Physiologically based pharmacokinetic modeling and simulation to predict drug–drug interactions of ivosidenib with CYP3A perpetrators in patients with acute myeloid leukemia
Chandra Prakash1 · Bin Fan1 · Alice Ke2 · Kha Le1 · Hua Yang1
Abstract
Purpose Develop a physiologically based pharmacokinetic (PBPK) model of ivosidenib using in vitro and clinical PK data from healthy participants (HPs), refine it with clinical data on ivosidenib co-administered with itraconazole, and develop a model for patients with acute myeloid leukemia (AML) and apply it to predict ivosidenib drug–drug interactions (DDI). Methods An HP PBPK model was developed in Simcyp Population-Based Simulator (version 15.1), with the CYP3A4 component refined based on a clinical DDI study. A separate model accounting for the reduced apparent oral clearance in patients with AML was used to assess the DDI potential of ivosidenib as the victim of CYP3A perpetrators.
Results For a single 250 mg ivosidenib dose, the HP model predicted geometric mean ratios of 2.14 (plasma area under concentration–time curve, to infinity [AUC0-∞]) and 1.04 (maximum plasma concentration [Cmax]) with the strong CYP3A4 inhibitor, itraconazole, within 1.26-fold of the observed values (2.69 and 1.0, respectively). The AML model reasonably predicted the observed ivosidenib concentration–time profiles across all dose levels in patients. Predicted ivosidenib geo- metric mean steady-state AUC0-∞ and Cmax ratios were 3.23 and 2.26 with ketoconazole, and 1.90 and 1.52 with fluconazole, respectively. Co-administration of the strong CYP3A4 inducer, rifampin, predicted a greater DDI effect on a single dose of ivosidenib than on multiple doses (AUC ratios 0.35 and 0.67, Cmax ratios 0.91 and 0.81, respectively).
Conclusion Potentially clinically relevant DDI effects with CYP3A4 inducers and moderate and strong inhibitors co-administered with ivosidenib were predicted. Considering the challenges of conducting clinical DDI studies in patients, this PBPK approach is valuable in ivosidenib DDI risk assessment and management.
Keywords Ivosidenib · Drug–drug interactions · Physiologically based pharmacokinetic model · CYP3A perpetrators · Acute myeloid leukemia
Introduction
Ivosidenib (AG-120) is a potent, targeted, orally active, small molecule inhibitor of mutant isocitrate dehydroge- nase 1 (mIDH1) [1]. mIDH1 has been identified in several hematologic and solid malignancies, including acute myeloid leukemia (AML) and glioma. Ivosidenib inhibits produc- tion of the oncometabolite D-2-hydroxyglutarate by mIDH1, restoring cellular differentiation in pre-clinical models of mIDH1 cancers [1]. In a phase I trial, ivosidenib mono- therapy showed tolerability and clinical activity, including complete remissions, at a daily dose of 500 mg in patients with relapsed or refractory (R/R) AML [2] and with newly diagnosed AML [3]. Ivosidenib is approved in the United States for treatment of adults with a susceptible IDH1 muta- tion with newly diagnosed AML aged ≥ 75 years or ineligi- ble for intensive chemotherapy, and those with R/R AML; development is ongoing for this and other malignancies.
Pharmacokinetic (PK) analysis in patients with advanced hematologic malignancies showed that oral ivosidenib was rapidly absorbed, with a mean half-life (t½) of 72–138 h after a single dose, and reached steady state within 14 days of repeat dosing in patients with AML. A less than dose-proportional increase of area under the plasma drug concentration–time curve (AUC) and maximum plasma concentration (Cmax) was observed with increasing doses (100 mg twice daily [BID] to 1200 mg once daily [QD] dose) [2, 4] as a result of autoinduc- tion of drug clearance. In the radiolabeled human absorption, metabolism, and excretion (AME) study, unchanged [14C] ivosidenib on average accounted for ~ 9.92 and ~ 67.4% of the total recovered dose in urine and feces, respectively [5]. These data suggest that ≥ 32.6% of the dose was absorbed from the suspension dose.
Ivosidenib is slowly metabolized and ~ 65% of the absorbed dose was eliminated by metabolism, largely by oxidation, with minor contributions from N-dealkylation and hydrolytic path- ways [5]. Additional phase I studies in healthy participants (HPs) have described the effects of food and itraconazole on the PK of a single ivosidenib dose [6]. Co-administration of the strong cytochrome P450 (CYP) 3A4 inhibitor itracona- zole (200 mg QD on days 1–18) with a single ivosidenib dose (250 mg on day 5) increased ivosidenib AUC by 169%, but had no effect on Cmax [6]. Comparison of PK profiles from the above studies indicated that the observed apparent oral clear- ance (CL/F) following administration of a single 500 mg dose in HPs was 1.4- to 2.1-fold higher than observed in patients with AML.
Here, we present studies identifying the P450 enzyme(s) involved in the oxidative metabolism of ivosidenib in human liver, and the development of a physiologically based PK (PBPK) model of ivosidenib using the results from these in vitro studies and clinical PK data from HPs. The model was further verified and refined using results from the clinical drug–drug interaction (DDI) study of ivosidenib co-administered with itra- conazole [6]. It was assumed that the absorption of ivosidenib was similar between HPs and patients with AML at the same dose level [6, 7]. The observed reduced oral clearance (CL/F) in patients was attributable to a potential disease or age effect on the systemic clearance (CL) of the drug. To account for the reduced CL/F in patients with AML, a separate model was gen- erated for these patients. The AML model was then applied to simulate the victim DDI potential of ivosidenib (with CYP3A4 inhibitors and inducers) at steady state. These results helped to define the United States Food and Drug Administration (US FDA)-approved language for the product label regarding DDI and dosing recommendations for ivosidenib.
