of such complex systems and incorporate new eliminating enzymes into existing models when adequate physiologic data are available.

At present, there is no appropriate dosage form of atRA to allow i.v. administration to humans, so the absorption and distribution kinetics of atRA are poorly understood. The studies presented here show that despite its high lipophilicity, atRA has relatively low permeability and a restricted volume of distribution similar to many other lipophilic acidic compounds that are highly bound to plasma albumin. The Caco-2 cell assay suggested that passive diffusion is the primary mechanism of atRA absorption in the gut and that transporters do not contribute to atRA absorption. The distribution volume of atRA appears to be predictable between species, and it is likely that the individual organ tissue to plasma partition ratios are similar between species. However, at present the distribution component of the PBPK model is not adequate to allow simulation of atRA distribution to specific organs or drug target sites except the liver, kidney, lung, pancreas, and brain.

The in vivo clearance of atRA was predicted from in vitro enzyme kinetic data with the addition of the in vivo information of lack of renal elimination of unchanged atRA and detection of conjugated metabolites such as glucuronides in human urine. The enzyme kinetic data and the dynamic model predicted approximately 90% clearance of atRA by hepatic CYP26A1 metabolism after a single dose of atRA, a value in good agreement with the previously predicted >90% contribution based on static prediction (Thatcher et al., 2010). Uniquely, this model also accurately predicted the saturation kinetics of atRA based on the in vitro enzyme kinetic values and the time-dependent induction of CYP26A1 based on HepG2 cell and hepatocyte induction data. In addition to atRA disposition when administered alone, the model also accurately predicted the complex DDIs between atRA and ketoconazole and atRA and liarozole.

The inhibition of atRA clearance has been an attractive therapeutic target for combating therapy resistance to atRA in treatment of APL and possibly other cancers. However, the development of inhibitors of atRA metabolism has been hindered by lack of understanding of the necessary potency and pharmacokinetic characteristics of novel inhibitors, especially in the target cancer population. The developed model together with the DDI simulations provides insight to the shortcomings of the clinical studies conducted with ketoconazole and liarozole as CYP26 inhibitors, and it will allow design of future dosing regimens that can restore atRA exposures to the levels observed after single doses. In addition, the model can be used to design atRA dosing regimens that achieve maximum exposure and efficacy.

It is well recognized that patient demographics and physiologic features are different for cancer patients and healthy subjects. To address these differences, a previously developed adult cancer population model (Cheeti et al., 2013) was adapted to test whether the healthy population PBPK model for atRA could be applied to cancer patients. Based on model verification and sensitivity analysis, all the data collectively suggested that atRA absorption is significantly decreased in cancer patients, while all the physiologic features reported in the previous cancer population model were applicable to the cancer populations included in the atRA clinical studies. The decrease in atRA absorption is not entirely surprising, as malnutrition, anorexia, and cachexia and alterations in nutrient absorption and nutritional status are commonly reported in cancer patients (Argilés, 2005). As atRA is also the active form of vitamin A, it is possible that alterations in its absorption are related to the general poor absorption of nutrients in cancer patients and will not reflect absorption processes for cancer medications. However, further studies are needed to investigate potential effects of cancer on drug absorption.

With the modification of the absorption of atRA in cancer patients, the developed PBPK model accurately predicted atRA disposition in this challenging patient population for which it is often difficult to obtain sufficient pharmacokinetic data. In addition, the developed PBPK model could be applied to predict atRA disposition in pediatric cancer patients, a uniquely vulnerable population. As such, the data obtained support the use of PBPK models developed in healthy populations to predict disposition in patient populations including pediatrics.

In summary, we have developed a mechanistic PBPK model predicting atRA disposition in both healthy volunteers and cancer patients. The model was successfully applied to simulate the pharmacokinetic profiles of atRA after single and multiple doses and to quantitatively predict the interaction between atRA and atRA metabolism inhibitors (ketoconazole and liarozole). Therefore, it could be used to investigate atRA disposition, to design drug-drug interaction studies in the development of novel atRA metabolism inhibitors, and to predict atRA pharmacokinetics in cancer patients.

