Tipifarnib physiologically‐based pharmacokinetic modeling to assess drug–drug interaction, organ impairment, and biopharmaceutics in healthy subjects and cancer patients

Abstract

A physiologically‐based pharmacokinetic (PBPK) model for tipifarnib, which included mechanistic absorption, was built and verified by integrating in vitro data and several clinical data in healthy subjects and cancer patients. The final PBPK model was able to recover the clinically observed single and multiple‐dose plasma concentrations of tipifarnib in healthy subjects and cancer patients under several dosing conditions, such as co‐administration with a strong CYP3A4 inhibitor and inducer, an acid‐reducing agent (proton pump inhibitor and H2 receptor antagonist), and with a high‐fat meal. In addition, the model was able to accurately predict the effect of mild or moderate hepatic impairment on tipifarnib exposure. The appropriately verified model was applied to prospectively simulate the liability of tipifarnib as a victim of CYP3A4 enzyme‐based drug–drug interactions (DDIs) with a moderate inhibitor and inducer as well as tipifarnib as a perpetrator of DDIs with sensitive substrates of CYP3A4, CYP2B6, CYP2D6, CYP2C9, and CYP2C19 in healthy subjects and cancer patients. The effect of a high‐fat meal, acid‐reducing agent, and formulation change at the therapeutic dose was simulated. Finally, the model was used to predict the effect of mild, moderate, or severe hepatic, and renal impairment on tipifarnib PK. This multipronged approach of combining the available clinical data with PBPK modeling‐guided dosing recommendations for tipifarnib under several conditions. This example showcases the totality of the data approach to gain a more thorough understanding of clinical pharmacology and biopharmaceutic properties of oncology drugs in development.

---

Study Highlights

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Tipifarnib, is a first-in-class nonpeptidomimetic quinolinone that inhibits farnesyltransferase and has the potential to improve outcomes for cancer patients with RAS-dependent malignancies. Tipifarnib is eliminated by CYP3A4- and UGT1A4-mediated metabolism, and it inhibits various CYP isoforms. It is a BCS class 2 molecule demonstrating pH-dependent changes in solubility and hence is prone to effects of food and acid-reducing agents on PK. The clinical dose of tipifarnib is 600 mg but clinical pharmacology studies in healthy subjects have been conducted at lower doses due to potential hematologic adverse events. PBPK model is applied to investigate effect of several extrinsic and intrinsic factors of tipifarnib PK at the therapeutic dose.

• WHAT QUESTION DID THIS STUDY ADDRESS?

Using the PBPK model, untested clinical scenarios were investigated for tipifarnib, that is, victim and perpetrator of DDI, food effect, the effect of ARAs, the effect of formulation changes, and PK in subjects with hepatic and renal impairment at the therapeutic dose level of 600 mg.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

Contribution estimated based on the in vitro metabolism study using HLM with cofactors for either CYPs (NADPH) or UGTs (UDPGA) was verified using the clinical DDI study with itraconazole. Higher exposure to tipifarnib in cancer patients was recovered by considering the reported down-regulation of UGT1A1 activity in patients. PK of tipifarnib in cancer patients with hepatic impairment was recovered by incorporating the HSA and Ccr data. In vitro solubility and dissolution data were parameterized and incorporated in the PBPK model that recovered the food effect and effect of ARAs.

• HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

This work demonstrates the utility of PBPK in supplementing clinical data to provide a comprehensive understanding of clinical pharmacology and biopharmaceutic properties of compounds in development and provide a totality of data approach to development and approval.

---

INTRODUCTION

Tipifarnib is a first-in-class non-peptidomimetic quinolinone that selectively binds and potently inhibits farnesyltransferase.¹ The farnesyltransferase enzyme catalyzes farnesylation, which is the predominant form of RAS prenylation. Prenylation, the attachment of a hydrophobic isoprenyl group to the C-terminal tail of mutant HRAS, is needed for its localization to the plasma membrane to activate downstream signaling. Thus, it was hypothesized that inhibiting farnesyltransferase would delocalize RAS and inhibit downstream signaling, translating to tumor regression in RAS-dependent malignancies. Tipifarnib has demonstrated dramatic tumor regression in HRAS mutant head and neck squamous cell carcinoma (HNSCC) patient-derived xenograft models.² In clinical trials, tipifarnib has demonstrated encouraging efficacy in patients with recurrent and/or metastatic HNSCC with HRAS mutations.^3,4 Additionally, tipifarnib in combination with alpelisib, a PI3Kα inhibitor, demonstrated its efficacy in PIK3CA- and HRAS-dysregulated HNSCC in preclinical models.⁵ This combination has the potential to improve outcomes for many patients with recurrent, metastatic HNSCC and is currently being evaluated in a phase I/II trial.⁶ Tipifarnib is a high permeability and low solubility molecule that is rapidly absorbed after oral administration.⁷ Administration with a high-fat meal increased tipifarnib exposure by ~30%.⁷ Hepatic metabolism is the predominant route of tipifarnib elimination, primarily by cytochrome P450 (CYP450) and uridine 5′-diphospho-glucuronosyltransferase (UGT).⁷