Materials and methods
Materials
Ivosidenib was synthesized by Agios Pharmaceuticals, Inc., and [14C]ivosidenib was from Moravek Biochemicals, Inc. (Brea, CA). The specific radioactivity for [14C] ivosidenib is 58.1 mCi/mmol with > 98% radiochemical purity.
Pooled human liver microsomes (HLM) (20 samples) and human complimentary DNA (cDNA)-expressed CYP enzymes (CYP1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1, 3A4, and 3A5) were from BD Biosciences (Woburn, MA). Cryopreserved human hepatocytes and pooled HLM (16 samples) were from XenoTech, LLC (Lenexa, KS). Benzyl- nirvanol, CYP3cide, diethyldithiocarbamate, ketoconazole, thiotepa, tranylcypromine, β-nicotinamide adenine phos- phate (β-NADPH), dimethyl sulfoxide, glucose-6-phosphate, and glucose-6-phosphate dehydrogenase were from Sigma- Aldrich (St Louis, MO); montelukast from Sequoia Research Products; and troleandomycin from Enzo Life Sciences (Exeter, UK). Liquid scintillation cocktails and 96-deep well LumaPlates were from PerkinElmer (Waltham, MA). Other reagents were analytical or American Chemical Society rea- gent grade and from Fisher Scientific (Hampton, NH).
In vitro CYP reaction phenotyping
In vitro metabolism
The metabolism of [14C]ivosidenib was examined using pooled HLM (20 samples) and human cDNA-expressed CYP enzymes. Incubation mixtures contained pooled HLM (1 mg/mL), [14C]ivosidenib (1 or 10 μM), NADPH (1 mM), and MgCl2 (15 mM) in phosphate buffer (0.1 M, pH 7.4). Incubations were started with addition of NADPH and car- ried out for 1 h at 37 °C. Additional HLM (1 mg/mL) and NADPH (1 mM) were added and the incubation extended for another hour. Controls were incubation mixtures in the absence of NADPH. Reactions were stopped by the addition of three volumes of ethyl acetate. Samples were vortexed and centrifuged, and the extraction was repeated three times. Pooled supernatants were dried, the residues reconstituted in methanol, and aliquots analyzed by high-performance liquid chromatography (HPLC) and liquid chromatography–mass spectrometry (LC–MS), with the radioactivity of HPLC fractions assessed by a TopCount scintillation plate reader. LC–MS/MS data were acquired on an AB Sciex 4000 QTrap (Foster City, CA, MA) with Turbo IonSpray in positive ion mode using Analyst 1.5.2 software. Details of LC–MS/MS analysis are described in the Online Resource, Sect. 2.
Metabolism of [14C]ivosidenib by human cDNA‑expressed CYP enzymes
Incubation mixtures contained each CYP enzyme in super- somes (0.1 nmol/mL), [14C]ivosidenib (10 μM), MgCl2 (15 mM), and NADPH (1 mM) in 0.2 mL phosphate buffer (0.1 M, pH 7.4). Incubations were started with addition of NADPH and proceeded for 1 h at 37 °C. Additional CYP enzyme (0.1 nmol/mL) and NADPH (1 mM) were added and the incubation was extended for another hour. Controls were the incubation mixtures of CYP3A4 in the absence of NADPH. Samples were analyzed as HLM samples.
Incubations with CYP‑selective inhibitors in HLM
Incubation mixtures contained HLM (1 mg/mL), [14C] ivosidenib (10 μM), MgCl2 (15 mM), and NADPH (1 mM) in 0.1 mL phosphate buffer (0.1 M, pH 7.4) in the presence or absence of a specific CYP inhibitor: furafylline (10 μM) for CYP1A2; tranylcypromine (2 μM) for CYP2A6; thiotepa (50 μM) for CYP2B6; montelukast (3 μM) for CYP2C8; sulfaphenazole (10 μM) for CYP2C9; benzylnirvanol (1 μM) for CYP2C19; quinidine (1 μM) for CYP2D6; diethyldithi- ocarbamate (50 μM) for CYP2E1; ketoconazole (1 μM) and troleandomycin (50 μM) for CYP3A4/5; and CYP3cide (1 μM) for CYP3A4. The metabolism-dependent inhibi- tors (furafylline, troleandomycin, and CYP3cide) were pre-incubated with HLM in the presence of NADPH for 10 min. Incubations were started with addition of NADPH or the substrate and carried out for 1 h at 37 °C. The reac- tion was stopped by the addition of acetonitrile containing the internal standard reserpine. Samples were vortexed and centrifuged, and supernatants analyzed by tandem LC–MS (LC–MS/MS). All experiments were carried out in triplicate.
Ivosidenib Simcyp PBPK model
Simcyp (version 15.1) (Simcyp Ltd., a Certara Company, Sheffield, UK) was used for all modeling and simulations. To recover the multi-exponential disposition kinetics of ivosidenib, the full PBPK model was used. Two PBPK mod- els were developed: one was developed based on the HP data and a separate model was developed based on the data from patients with AML to account for the reduced CL/F in these patients, as shown in Fig. 1.
Input parameters
All physicochemical and PK parameters of ivosidenib used for the PBPK model are summarized in Table 1. A first-order absorption model parameterized with fraction absorbed (fa) and first-order absorption rate constant (ka) based on clinical data was used to describe the absorp- tion kinetics of ivosidenib. The fa value was user defined as 0.89 or 0.59 in HPs for a 250 or 500 mg ivosidenib dose, respectively, and 0.69 in patients with AML for a 500 mg dose (Table 1). These values were determined by matching the observed clinical PK data of these doses in HPs or patients as described below. In the human AME study, fa was estimated to be 0.33 [5] from the formula- tion made by the dry blending method, as opposed to the wet granulation method used for tablets in other clinical studies. The exposure (AUC0-∞) of ivosidenib with tablets Assumed to be the same as the relative bioavailability of the film- coated tablet formulation at 500 mg determined from the food effect study conducted in the HP in the fed state [6] was 1.8-fold higher in HPs [6] than in the human AME study at the same dose level of ~ 500 mg, suggesting lower absorption with oral powder suspension compared with tablets. Therefore, the fa values used in the model for HPs and patients were estimated by multiplying the fa from the human AME study to those of ratios of respective plasma exposures. The full PBPK model was used with a Simcyp- predicted volume of distribution at steady state of 2.45 L/kg. Ivosidenib is predominantly eliminated by metabolism in humans with only ~ 9.92% of the dose excreted as unchanged in the urine [5]. Based on these data, the renal clearance was calculated as 0.537 L/h in HPs. The renal clearance in patients with AML was predicted to be 0.36 L after correcting for the age effect on renal function, which is consistent with the estimated median glomerular filtra- tion rate value in the patients [7].