Acknowledgments

The authors thank Michael D. Haenisch from the Department of Comparative Medicine at University of Washington for his skillful assistance in the mouse experiments.

Authorship Contributions

Participated in research design: Jing, Nelson, Shirasaka, Paik, Amory, Isoherranen.

Conducted experiments: Jing, Nelson, Amory, Paik, Shirasaka.

Performed data analysis: Jing, Nelson, Shirasaka, Isoherranen.

Wrote or contributed to the writing of the manuscript: Jing, Nelson, Shirasaka, Isoherranen.

Footnotes

    • Received February 1, 2017.
    • Accepted March 6, 2017.

  • This work was supported in part by funding from the National Institutes of Health National Institute of General Medical Sciences [Grant R01 GM111772] and the Eunice Kennedy Shriver National Institute of Child Health and Human Development through cooperative agreement [Grant U54 HD42454] as part of the Cooperative Contraceptive Research Centers Program.

  • https://doi.org/10.1124/jpet.117.240523.

  • ↵Embedded ImageThis article has supplemental material available at jpet.aspetjournals.org.

Abbreviations

APL
acute promyelocytic leukemia
atRA
all-trans retinoic acid
AUC
area under plasma concentration–time curve
BID
twice a day
CI
confidence interval
CLpo
oral clearance
DDI
drug-drug interaction
Emax
maximum fold induction
fu,inc
fraction of unbound drug in the in vitro incubation
ISEF
intersystem extrapolation factor
LC-MS/MS
liquid chromatography with tandem mass spectrometry
PBPK
physiologically based pharmacokinetics
PO
by mouth
RAR
retinoic acid receptor
TEER
transepithelial electrical resistance
Vss
steady-state volume of distribution