PBPK modeling is a mainstay in drug development and regulatory reviews.⁸ There have been significant applications of PBPK in the assessment of drug–drug interactions (DDIs), food effects, biopharmaceutics, organ impairment, and pediatrics. In the case of DDI assessment, transformational changes have happened in the last several years due to an enhanced understanding of the mechanisms of DDI and PBPK model-based approaches to evaluate them and guide dosing recommendations.⁹ The US FDA's 2020 clinical DDI guidance states that PBPK models can be used in lieu of some prospective DDI studies.¹⁰ There are many published applications of PBPK models to investigate the effect of CYP450 inhibitors and inducers that have been used in regulatory submissions.¹¹ PBPK is also routinely used for prediction and mechanistic understanding of the effect of food on drug absorption.¹² The FDA's 2022 guidance on assessing the effect of food on drugs states that PBPK modeling can be used to assess food effect in conjunction with a clinical food effect study.¹³ Similar advances have been made in the biopharmaceutics field, where physiologically-based biopharmaceutics models (PBBM), which link in vitro formulation dissolution data to PK, have been used to guide formulation development, develop bioequivalence safe space, set specifications, and investigate the effect of acid-reducing agents (ARA).¹⁴ The FDA's 2020 guidance on the application of PBPK modeling in the biopharmaceutics area further validated the utility of these models in drug product development.¹⁵ The utility of PBPK models to provide an integrated assessment of hepatic¹⁶ and renal¹⁷ impairment has also been explored and can be applied to fill in gaps of untested clinical scenarios.

In this paper, the development, verification, and application of a PBPK model for tipifarnib is described. The PBPK model was developed by integrating in vitro and clinical data. The model performance was then verified against several independent clinical data in healthy subjects and cancer patients. The verified model was applied to simulate tipifarnib DDI as a victim and perpetrator of CYP enzymes, food effect, effect of ARAs, formulation change, and PK in hepatic and renal impairment patients. The objective of this work was to demonstrate the utility of PBPK in supplementing clinical data to provide a comprehensive understanding of clinical pharmacology and biopharmaceutic properties of compounds in development.

METHODS

In vitro data

The metabolic study in a panel of recombinant CYP and UGT isoforms showed that CYP3A4 and UGT1A4 are involved in the metabolism of tipifarnib. The fraction metabolized by CYP3A4 (fm_CYP3A4) and UGT1A4 (fm_UGT1A4) was determined from experiments conducted in human liver microsomes with either NADPH or UDPGA as cofactors for CYP- and UGT-mediated metabolism, respectively. The fm_CYP3A4 was determined from the fraction of CYP-mediated intrinsic clearance that was inhibited by ketoconazole (1 μM), whereas all UGT-mediated clearance was assigned to UGT1A4.

The permeability (Papp) across C2BBel cells (a clone of Caco-2, ATCC® CRL-2102) cells cultured to confluence on Transwell® apparatus was determined. Atenolol and propranolol were used for calibration.

The solubility of tipifarnib (Tables S1, S2) was modeled using the Simcyp in vitro analysis (SIVA) Toolkit (Version 4.0). The intrinsic solubility, solubility factor, and bile micelle partitioning coefficient (unionized and ionized species) were estimated. The particle size distribution data were analyzed assuming the log-linear distribution. Dissolution data for whole and crushed tipifarnib tablets were analyzed using SIVA to estimate the disintegration rate constant (Kd1).

The potential to inhibit CYP enzymes was studied in HLM with selective probe substrates. Fraction unbound in the incubation (fu_inc) was corrected using the predicted value based on the physicochemical properties. Enzyme induction was evaluated in cultured human hepatocytes from three donors. The fold induction of mRNA was fitted with an E_max slope model.