The observed CL/F values following the administration of a single oral dose of 500 mg in HPs were, on average, 1.4- to 2.1-fold higher than those observed in the patient population. Input parameters for intrinsic metabolic clear- ance (CLint) were back-calculated from the observed mean CL/F using the well-stirred liver model (Eq. 1); fraction metabolized (Fm) for individual CYPs; average popula- tion values for liver weight (1648 g in HPs vs. 1460 g in patients); microsomal protein per gram of liver (39.8 in HPs vs. 29.1 mg in patients); and hepatic CYP enzyme abundance (137 pmol/mg for CYP3A4): CLpo, the observed CL/F (2.31 L/h for HPs and 1.63 L/h for patients with AML); B:P, the concentration ratio of drug in blood to plasma (Table 1); fub, fraction of unbound drug in blood (calculated from unbound frac- tion in plasma [fup]/B:P, 1); QH, blood flow in the hepatic vein (90 L/h in HPs vs. 68 L/h in patients); fg, the fraction escaping first-pass metabolism in the gut (assumed to be one due to low CL of the drug); CLR, the renal clearance; uptake, a factor that accounts for any active hepatic uptake (default value = 1).
Ivosidenib induced CYP3A4 at both the messenger RNA (mRNA) and activity level in cultured human hepat- ocytes. On average, the observed maximum effect (Emax) and half-maximal effective concentration (EC50) values for CYP3A4 mRNA were 78.2-fold and 12.5 µM, respec- tively (data not shown). The observed Emax for CYP3A4 mRNA was subsequently calibrated against the in vitro rifampin Emax (58.2-fold), as well as the in vivo rifampin Emax (16-fold), to obtain an Emax of 21.25-fold. CYP3A4 induction in patients with AML was optimized by use of 4β-hydroxy-cholesterol (4β-OHC) biomarker data from the clinical study to verify the predicted hepatic CYP3A4 induction effect after multiple ivosidenib doses [8]. The measured in vitro half-maximal induction concentration (IndC50) value (12.5 µM) was directly used in the single- dose HP simulations, and an in vivo CYP3A4 IndC50 value (2.5 µM), optimized by use of 4β-OHC biomarker data, was used in all AML multiple-dose simulations, including model development and application.
PBPK modeling of ivosidenib PK and DDI
The Simcyp “Healthy Volunteer” population was used for HP simulations. The proportion of female participants was set as 0.5. Ten trials of ten participants were simulated for each dosing regimen. Two different ivosidenib Simcyp models were developed to predict single- or multiple-dosing regimens in HPs and in patients with AML, as summa- rized above. In the base model, 98% of the metabolism was assigned to CYP3A4. Initial simulations involving the devel- oped base model for ivosidenib led to the underprediction of the DDI effects with itraconazole. Thus, the base model for ivosidenib was refined by assigning 100% of the metabolism to CYP3A4. This corresponds to an fm and fa of 67 and 33%, respectively.
The predictions of plasma drug concentration–time pro- files, clearance, and DDI in the patient population were per- formed in the Simcyp Simulator using a modified Sim-NEur Caucasian population to match the age distribution of the oncology population. The single modification was to change the default age distribution to account for the older age of the disease population used in clinical studies compared with the default file settings, an important change, because increased age is associated with reduced drug clearance. In the Simcyp Simulator, a Weibull distribution is implemented to generate the age of virtual participants (Eq. 2): library values provided in the Simcyp Simulator. To assess the effect of ketoconazole, the Simcyp ketoconazole com- pound files were developed as “fit-for-purpose” models, which are described by relevant CL/F rather than in vitro metabolism data. Because the ketoconazole compound files do not have a CYP3A4 component, the ivosidenib treatment does not reduce the exposure of ketoconazole, representing the maximal DDI effects with ivosidenib due to CYP3A4 inhibition. The maximum induction (i.e., maximum fold increase over vehicle control + 1) of 16 for rifampin is used in current Simcyp versions and has been found to be more predictive of rifampin clinical DDI with CYP3A substrates [11, 12]. The efavirenz model was recently published [13].
Results
In vitro CYP profiling
[14C]Ivosidenib was incubated with HLM in the presence of NADPH. Radioactive metabolites were analyzed by HPLC and four peaks were observed (data not shown). These peaks were small and their radioactivity accounted for only 2.59% of the total integrated radioactivity injected for the HPLC analysis. Further analysis of the radioactive peaks by LC–MS/MS suggested that these peaks corresponded to monohydroxylated [14C]ivosidenib metabolites, named M1, M2, M3, and M4. The formation of M1–M4 was linear up to 60 min, and microsomal protein concentrations up to 1 mg/ mL. The enzyme kinetic parameters Km, Vmax, and Vmax/ Km ratio (CLint, intrinsic clearance) values for the formation of M1–M4 are shown in Online Resource, Table 1.