References

  1. ↵
    1. Abduljalil K,
    2. Cain T,
    3. Humphries H, and
    4. Rostami-Hodjegan A
    (2014) Deciding on success criteria for predictability of pharmacokinetic parameters from in vitro studies: an analysis based on in vivo observations. Drug Metab Dispos 42:1478–1484.
    OpenUrlAbstract/FREE Full Text
  2. ↵
    1. Ablain J and
    2. de The H
    (2011) Revisiting the differentiation paradigm in acute promyelocytic leukemia. Blood 117:5795–5802.
    OpenUrlAbstract/FREE Full Text
  3. ↵
    1. Adamson PC
    (1996) All-trans-retinoic acid pharmacology and its impact on the treatment of acute promyelocytic leukemia. Oncologist 1:305–314.
    OpenUrlAbstract/FREE Full Text
  4. ↵
    1. Adamson PC,
    2. Bailey J,
    3. Pluda J,
    4. Poplack DG,
    5. Bauza S,
    6. Murphy RF,
    7. Yarchoan R, and
    8. Balis FM
    (1995) Pharmacokinetics of all-trans-retinoic acid administered on an intermittent schedule. J Clin Oncol 13:1238–1241.
    OpenUrlAbstract
  5. ↵
    1. Argilés JM
    (2005) Cancer-associated malnutrition. Eur J Oncol Nurs 9 (Suppl 2):S39–S50.
    OpenUrlCrossRefPubMed
  6. ↵
    1. Arnold SLM,
    2. Amory JK,
    3. Walsh TJ, and
    4. Isoherranen N
    (2012) A sensitive and specific method for measurement of multiple retinoids in human serum with UHPLC-MS/MS. J Lipid Res 53:587–598.
    OpenUrlAbstract/FREE Full Text
  7. ↵
    1. Arnold SLM,
    2. Kent T,
    3. Hogarth CA,
    4. Griswold MD,
    5. Amory JK, and
    6. Isoherranen N
    (2015) Pharmacological inhibition of ALDH1A in mice decreases all-trans retinoic acid concentrations in a tissue specific manner. Biochem Pharmacol 95:177–192.
    OpenUrlCrossRefPubMed
  8. ↵
    1. Arrieta O,
    2. González-De la Rosa CH,
    3. Aréchaga-Ocampo E,
    4. Villanueva-Rodríguez G,
    5. Cerón-Lizárraga TL,
    6. Martínez-Barrera L,
    7. Vázquez-Manríquez ME,
    8. Ríos-Trejo MA,
    9. Alvarez-Avitia MA,
    10. Hernández-Pedro N, et al.
    (2010) Randomized phase II trial of all-trans-retinoic acid with chemotherapy based on paclitaxel and cisplatin as first-line treatment in patients with advanced non-small-cell lung cancer. J Clin Oncol 28:3463–3471.
    OpenUrlAbstract/FREE Full Text
  9. ↵
    1. Baxter JG,
    2. Brass C,
    3. Schentag JJ, and
    4. Slaughter RL
    (1986) Pharmacokinetics of ketoconazole administered intravenously to dogs and orally as tablet and solution to humans and dogs. J Pharm Sci 75:443–447.
    OpenUrlCrossRefPubMed
  10. ↵
    1. Bryson HM and
    2. Wagstaff AJ
    (1996) Liarozole. Drugs Aging 9:478–484.
    OpenUrlPubMed
  11. ↵
    1. Cheeti S,
    2. Budha NR,
    3. Rajan S,
    4. Dresser MJ, and
    5. Jin JY
    (2013) A physiologically based pharmacokinetic (PBPK) approach to evaluate pharmacokinetics in patients with cancer. Biopharm Drug Dispos 34:141–154.
    OpenUrlCrossRefPubMed
  12. ↵
    1. Clewell 3rd HJ,
    2. Andersen ME,
    3. Wills RJ, and
    4. Latriano L
    (1997) A physiologically based pharmacokinetic model for retinoic acid and its metabolites. J Am Acad Dermatol 36:S77–S85.
    OpenUrlCrossRefPubMed
  13. ↵
    1. Conley BA,
    2. Egorin MJ,
    3. Sridhara R,
    4. Finley R,
    5. Hemady R,
    6. Wu S,
    7. Tait NS, and