Clinical data

Clinical trial data were available from several studies, including single and multiple ascending dose (SAD/MAD) studies in healthy subjects, cancer patients, and cancer patients with hepatic impairment. In studies R115777-BEL-3, R115777-BEL-4, and R115777-P01-101, the effect of food on tipifarnib PK was investigated. Plasma exposure of tipifarnib after administration of tablets in the fed state was ~30% higher than in the fasted state. The food effect was negligible after administration of tipifarnib solution. Study R115777-BEL-5 was an absolute bioavailability study where tipifarnib was administered as a single intravenous (IV) dose of 25 mg and a single oral dose of 50 mg in healthy subjects. Geometric mean clearance (CL) and volume of distribution at steady-state (V_SS) values were estimated to be 35 L/h and 126 L, respectively. Following oral administration, the absolute bioavailability (F) of tipifarnib was found to be 34%. In the mass balance study (study R115777-BEL-13), healthy adult male subjects received an oral dose of 50 mg [¹⁴C]-tipifarnib. The tipifarnib-glucuronide excreted in urine accounted for 10% of drug-related material, whereas negligible amounts of tipifarnib-glucuronide and only 6% of unchanged drug were excreted in feces. In comparison, ~80% of total radioactivity was excreted in feces with ~30% of the dose accounted for by three oxidative metabolites (8–12% of dose). In a study R115777-BEL-15, tipifarnib was administered once daily (QD) and twice daily (b.i.d.) to cancer patients at doses ranging from 30 to300 mg. Steady state was reached within 3 days and no accumulation was observed. Dose-proportional increases in maximal drug concentration (C_max) and area under the curve (AUC_0–12h) were observed from 100 to 600 mg in a range of studies, with little or no accumulation of tipifarnib in plasma.⁷ In the clinical DDI study (study KO-TIP-014), geometric mean AUC_0–inf and C_max ratios for tipifarnib in the absence and presence of itraconazole were 2.14 and 1.82, respectively. Geometric mean AUC_0–inf and C_max ratios in the absence and presence of rifampicin were 0.15 and 0.22, respectively. In the study, KO-TIP-008 in healthy subjects, relative bioavailability between two tablet formulations ranged between 99.23% and 113.39% and tipifarnib AUC_0–inf decreased by 20% after co-administration of rabeprazole. Plasma exposure to tipifarnib was reduced when co-administered with omeprazole or ranitidine in healthy subjects (study R115777-BEL-16). In cancer patients with mild and moderate hepatic impairment, elevated plasma exposure to tipifarnib was observed (study R115777-NED-3). All clinical studies were conducted in full accordance with the principles stated in ICH, and GCP guidelines, as well as with IRB approval and informed consent from the study participants. The trial number and titles are listed in the Supplementary Materials S1. The clinical data used in the model development and verification are summarized in Figure 1.

[IMAGE OMITTED. SEE PDF]

PBPK modeling and simulations

Version 22 of the Simcyp Population-Based Simulator () was used for all PBPK modeling and simulation. The modeling strategy is shown in Figure 1.

The absorption of tipifarnib was modeled using the advanced dissolution absorption and metabolism (ADAM) model. Intestinal permeability (P_eff,man) was predicted based on the in vitro permeability across Caco-2 cells. Distribution was described using the minimal PBPK model with parameters obtained by fitting with the PK profile after intravenous administration (study R115777-BEL-5). The unbound fraction in the gut (fu_Gut) was optimized to recover the PK profile after oral administration of 50 mg tipifarnib (study R115777-BEL-5).

HV population file, North European Caucasian population file (NEurCaucasiaon), cancer population file, and cirrhosis population (Child-Pugh (CP)-A, CP-B and CP-C) files provided in the Simcyp simulator were used to generate virtual populations. 25% down-regulation of UGT1A4 expression in the liver, intestine, and kidney was assumed in cancer patients to recover the lower clearance of tipifarnib in these patients, which was qualitatively supported by literature data.¹⁸ To simulate the effect of hepatic impairment on the tipifarnib PK in cancer patients (study R115777-NED-3), Sim-NEurCaucasian population file with a 25% reduction of UGT1A4 abundances and the default CP-A and CP-B population file were used. The hepatic abundance of UGT1A4 in the CP-A population file is lower than that in healthy subjects by ~40%. Individual serum HSA and serum creatinine levels were also used to generate virtual populations.

The effect of ARAs on the plasma exposure of tipifarnib after administration in the fed state was simulated by changing the gastric pH profile to the reported pH profiles after administration of ranitidine¹⁹ and omeprazole.²⁰ Without ARAs gastric pH in the fed state was 5 and decreased to 1.5 after 2 h. When ranitidine was administered, gastric pH was ~7. In subjects treated with omeprazole, gastric pH in the fed state was 6 and decreased to 4 at 2 h after food intake.