The contribution of CYP enzymes to the metabolism of ivosidenib was examined using recombinant CYP1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1, 2J2, 3A4, and 3A5 enzymes. Radioactive M1–M4 peaks emerged in the incu- bation mixture containing CYP3A4, but not with other enzymes. The formation of M1–M4 was linear with micro- somal protein concentrations up to 0.05 nmol/mL. These
The distribution of ages for male participants set as a Weibull distribution used values of α (6.12) and β (67.6). The distribution of ages for female participants set as a Weibull distribution used the values of α (9.49) and β (72.1). Any other changes to population parameters that may occur in the oncology population were not accounted for in the model. Default Simcyp parameters for creating a virtual HP (physiological parameters including liver volume, blood flows, and enzyme abundances) have been described previ- ously [9, 10]. The CYP abundance values for the virtual patient population are kept the same as those used in the Sim-NEurCaucasian population.
The input parameters for itraconazole, ketoconazole, fluconazole, fluvoxamine, rifampin, and efavirenz were the results suggested that CYP3A4 is involved in the formation of M1–M4.
To further identify the CYP enzymes responsible for M1–M4 formation, 10 µM of [14C]ivosidenib was incubated in the presence of specific CYP inhibitors. The formations of M1–M4 were inhibited by the CYP3A4/5 inhibitor keto- conazole (1 µM) up to 97.9, 96.5, 94.8, and 91.7%, respectively, and troleandomycin (50 µM) up to 92.7, 87.8, 82.2, and 82.4%, respectively, and by the CYP3A4 inhibitor, CYP- 3cide (1 µM), up to 95.0, 93.5, 91.5, and 90.0%, respec- tively. The formations M3 and M4 were inhibited by the CYP2B6 inhibitor thiotepa (50 µM) up to 11.3 and 11.1%, respectively. The formations of M1–M4 were inhibited by the CYP2C8 inhibitor montelukast (3 µM) up to 16.6, 12.9, 7.1, and 13.6%, respectively. There was no substantial inhi- bition with other CYP inhibitors.
Using the relative activity factor, the relative contribu- tion of CYP3A4 to the formation of M1–M4 was estimated to be ~ 100, ~ 66, ~ 100, and ~ 100%, respectively. These results combined with chemical inhibition data suggested that CYP3A4 contributes to the majority of the metabolism of ivosidenib, with a minor contribution from other enzymes such as CYP2B6 and CYP2C8.
Optimization of CYP3A4 induction parameters
4β-OHC is an endogenous biomarker of hepatic CYP3A4 activity. In the phase I study in patients with AML receiving 200–1200 mg ivosidenib daily, baseline and induced levels of 4β-OHC were examined. Mean 4β-OHC plasma levels increased up to ~ 3- to 4-fold from baseline after 28 days of treatment with daily 200–1200 mg ivosidenib [4]. The magnitude of CYP3A4 induction (146–244% increase in 4β-OHC plasma concentration) following the administra- tion of 500 mg ivosidenib is lower than or similar to the strong CYP3A4 inducer rifampin: literature data showed 4β-OHC increased 290% after 14 days of treatment with 600 mg rifampin QD [11]. The observed 4β-OHC percent change due to ivosidenib treatment was then mapped to the corresponding midazolam AUC ratio, based on a published relationship: mean 4β-OHC levels were increased by ~ 199% following treatment with 500 mg ivosidenib, correspond- ing to a midazolam AUC ratio of 0.15. Mean 4β-OHC levels increased by 266–279% following treatment with daily 800–1200 mg ivosidenib, which corresponded to a midazolam AUC ratio of ~ 0.10. Sensitivity analysis results showed that CYP3A4 IndC50 in the range of 2.0–3.0 µM allowed the recovery of midazolam AUC ratio of 0.15. Additional sensitivity analysis on CYP3A4 IndC50 showed that CYP3A4 IndC50 set to ~ 2.5 µM allowed the recovery of midazolam AUC ratio of 0.10, observed with 1200 mg ivosidenib (Online Resource, Table 2 and Fig. 1). Thus, in the final model, a maximum induction (Indmax) of 21.25-fold and an IndC50 of 2.5 µM were used, compared with the range of the in vitro EC50 values (5.27–22.9 µM).
In the current model, first-pass metabolism in liver and gut has been minimal due to the observed low clearance of the drug. The simulated effect of induction of CYP3A4 in the liver and in the gut is shown following ivosidenib treat- ment for 19 days in patients in Online Resource Fig. 2.
Simulation of ivosidenib plasma profiles in HPs and patients with AML
Simulations using different doses of ivosidenib were gener- ated using the developed HP and AML models; the predicted concentration–time profiles and key PK parameters (AUC, Cmax, and time at which the maximum plasma concentration of drug is reached [tmax]) were compared with the observa- tions from the clinical studies.
Simulated and observed concentration–time profiles fol- lowing a single 500 mg ivosidenib dose in HPs in the fasted state are shown in Fig. 2a. The predicted mean AUC and Cmax values were 0.99- and 1.36-fold of the observed values, respectively.
Simulated and observed concentration–time profiles of ivosidenib following 500 mg QD for 19 days (single dose on day 1 with plasma sampling for 72 h followed by QD on days 4–19) in patients with AML are shown in Fig. 2b. Predicted and observed PK parameters are summarized in Table 2. For all doses (100 mg BID, 300 mg QD, 800 mg QD, and 1200 mg QD), the predictions of geometric mean steady- state (day 15) AUC and Cmax values were within 0.78- and 1.19-fold, and 0.76- and 1.58-fold of the observed values, respectively (Table 2). At the 500 mg therapeutic dose, pre- dicted day 3 AUC and Cmax values were within 0.78- and 1.03-fold of the observations, respectively; the predicted day 15 AUC and Cmax values were within 0.83- and 1.11-fold of the observations, respectively. Because the target dose level for ivosidenib is 500 mg, the developed model was deemed robust for the subsequent model application.
Predictions of the victim DDI potential of ivosidenib with CYP3A4 inhibitors
The models were applied to assess the victim DDI poten- tial of ivosidenib in HPs and patients with AML receiv- ing ivosidenib with strong (itraconazole and ketoconazole), moderate (fluconazole), and mild (fluvoxamine) CYP3A inhibitors.