    8. Van Echo DA
    (1997) Phase I clinical trial of all-trans-retinoic acid with correlation of its pharmacokinetics and pharmacodynamics. Cancer Chemother Pharmacol 39:291–299.
    OpenUrlCrossRefPubMed
  14. ↵
    1. Coombs CC,
    2. Tavakkoli M, and
    3. Tallman MS
    (2015) Acute promyelocytic leukemia: where did we start, where are we now, and the future. Blood Cancer J 5:e304.
    OpenUrlCrossRefPubMed
  15. ↵
    1. Daneshmend TK,
    2. Warnock DW,
    3. Ene MD,
    4. Johnson EM,
    5. Potten MR,
    6. Richardson MD, and
    7. Williamson PJ
    (1984) Influence of food on the pharmacokinetics of ketoconazole. Antimicrob Agents Chemother 25:1–3.
    OpenUrlAbstract/FREE Full Text
  16. ↵
    1. Daneshmend TK,
    2. Warnock DWTA,
    3. Turner A, and
    4. Roberts CJ
    (1981) Pharmacokinetics of ketoconazole in normal subjects. J Antimicrob Chemother 8:299–304.
    OpenUrlAbstract/FREE Full Text
  17. ↵
    1. De Buck SS,
    2. Sinha VK,
    3. Fenu LA,
    4. Gilissen RA,
    5. Mackie CE, and
    6. Nijsen MJ
    (2007) The prediction of drug metabolism, tissue distribution, and bioavailability of 50 structurally diverse compounds in rat using mechanism-based absorption, distribution, and metabolism prediction tools. Drug Metab Dispos 35:649–659.
    OpenUrlAbstract/FREE Full Text
  18. ↵
    1. Denis L,
    2. Debruyne F,
    3. De Porre P, and
    4. Bruynseels J
    (1998) Early clinical experience with liarozole (Liazal) in patients with progressive prostate cancer. Eur J Cancer 34:469–475.
    OpenUrlCrossRefPubMed
  19. ↵
    1. Huang YC,
    2. Colaizzi JL,
    3. Bierman RH,
    4. Woestenborghs R, and
    5. Heykants J
    (1986) Pharmacokinetics and dose proportionality of ketoconazole in normal volunteers. Antimicrob Agents Chemother 30:206–210.
    OpenUrlAbstract/FREE Full Text
  20. ↵
    1. Jing J,
    2. Isoherranen N,
    3. Robinson-Cohen C,
    4. Petrie I,
    5. Kestenbaum BR, and
    6. Yeung CK
    (2016) Chronic kidney disease alters vitamin A homeostasis via effects on hepatic RBP4 protein expression and metabolic enzymes. Clin Transl Sci 9:207–215.
    OpenUrl
  21. ↵
    1. Jones HM,
    2. Chen Y,
    3. Gibson C,
    4. Heimbach T,
    5. Parrott N,
    6. Peters SA,
    7. Snoeys J,
    8. Upreti VV,
    9. Zheng M, and
    10. Hall SD
    (2015) Physiologically based pharmacokinetic modeling in drug discovery and development: a pharmaceutical industry perspective. Clin Pharmacol Ther 97:247–262.
    OpenUrlCrossRefPubMed
  22. ↵
    1. Lee JS,
    2. Newman RA,
    3. Lippman SM,
    4. Fossella FV,
    5. Calayag M,
    6. Raber MN,
    7. Krakoff IH, and
    8. Hong WK
    (1995) Phase I evaluation of all-trans retinoic acid with and without ketoconazole in adults with solid tumors. J Clin Oncol 13:1501–1508.
    OpenUrlAbstract
  23. ↵
    1. Lim YC,
    2. Kang HJ,
    3. Kim YS, and
    4. Choi EC
    (2012) All-trans-retinoic acid inhibits growth of head and neck cancer stem cells by suppression of Wnt/β-catenin pathway. Eur J Cancer 48:3310–3318.
    OpenUrlCrossRefPubMed
  24. ↵
    1. Louisse J,
    2. Bosgra S,
    3. Blaauboer BJ,
    4. Rietjens IMCM, and
    5. Verwei M