The ability of the model to recover tipifarnib PK in clinical studies conducted on healthy subjects and cancer patients was evaluated. The criteria for an accurate recovery were defined as when simulated summary PK parameters (geomean AUC and C_max, median t_max) were within 0.8 to 1.25-fold of corresponding observed values. For each simulated clinical trial, 10 trials with the same number of individuals as the observed clinical trial were simulated. The model was applied to simulate DDI liability as a victim and perpetrator of several CYP enzymes, food effect, effect of ARAs, formulation change effects, and organ impairment effects.

RESULTS

Model development and verification

The input parameters are listed in Table 1. Tipifarnib is a substrate of UGT 1A4 and cytochrome P450 (CYP) 3A4 enzymes in recombinant CYP and HLM systems.

TABLE 1 Input parameters of tipifarnib.

Parameter	Value	Reference
Physicochemical and binding parameters
MW (g/mol)	489.4	Tipifarnib Investigator's Brochure
Log P	4.41
Compound type	Diprotic base
pKa 1	6.1
pKa 2	3.3
B:P	0.74	Kura Oncology data on file
fu (Healthy)	0.0078	Kura Oncology data on file
fu (Cancer)	0.0062	Clinical study R115777-BEL-15 (Kura Oncology data on file)
Main binding protein	HSA	Kura Oncology data on file
Cancer patient HSA [P]ref (g/L)	37.5	Optimized
Absorption model – ADAM model
fugut	0.55	Optimized
Caco-2 Papp (×10−6 cm/s)	13.3	Kura Oncology data on file
Calibrator Papp (×10−6 cm/s) – atenolol	0.286
Calibrator Papp (×10−6 cm/s) – propranolol	25.2
Peff,man (pred) (×10−4 cm/s)	2.75	Simcyp predicted
Formulation type	IR tablet (DLM model)
Intrinsic solubility (mg/mL)	0.000812	Estimated from SIVA
Solubility factor	4181
LogKm:w – neutral	5.37
LogKm:w – ion	5.24
Mean radius (μm)	6.895	Estimated from Simcyp
CV (%)	111.5
Minimum radius (μm)	0.310
Maximum radius (μm)	43.6
Disintegration rate constant (Kd1, h−1)	0.210 (KF001)	Estimated from SIVA
0.248 (F032)
0.229 (other tablets)
Number of bins (simulation)	30	–
Distribution model – Minimal PBPK model
VSS (L/kg)	1.54	Clinical study R115777-BEL-5 (Kura Oncology data on file)
kin (1/h)	0.152
kout (1/h)	0.226
VSAC (L/kg)	0.54
Elimination parameters
CYP3A4 CLint (μL/min/pmol)	5.25	Iterative retrograde model. CLIV (33.9 L/h) obtained from Clinical study R115777-BEL-5; fmCYP3A4 (36.4%) and fmUGT1A4 (60.5%) obtained from in vitro reaction phenotyping studies (Kura Oncology data on file)
UGT1A4 CLint (μL/min/pmol)	21.1
Additional HLM CLint (μL/min/mg)	56.3
CLR (L/h)	0	Clinical study R115777-BEL-13 (Kura Oncology data on file)
Interaction parameters
CYP2B6 Ki (μM)	2.90	Kura Oncology data on file
CYP2C8 Ki (μM)	1.15
CYP2C9 Ki (μM)	0.35
CYP2C19 Ki (μM)	0.75
CYP3A4 Ki (μM)	1.05
fumic (for Ki inputs)	0.726	Simcyp predicted (0.1 mg/mL HLM)
CYP2B6 Indmax	8.75	Kura Oncology data on file
CYP2B6 IndC50 (μM)	14.9
CYP2B6 γ	1.86
CYP3A4 Indmax	9.73	Kura Oncology data on file
CYP3A4 IndC50 (μM)	15.3
fuhep (for induction inputs)	0.252	Predicted

The hepatic clearance of tipifarnib was apportioned with fraction metabolized (fm) values for UGT1A4 and CYP3A4 of ~ 61 and 36%, respectively. These fm's were assigned based on the relative tipifarnib intrinsic clearance in human liver microsomes in the presence of cofactors for either CYPs (NADPH) or UGTs (UDPGA) and the impact of the selective inhibitor ketoconazole on CYP-mediated metabolism.

The simulated fraction absorbed (fa) of 0.92 ± 0.09 (n = 12) after an oral dose of 50 mg was in line with mass balance data, which suggested near-complete oral absorption of tipifarnib (study R115777-BEL-13). Using the simulated AUC_0–inf after an oral dose of 50 mg tipifarnib and after an intravenous dose of 25 mg tipifarnib, the estimated absolute bioavailability was 36.4 ± 9.86%. This is in close agreement with the observed absolute bioavailability of 34.0 ± 10.0% (study R115777-BEL-5). The model was able to predict independent clinical PK following the administration of a single dose of tipifarnib in healthy subjects and multiple doses in cancer patients at different doses (Figure 2).