Simulated and observed concentration–time profiles and exposure (Cmax and AUC) of ivosidenib after a single oral dose (250 mg) in the absence and presence of multiple daily doses of itraconazole (200 mg QD for 18 days) using the ivosidenib HP model (fasted) are shown in Fig. 3. Predicted and observed Cmax and AUC values in the control arm were reasonably consistent with the observed data. However, the inhibitory effect on the geometric mean AUC ratio of ivosidenib was slightly underpredicted: the predicted versus observed ratios were 1.94 and 2.69, respectively (Fig. 3a). One of the factors that may explain the underprediction of the inhibitory effects of itraconazole is that alternative ivosidenib clearance mechanisms are also inhibited by itra- conazole. Ivosidenib was identified in vitro as a P-glycopro- tein (P-gp) substrate in Caco-2 cells. However, the observed DDI data showed that itraconazole treatment had no effect on the Cmax of ivosidenib, suggesting that the involvement of intestinal P-gp in ivosidenib disposition in vivo is likely to be minimal.
Itraconazole has been shown to inhibit P-gp activity, resulting in a digoxin (0.5 mg orally) AUC0-72 h ratio of 1.52 [14]. Furthermore, itraconazole treatment (200 mg tablets QD for 6 days) reduced digoxin renal clearance from 9.66 to 7.7 L/h (i.e., 20% reduction) in HPs. Thus, a separate simulation was conducted in which the renal clearance of ivosidenib (0.537 L/h) was reduced by 20%. This corre- sponds to a reduced renal clearance of 0.43 L/h. Simulated and observed concentration–time profiles of ivosidenib in Caucasian participants, incorporating the inhibitory effects of itraconazole on renal clearance, are shown in Fig. 3b. Predicted versus observed AUC ratios of ivosidenib were 2.14 and 2.69, respectively. These values were closer to the observed data and the predicted ivosidenib AUC ratio was within 1.26-fold of the observed value; this ivosidenib model was applied to subsequent simulations without further refinement.
The verified ivosidenib model was applied to assess the impact of co-administration of the strong CYP3A4 inhibi- tor itraconazole on the PK of ivosidenib following a multi- ple daily dose of 500 mg in patients with AML. Simulated concentration–time profiles of ivosidenib in patients with AML receiving multiple oral doses of ivosidenib for 15 days in the presence and absence of itraconazole treatment (200 mg tablets QD for 15 days, in the fed state) are shown in Fig. 4a. Predicted ivosidenib AUC and Cmax ratios were 1.44 (95% confidence interval [CI] 1.41–1.48) and 1.29 (95% CI 1.26–1.31), respectively, following concurrent itracona- zole treatment (Table 3). These data suggest that the DDI effect on the kinetics of multiple-dose ivosidenib (AUC ratio 1.44) is smaller than that of single-dose ivosidenib (AUC ratio 2.14). This result is plausible as the strong CYP3A4 induction effect as a result of multiple-dose administra- tion of ivosidenib led to much lower itraconazole and its hydroxyl metabolite exposures, and lower inhibitory effects on CYP3A4.
Because itraconazole is a CYP3A4 substrate, the DDIs observed with itraconazole might not represent the mag- nitude of DDIs between ivosidenib and a strong CYP3A4 inhibitor that is not significantly metabolized by CYP3A4. Therefore, additional simulations were conducted with keto- conazole, the most potent CYP3A4 inhibitor, to assess the worst-case scenario of the DDI effects with ivosidenib due to CYP3A4 inhibition. Simulated concentration–time pro- files of ivosidenib in patients with AML receiving multi- ple oral doses of ivosidenib for 15 days in the presence and absence of ketoconazole treatment (200 mg BID for 15 days, in the fed state) are shown in Fig. 4b. Predicted ivosidenib AUC and Cmax geometric mean (GM) ratios were 2.46 (95% CI 2.33–2.60) and 1.03 (95% CI 1.02–1.03), respectively, with a single 500 mg dose and 3.23 (95% CI 3.05–3.41) and 2.26 (95% CI 2.17–2.35), respectively, with multiple 500 mg doses following concurrent ketoconazole treatment (Table 3).
The impact of the co-administration of fluconazole, a moderate CYP3A4 inhibitor, on the PK of ivosidenib following single- and multiple-dose administration of 500 mg in patients with AML was predicted. Simulated concentration–time profiles of ivosidenib in patients with AML receiving multiple oral doses of 500 mg ivosidenib for 15 days in the presence and absence of fluconazole treatment (400 mg on day 1 followed by 200 mg QD for 14 days) are shown in Fig. 4c. Predicted ivosidenib AUC and Cmax GM ratios were 1.73 (95% CI 1.68–1.78) and 1.02 (95% CI 1.02–1.02), respectively, with a sin- gle 500 mg dose and 1.90 (95% CI 1.86–1.94) and 1.52 (95% CI 1.50–1.55), respectively, with multiple 500 mg doses following concurrent fluconazole treatment (Table 3). The simulated fluconazole DDI effects on mul- tiple-dose ivosidenib were slightly greater than the effects on single-dose ivosidenib. This is because when ivosidenib was dosed to steady state, the fm value became > 64% due to autoinduction; therefore, the predicted AUC ratio due to CYP3A4 inhibition became higher. The exception was itraconazole, as the exposure of itraconazole itself was affected by ivosidenib due to induction of CYP3A4.
A weak CYP3A4 inhibitor, fluvoxamine, was predicted to have negligible DDI effects on both single- and multiple- dose ivosidenib (Table 3). Simulated and observed plasma concentration–time profiles of ivosidenib in patients with AML receiving multiple oral doses of 500 mg ivosidenib for 15 days in the presence and absence of fluvoxamine treat- ment (100 mg BID for 15 days) are shown in Fig. 4d. Pre- dicted ivosidenib AUC and Cmax ratios were 1.11 and 1.01, respectively, with a single 500 mg dose and 1.11 and 1.08, respectively, with multiple 500 mg doses following concur- rent fluvoxamine treatment (Table 3).