    (2015) Prediction of in vivo developmental toxicity of all-trans-retinoic acid based on in vitro toxicity data and in silico physiologically based kinetic modeling. Arch Toxicol 89:1135–1148.
    OpenUrl
  25. ↵
    1. Lutz JD,
    2. Dixit V,
    3. Yeung CK,
    4. Dickmann LJ,
    5. Zelter A,
    6. Thatcher JE,
    7. Nelson WL, and
    8. Isoherranen N
    (2009) Expression and functional characterization of cytochrome P450 26A1, a retinoic acid hydroxylase. Biochem Pharmacol 77:258–268.
    OpenUrlCrossRefPubMed
  26. ↵
    1. Lutz JD and
    2. Isoherranen N
    (2012) In vitro-to-in vivo predictions of drug-drug interactions involving multiple reversible inhibitors. Expert Opin Drug Metab Toxicol 8:449–466.
    OpenUrlCrossRefPubMed
  27. ↵
    1. Miller VA,
    2. Rigas JR,
    3. Muindi JRF,
    4. Tong WP,
    5. Venkatraman E,
    6. Kris MG, and
    7. Warrell Jr RP
    (1994) Modulation of all-trans retinoic acid pharmacokinetics by liarozole. Cancer Chemother Pharmacol 34:522–526.
    OpenUrlPubMed
  28. ↵
    1. Muindi J,
    2. Frankel SR,
    3. Miller Jr WH,
    4. Jakubowski A,
    5. Scheinberg DA,
    6. Young CW,
    7. Dmitrovsky E, and
    8. Warrell Jr RP
    (1992a) Continuous treatment with all-trans retinoic acid causes a progressive reduction in plasma drug concentrations: implications for relapse and retinoid “resistance” in patients with acute promyelocytic leukemia. Blood 79:299–303.
    OpenUrlAbstract/FREE Full Text
  29. ↵
    1. Muindi JR,
    2. Frankel SR,
    3. Huselton C,
    4. DeGrazia F,
    5. Garland WA,
    6. Young CW, and
    7. Warrell Jr RP
    (1992b) Clinical pharmacology of oral all-trans retinoic acid in patients with acute promyelocytic leukemia. Cancer Res 52:2138–2142.
    OpenUrlAbstract/FREE Full Text
  30. ↵
    1. Muindi JR,
    2. Young CW, and
    3. Warrell Jr RPJ
    (1994) Clinical pharmacology of all-trans retinoic acid. Leukemia 8:1807–1812.
    OpenUrlPubMed
  31. ↵
    1. Nakai D,
    2. Kumamoto K,
    3. Sakikawa C,
    4. Kosaka T, and
    5. Tokui T
    (2004) Evaluation of the protein binding ratio of drugs by a micro-scale ultracentrifugation method. J Pharm Sci 93:847–854.
    OpenUrlCrossRefPubMed
  32. ↵
    1. Nelson CH,
    2. Buttrick BR, and
    3. Isoherranen N
    (2013) Therapeutic potential of the inhibition of the retinoic acid hydroxylases CYP26A1 and CYP26B1 by xenobiotics. Curr Top Med Chem 13:1402–1428.
    OpenUrlCrossRefPubMed
  33. ↵
    1. Njar VC,
    2. Gediya L,
    3. Purushottamachar P,
    4. Chopra P,
    5. Vasaitis TS,
    6. Khandelwal A,
    7. Mehta J,
    8. Huynh C,
    9. Belosay A, and
    10. Patel J
    (2006) Retinoic acid metabolism blocking agents (RAMBAs) for treatment of cancer and dermatological diseases. Bioorg Med Chem 14:4323–4340.
    OpenUrlCrossRefPubMed
  34. ↵
    1. Ozpolat B,
    2. Lopez-Berestein G,
    3. Adamson P,
    4. Fu CJ, and
    5. Williams AH
    (2003) Pharmacokinetics of intravenously administered liposomal all-trans-retinoic acid (ATRA) and orally administered ATRA in healthy volunteers. J Pharm Pharm Sci 6:292–301.
    OpenUrlPubMed
  35. ↵
    1. Peng J-B,
    2. Luo C-H,
    3. Wang Y-C,
    4. Huang W-H,
    5. Chen Y,
    6. Zhou H-H, and
    7. Tan Z-R