Predictions of the victim DDI potential of ivosidenib with CYP3A4 inducers
The developed AML model for ivosidenib was applied to assess the impact of the co-administration of the strong CYP3A4 inducer, rifampin, and a moderate inducer, efa- virenz, on the PK of ivosidenib following single- and mul- tiple-dose administration.
Simulated concentration–time profiles of ivosidenib in patients with AML receiving multiple 500 mg oral doses for 30 days in the presence and absence of rifampin treat- ment (600 mg QD on days 15–30) are shown in Fig. 4e. Predicted ivosidenib AUC and Cmax GM ratios were 0.35 (95% CI 0.33–0.37) and 0.91 (95% CI 0.90–0.92), respectively, with a single 500 mg dose and 0.67 (95% CI 0.63–0.71) and 0.81 (95% CI 0.79–0.83), respectively, with multiple 500 mg doses following concurrent rifampin treatment (Table 3).
A greater DDI effect on the kinetics of a single dose than that of multiple-dose ivosidenib was predicted (AUC ratios 0.35 and 0.67, respectively). This result is plausible as the CYP3A4 activity has been induced to a significant extent as a result of multiple-dose administration of ivosidenib; thus, the inducible effect as a result of rifampin treatment appears to be diminished as the magnitude of enzyme induction is inversely related to the baseline levels of enzyme [15].
The impact of the co-administration of the moderate CYP3A4 inducer efavirenz on the PK of ivosidenib fol- lowing single- and multiple-dose administration was also assessed. The CYP3A4 induction data have been included in the efavirenz model (Online Resource, Table 3). Simulated concentration–time profiles of ivosidenib in patients with AML receiving multiple oral doses of 500 mg ivosidenib for 30 days in the presence and absence of efavirenz treatment (600 mg QD on days 15–30) are shown in Fig. 4f. Predicted ivosidenib AUC and Cmax GM ratios were 0.50 (95% CI 0.47–0.52) and 0.94 (95% CI 0.94–0.95), respectively, with a single 500 mg dose and 0.89 (95% CI 0.86–0.92) and 0.93 (95% CI 0.92–0.95), respectively, with multiple 500 mg doses following concurrent efavirenz treatment (Table 3).
Discussion
The use of PBPK modeling and simulation at different stages of drug development and in regulatory submissions has become increasingly apparent from the literature and recent product labels [16–18], supporting that a well-qualified PBPK model can be used to estimate the risk of concomi- tant medications in clinical trials, even before in vivo DDI studies [19, 20]. Based on the use of PBPK modeling in recent regulatory submissions, there was a notable trend in the predictions of enzyme-based DDIs (60%), especially for inhibitors and/or inducers of CYP3A4 and the dose recom- mendations [21]. The modeling and simulation of ivosidenib presented in this manuscript was influential for the dosing strategy when co-administered with perpetrators of CYP3A4 in the product label for patients.
A slow metabolic turnover of ivosidenib was observed in the incubation with HLM and recombinant CYP enzymes. Incubation of HLM with ivosidenib and specific CYP inhibi- tors followed by HPLC and LC–MS identified CYP enzymes responsible for metabolizing ivosidenib. In conjunction with these results, the present data suggest that CYP3A4 plays a major role in the elimination of ivosidenib. Although the CL/F values following the administration of a single oral dose of 500 mg in HPs in the fasted condition were, on aver- age, 1.4-fold to 2.1-fold higher than patients with AML, the absorption of ivosidenib is expected to be similar between two populations. This assumption is supported by the fol- lowing observations: (1) the in vitro dissolution profiles for the coated and uncoated tablets were superimposable, and the disintegration times for the core tablets were rapid (~ 2 min); and (2) ivosidenib has high passive permeability and the estimated fa value from the human AME study is moderate-to-high. As such, the observed difference in the values for CL/F was mainly attributed to the potential dis- ease or age effect on the CL of the drug [22].
Clinical data obtained with different doses suggest that the absorption of ivosidenib may be dose-dependent. Here, dose-dependent fa values derived from the clinical data were used to simulate the kinetics of ivosidenib at 250 and 500 mg dose levels in HPs. On the other hand, the simula- tions in patients with AML used a fixed fa value based on the 500 mg dose level and were found to reasonably recover the observed systemic exposure of ivosidenib across all dose levels. Notably, it is challenging to reliably estimate dose-dependent fa values from the clinical data obtained in patients with AML due to confounding effects from food intake, as well as the PK variability.
Because the low turnover of ivosidenib in vitro precluded the quantitative estimate of a CLint value, a retrograde approach was used to derive CLint from the observed CL/F of ivosidenib by making assumptions on values of fa, fg, and fraction escaping hepatic clearance (fh). The developed HP model has a CL of 1.85 L/h, with percentages of fm and frac- tion excreted (fe) of 67 and 33%, respectively. The developed model for the patient population has a CL of 1.1 L/h with fm and fe of 64 and 36%, respectively. Ivosidenib metabolism was entirely assigned to CYP3A4. The patient model is asso- ciated with a slightly smaller fmCYP3A4 than the HP model, mainly due to a reduced CL value.
To verify the in vivo contribution of CYP3A4 to the dis- position of ivosidenib, the DDI effect with itraconazole was assessed and compared with the observed data. The inhibi- tory effects on the AUC ratio of ivosidenib was initially underpredicted (predicted 1.94 vs. observed 2.69) when only the inhibition of CYP3A4-mediated pathway was consid- ered. After accounting for the inhibitory effects on the renal clearance of ivosidenib, the predicted and observed Cmax and AUC values were closer to the observed data. Because the predicted ivosidenib AUC ratio was within 1.25-fold of the observed value, the ivosidenib model was applied to subsequent DDI simulations without further refinement. The DDI effects with ivosidenib due to CYP3A4 inhibition were assessed with ketoconazole, which is not assumed to be a CYP3A substrate in the simulations. The maximal DDI effect with ivosidenib due to CYP3A4 inhibition is a 3.23-fold increase in AUC and a 2.26-fold increase in Cmax. A population-PK analysis has identified strong CYP3A4 inhibitors as a significant covariate of drug clearance. Strong concomitant CYP3A4 inhibitors, voriconazole and posaconazole, increased steady-state AUC by 57 and 53%, respectively, from nominal exposures [7]. The population- PK analysis represents an estimation of the real-time effect of these comedications in the phase I hematologic cancer population, and the estimated effects are contained within the range of predicted DDI effects obtained from PBPK analysis [7].
Simulation of the effects of itraconazole or rifampin on the PK of ivosidenib suggested that the DDI effect on multiple-dose ivosidenib is smaller than on single-dose ivosidenib. This could be due to multiple-dose ivosidenib administration causing stronger CYP3A4 induction com- pared with single-dose ivosidenib; high-level CYP3A4 expression may decrease both the exposure of itraconazole and its hydroxyl metabolite, lowering inhibitory effects on CYP3A4 and the induction capability of rifampin, because the magnitude of CYP3A4 induction is inversely corre- lated with the baseline levels of enzyme [15]. A moderate CYP3A4 inhibitor, fluconazole, demonstrated a similar pre- dicted DDI effect as itraconazole, although multiple-dose administration of ivosidenib had no effect on fluconazole exposures. This was probably due to the fact that fluconazole is mainly eliminated through the renal pathway [23]. Keto- conazole also predicted a higher DDI with multiple doses of ivosidenib. In summary, the application of the ivosidenib AML model predicted a greater DDI effect on the exposure of ivosidenib after multiple doses (AUC and Cmax ratios: 3.23 and 2.26) relative to that following a single dose of ivosidenib (AUC and Cmax ratios: 2.46 and 1.03) due to con- current ketoconazole administration. This result is likely to be a consequence of the greater contribution of CYP3A4 to ivosidenib metabolism at steady state, due to autoinduc- tion of CYP3A4 following multiple-dose administration of ivosidenib. Therefore, the inhibitory effect by ketoconazole treatment is also greater at steady state. The maximal DDI effect with ivosidenib due to CYP3A4 inhibition is a 3.23- fold increase in AUC and a 2.26-fold increase in Cmax.
In conclusion, the developed PBPK model reasonably predicted the observed steady-state exposures of ivosidenib as well as the autoinduction effect across all dose levels in patients with AML. Considering the challenges in con- ducting multiple-dose clinical DDI studies of ivosidenib in patients, this PBPK-based DDI approach to assess ivosidenib as a victim is valuable to examine various clinical scenarios with reasonable accuracy. Notably, the case example pre- sented here is the first known example of drug approval by the US FDA during the new drug application process for a CYP3A-substrate drug based on PBPK simulations in lieu of a clinical DDI study with a strong inducer such as rifampin. The magnitude of interaction predicted was important for decisions made regarding dosing recommendations, espe- cially while administered with CYP3A4 inhibitors/inducers, and had an impact on the current US FDA labeling language for ivosidenib.
References
1. Popovici-Muller J, Lemieux RM, Artin E, Saunders JO, Sali- turo FG, Travins J, Cianchetta G, Cai Z, Zhou D, Cui D, Chen P, Straley K, Tobin E, Wang F, David MD, Penard-Lacronique V, Quivoron C, Saada V, de Botton S, Gross S, Dang L, Yang H, Utley L, Chen Y, Kim H, Jin S, Gu Z, Yao G, Luo Z, Lv X, Fang C, Yan L, Olaharski A, Silverman L, Biller S, Su SM, Yen K (2018) Discovery of AG-120 (ivosidenib): a first-in-class mutant IDH1 inhibitor for the treatment of IDH1 mutant cancers. ACS Med Chem Lett 9(4):300–305
2. DiNardo CD, Stein EM, de Botton S, Roboz GJ, Altman JK, Mims AS, Swords R, Collins RH, Mannis GN, Pollyea DA, Donnellan W, Fathi AT, Pigneux A, Erba HP, Prince GT, Stein AS, Uy GL, Foran JM, Traer E, Stuart RK, Arellano ML, Slack JL, Sekeres MA, Willekens C, Choe S, Wang H, Zhang V, Yen KE, Kapsalis SM, Yang H, Dai D, Fan B, Goldwasser M, Liu H, Agresta S, Wu B, Attar EC, Tallman MS, Stone RM, Kantarjian HM (2018) Durable remissions with ivosidenib in IDH1-mutated relapsed or refractory AML. N Engl J Med 378(25):2386–2398