    (2014) Validation of a liquid chromatography-electrospray ionization-tandem mass spectrometry method for determination of all-trans retinoic acid in human plasma and its application to a bioequivalence study. Molecules 19:1189–1200.
    OpenUrl
  36. ↵
    1. Phuphanich S,
    2. Scott C,
    3. Fischbach AJ,
    4. Langer C, and
    5. Yung WK
    (1997) All-trans-retinoic acid: a phase II Radiation Therapy Oncology Group study (RTOG 91-13) in patients with recurrent malignant astrocytoma. J Neurooncol 34:193–200.
    OpenUrlCrossRefPubMed
  37. ↵
    1. Regazzi MB,
    2. Iacona I,
    3. Gervasutti C,
    4. Lazzarino M, and
    5. Toma S
    (1997) Clinical pharmacokinetics of tretinoin. Clin Pharmacokinet 32:382–402.
    OpenUrlPubMed
  38. ↵
    1. Rigas JR,
    2. Miller VA,
    3. Zhang ZF,
    4. Klimstra DS,
    5. Tong WP,
    6. Kris MG, and
    7. Warrell Jr RP
    (1996) Metabolic phenotypes of retinoic acid and the risk of lung cancer. Cancer Res 56:2692–2696.
    OpenUrlAbstract/FREE Full Text
  39. ↵
    1. Rigas JR,
    2. Francis PA,
    3. Muindi JR,
    4. Kris MG,
    5. Huselton C,
    6. DeGrazia F,
    7. Orazem JP,
    8. Young CWWRJ, and
    9. Warrell Jr RP
    (1993) Constitutive variability in the pharmacokinetics of the natural retinoid, all-trans-retinoic acid, and its modulation by ketoconazole. J Natl Cancer Inst 85:1921–1926.
    OpenUrlAbstract/FREE Full Text
  40. ↵
    1. Rodgers T and
    2. Rowland M
    (2006) Physiologically based pharmacokinetic modelling 2: predicting the tissue distribution of acids, very weak bases, neutrals and zwitterions. J Pharm Sci 95:1238–1257.
    OpenUrlCrossRefPubMed
  41. ↵
    1. Russo D,
    2. Regazzi M,
    3. Sacchi S,
    4. Visani G,
    5. Lazzarino M,
    6. Avvisati G,
    7. Pelicci PG,
    8. Dastoli G,
    9. Grandi C,
    10. Iacona I, et al.
    (1998) All-trans retinoic acid (ATRA) in patients with chronic myeloid leukemia in the chronic phase. Leukemia 12:449–454.
    OpenUrlPubMed
  42. ↵
    1. Torres-Molina F,
    2. Cárcel-Trullols J,
    3. Araico A, and
    4. Peris JE
    1. Saadeddin A
    ,Torres-Molina F, Cárcel-Trullols J, Araico A, and Peris JE (2004) Pharmacokinetics of the time-dependent elimination of all-trans-retinoic acid in rats. AAPS PharmSci 6:E1.
    OpenUrlCrossRefPubMed
  43. ↵
    1. Saiag P,
    2. Pavlovic M,
    3. Clerici T,
    4. Feauveau V,
    5. Nicolas JC,
    6. Emile D, and
    7. Chastang C
    (1998) Treatment of early AIDS-related Kaposi’s sarcoma with oral all-trans-retinoic acid: results of a sequential non-randomized phase II trial. Kaposi’s Sarcoma ANRS Study Group. Agence Nationale de Recherches sur le SIDA. AIDS 12:2169–2176.
    OpenUrlCrossRefPubMed
  44. ↵
    1. Sandberg JA,
    2. Eckhoff C,
    3. Nau H, and
    4. Slikker Jr W
    (1994) Pharmacokinetics of 13-cis-, all-trans-, 13-cis-4-oxo-, and all-trans-4-oxo retinoic acid after intravenous administration in the cynomolgus monkey. Drug Metab Dispos 22:154–160.
    OpenUrlAbstract
  45. ↵
    1. Shirasaka Y,
    2. Sager JE,
    3. Lutz JD,
    4. Davis C, and
    5. Isoherranen N
    (2013) Inhibition of CYP2C19 and CYP3A4 by omeprazole metabolites and their contribution to drug-drug interactions. Drug Metab Dispos 41:1414–1424.
    OpenUrlAbstract/FREE Full Text
    1. Smith MA,
    2. Adamson PC,