3. Roboz GJ, DiNardo CD, Stein EM, de Botton S, Mims AS, Prince GT, Altman JK, Arellano ML, Donnellan W, Erba HP, Mannis GN, Pollyea DA, Stein AS, Uy GL, Watts JM, Fathi AT, Kantar- jian HM, Tallman MS, Choe S, Dai D, Fan B, Wang H, Zhang V, Yen KE, Kapsalis SM, Hickman D, Liu H, Agresta SV, Wu B, Attar EC, Stone RM (2020) Ivosidenib induces deep durable remissions in patients with newly diagnosed IDH1-mutant acute myeloid leukemia. Blood 135(7):463–471
4. Fan B, Mellinghoff IK, Wen PY, Lowery MA, Goyal L, Tap WD, Pandya SS, Manyak E, Jiang L, Liu G, Nimkar T, Gliser C, Prahl Judge M, Agresta S, Yang H, Dai D (2020) Clinical pharmacoki- netics and pharmacodynamics of ivosidenib, an oral, targeted inhibitor of mutant IDH1, in patients with advanced solid tumors. Invest New Drugs 38(2):433–444
5. Prakash C, Fan B, Altaf S, Agresta S, Liu H, Yang H (2019) Phar- macokinetics, absorption, metabolism, and excretion of [(14)C] ivosidenib (AG-120) in healthy male subjects. Cancer Chemother Pharmacol 83(5):837–848
6. Dai D, Yang H, Nabhan S, Liu H, Hickman D, Liu G, Zacher J, Vutikullird A, Prakash C, Agresta S, Bowden C, Fan B (2019) Effect of itraconazole, food, and ethnic origin on the pharma- cokinetics of ivosidenib in healthy subjects. Eur J Clin Pharmacol 75(8):1099–1108
7. Jiang X, Wada R, Poland B, Kleijn HJ, Fan B, Liu G, Liu H, Kapsalis S, Yang H, Le K (2020) Population pharmacokinetic and exposure-response analyses of ivosidenib (AG-120) in patients with IDH1-mutant advanced hematologic malignancies. Clin Pharmacol Ther (In review)
8. Almond LM, Mukadam S, Gardner I, Okialda K, Wong S, Hat- ley O, Tay S, Rowland-Yeo K, Jamei M, Rostami-Hodjegan A, Kenny JR (2016) Prediction of drug-drug interactions arising from CYP3A induction using a physiologically based dynamic model. Drug Metab Dispos 44(6):821–832
9. Howgate EM, Rowland Yeo K, Proctor NJ, Tucker GT, Rostami- Hodjegan A (2006) Prediction of in vivo drug clearance from in vitro data. I: impact of inter-individual variability. Xenobiotica 36(6):473–497
10. Inoue S, Howgate EM, Rowland-Yeo K, Shimada T, Yamazaki H, Tucker GT, Rostami-Hodjegan A (2006) Prediction of in vivo drug clearance from in vitro data. II: potential inter-ethnic differ- ences. Xenobiotica 36(6):499–513
11. Leil TA, Kasichayanula S, Boulton DW, LaCreta F (2014) Evalu- ation of 4beta-hydroxycholesterol as a clinical biomarker of CYP3A4 drug interactions using a Bayesian mechanism-based pharmacometric model. CPT Pharmacometrics Syst Pharmacol 3(6):e120
12. Wagner C, Pan Y, Hsu V, Sinha V, Zhao P (2016) Predicting the effect of CYP3A inducers on the pharmacokinetics of substrate drugs using physiologically based pharmacokinetic (PBPK) mod- eling: an analysis of PBPK submissions to the US FDA. Clin Pharmacokinet 55(4):475–483
13. Ke A, Barter Z, Rowland-Yeo K, Almond L (2016) Towards a best practice approach in PBPK modeling: case example of developing a unified efavirenz model accounting for induction of CYPs 3A4 and 2B6. CPT Pharmacometrics Syst Pharmacol 5(7):367–376
14. Jalava KM, Partanen J, Neuvonen PJ (1997) Itraconazole decreases renal clearance of digoxin. Ther Drug Monit 19(6):609–613
15. Gorski JC, Vannaprasaht S, Hamman MA, Ambrosius WT, Bruce MA, Haehner-Daniels B, Hall SD (2003) The effect of age, sex, and rifampin administration on intestinal and hepatic cytochrome P450 3A activity. Clin Pharmacol Ther 74(3):275–287
16. Jones HM, Chen Y, Gibson C, Heimbach T, Parrott N, Peters SA, Snoeys J, Upreti VV, Zheng M, Hall SD (2015) Physiologically based pharmacokinetic modeling in drug discovery and develop- ment: a pharmaceutical industry perspective. Clin Pharmacol Ther 97(3):247–262
17. Luzon E, Blake K, Cole S, Nordmark A, Versantvoort C, Berglund EG (2017) Physiologically based pharmacokinetic modeling in regulatory decision-making at the European Medicines Agency. Clin Pharmacol Ther 102(1):98–105
18. Sager JE, Yu J, Ragueneau-Majlessi I, Isoherranen N (2015) Physiologically based pharmacokinetic (PBPK) modeling and simulation approaches: a systematic review of published mod- els, applications, and model verification. Drug Metab Dispos 43(11):1823–1837
19. European Medicines Agency (2012) Guideline on the investiga- tion of drug interactions. https://www.ema.europa.eu/en/docum ents/scientific-guideline/guideline-investigation-drug-interactio ns_en.pdf. Accessed 17 April 2020
20. U.S. Department of Health and Human Services Food and Drug Administration Center for Drug Evaluation and Research (CDER) (2017) Clinical drug interaction studies — study design, data anal- ysis, and clinical implications guidance for industry. https://www. fda.gov/files/drugs/published/Clinical-Drug-Interaction-Studies-%E2%80%94-Study-Design–Data-Analysis–and-Clinical-Impli cations-Guidance-for-Industry.pdf. Accessed 17 April 2020
21. Grimstein M, Yang Y, Zhang X, Grillo J, Huang SM, Zineh I, Wang Y (2019) Physiologically based pharmacokinetic modeling in regulatory science: an update from the U.S. Food and Drug Administration’s Office of Clinical Pharmacology. J Pharm Sci 108(1):21–25
22. Coutant DE, Kulanthaivel P, Turner PK, Bell RL, Baldwin J, Wijayawardana SR, Pitou C, Hall SD (2015) Understanding dis- ease-drug interactions in cancer patients: implications for dosing within the therapeutic window. Clin Pharmacol Ther 98(1):76–86
23. Cousin L, Berre ML, Launay-Vacher V, Izzedine H, Deray G (2003) Dosing guidelines for fluconazole in patients with renal failure. Nephrol Dial Transplant 18(11):2227–2231
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.