    3. Balis FM,
    4. Feusner J,
    5. Aronson L,
    6. Murphy RF,
    7. Horowitz ME,
    8. Reaman G,
    9. Hammond GD, and
    10. Fenton RM
    (1992) Phase I and pharmacokinetic evaluation of all-trans-retinoic acid in pediatric patients with cancer. J Clin Oncol 10:1666–73.
    OpenUrlAbstract/FREE Full Text
    1. Takitani K,
    2. Tamai H,
    3. Morinobu T,
    4. Kawamura N,
    5. Miyake M,
    6. Fujimoto T, and
    7. Mino M
    (1995) Pharmacokinetics of All- trans Retinoic Acid in Pediatric Patients with Leukemia. Japanese J Cancer Res 86:400–405.
    OpenUrl
  48. ↵
    1. Tay S,
    2. Dickmann L,
    3. Dixit V, and
    4. Isoherranen N
    (2010) A comparison of the roles of peroxisome proliferator-activated receptor and retinoic acid receptor on CYP26 regulation. Mol Pharmacol 77:218–227.
    OpenUrlAbstract/FREE Full Text
  49. ↵
    1. Thatcher JE,
    2. Buttrick B,
    3. Shaffer SA,
    4. Shimshoni JA,
    5. Goodlett DR,
    6. Nelson WL, and
    7. Isoherranen N
    (2011) Substrate specificity and ligand interactions of CYP26A1, the human liver retinoic acid hydroxylase. Mol Pharmacol 80:228–239.
    OpenUrlAbstract/FREE Full Text
  50. ↵
    1. Thatcher JE,
    2. Zelter A, and
    3. Isoherranen N
    (2010) The relative importance of CYP26A1 in hepatic clearance of all-trans retinoic acid. Biochem Pharmacol 80:903–912.
    OpenUrlCrossRefPubMed
  51. ↵
    1. Thudi NR,
    2. Shrivastav VK,
    3. Monif T,
    4. Khuroo A,
    5. Gurule S,
    6. Partani PO,
    7. Tandon M, and
    8. Mathur R
    (2011) Pharmacokinetic and bioequivalence study of endogenous compound tretinoin 10 mg capsules in healthy volunteers by base line correction approach. Clin Res Regul Aff 28:68–73 DOI: 10.3109/10601333.2011.597762.
    OpenUrl
  52. ↵
    1. Topletz AR,
    2. Thatcher JE,
    3. Zelter A,
    4. Lutz JD,
    5. Tay S,
    6. Nelson WL, and
    7. Isoherranen N
    (2012) Comparison of the function and expression of CYP26A1 and CYP26B1, the two retinoic acid hydroxylases. Biochem Pharmacol 83:149–163.
    OpenUrlCrossRefPubMed
  53. ↵
    1. Topletz AR,
    2. Tripathy S,
    3. Foti RS,
    4. Shimshoni JA,
    5. Nelson WL, and
    6. Isoherranen N
    (2015) Induction of CYP26A1 by metabolites of retinoic acid: evidence that CYP26A1 is an important enzyme in the elimination of active retinoids. Mol Pharmacol 87:430–441.
    OpenUrlAbstract/FREE Full Text
  54. ↵
    1. Wang CC,
    2. Campbell S,
    3. Furner RL, and
    4. Hill DL
    (1980) Disposition of all-trans- and 13-cis-retinoic acids and n-hydroxyethylretinamide in mice after intravenous administration. Drug Metab Dispos 8:8–11.
    OpenUrlAbstract/FREE Full Text
Back to top

In this issue

Download PDF
Article Alerts
Email Article
Citation Tools

Research ArticleMetabolism, Transport, and Pharmacogenomics

PBPK Model of atRA Disposition and Interaction in Humans

Jing Jing, Cara Nelson, Jisun Paik, Yoshiyuki Shirasaka, John K. Amory and Nina Isoherranen
Journal of Pharmacology and Experimental Therapeutics May 1, 2017, 361 (2) 246-258; DOI: https://doi.org/10.1124/jpet.117.240523
del.icio.us logo Digg logo Reddit logo Twitter logo CiteULike logo Facebook logo Google logo Mendeley logo

Jump to section

Advertisement