[IMAGE OMITTED. SEE PDF]

The assumption of fm_CYP3A4 (0.36) was verified using the clinical DDI study data (Clinical DDI Study KO-TIP-014). The simulated geometric mean ratio (GMR) of AUC and C_max in the presence versus absence of itraconazole were accurately simulated by the model with ratio of simulated to observed GMR of AUC_0–inf and C_max being 0.97 and 0.92, respectively (Table 2 and Figure 3a). The effect of rifampicin on tipifarnib PK was slightly underestimated, as indicated by the ratio of simulated to observed GMR of AUC_0–inf and C_max of 1.42 and 1.32, respectively (Table 2 and Figure 3a).

TABLE 2 Summary of DDI model development, verification and application using various CYP inhibitors, inducers and substrates in healthy subjects and cancer patients.

Compound	Classification	Population	Single dose (600 mg)
AUC0–inf GMR	Cmax GMR
Model development and verification
Itraconazole	Strong CYP3A4 inhibitor	Healthy subjects	2.14 (observed)	1.82 (observed)
2.07 (simulated)	1.68 (simulated)
Rifampicin	Strong CYP3A4 inducer	Healthy subjects	0.15 (observed)	0.22 (observed)
0.21 (simulated)	0.29 (simulated)
Model Application – Simulation of tipifarnib as a victim of DDI
Fluconazole	Moderate CYP3A4 inhibitor	Healthy subjects	1.41	1.28
Cancer patients	1.50	1.33
Efavirenz	Moderate CYP3A4 inducer	Healthy subjects	0.51	0.61
Cancer patients	0.49	0.60
Model application – Simulation of tipifarnib as a perpetrator of DDI
Midazolam	CYP3A4 substrate	Healthy subjects	Induction + Inhibition	1.15	1.14
Induction	0.79	0.79
Inhibition	1.32	1.31
Cancer patients	Induction + Inhibition	1.07	1.05
Induction	0.74	0.75
Inhibition	1.29	1.27
Bupropion	CYP2B6 substrate	Healthy subjects	1.01	1.01
Cancer patients	1.01	1.01
Repaglinide	CYP2B8 substrate	Healthy subjects	1.08	1.08
Cancer patients	1.06	1.08
Warfarin	CYP2C9 substrate	Healthy subjects	1.03	1.01
Cancer patients	1.03	1.01
Omeprazole	CYP2C19 substrate	Healthy subjects	1.06	1.06
Cancer patients	1.05	1.05

[IMAGE OMITTED. SEE PDF]

The model was able to predict the positive food effect (R115777-BEL-4) at both 50 and 100 mg (Table 3). The ratio of simulated to observed AUC_0–inf and C_max were in the range of 0.9–1.08. By incorporating the gastric pH changes mimicking the reported pH-time profiles^19,20 after administration of ranitidine and omeprazole in the population model, reduced plasma exposure after administration of tipifarnib in the presence of ARAs was recovered well (Table 3). The ratios of simulated to observed AUC_0–inf and C_max were 1.10 and 0.85 for ranitidine, and were 0.98 and 0.93 for omeprazole, respectively.

TABLE 3 Model development/verification of the effect of food, gastric pH, and formulation on tipifarnib PK in healthy subjects and its application in the prediction of effect at higher dose.

Scenarios	Tipifarnib dose (mg)	Treatment	AUC0–inf (ng/mL*h)	Cmax (ng/mL)
Model development and verification
Food effect	50	Fasted	442 (observed) 497 (simulated)	110 (observed) 93.7 (simulated)
Fed (high-fat meal)	523 (observed) 667 (simulated)	136 (observed) 111 (simulated)
GMR (Fed/Fasted)	1.34 (observed) 1.39 (simulated)	1.39 (observed) 1.24 (simulated)
100	Fasted	1050 (observed) 1128 (simulated)	260 (observed) 218 (simulated)
Fed (high-fat meal)	1241 (observed) 1486 (simulated)	260 (simulated) 251 (observed)
GMR (Fed/Fasted)	1.26 (observed) 1.36 (simulated)	1.14 (observed) 1.21 (simulated)
Gastric pH change	50	Control	417 (observed) 660 (simulated)	104 (observed) 110 (simulated)
Ranitidine (300 mg)	322 (observed) 516 (simulated)	76.1 (observed) 77.7 (simulated)
GMR (ranitidine/control)	0.77 (observed) 0.78 (simulated)	0.73 (observed) 0.71 (simulated)
Omeprazole (20 mg)	372 (observed) 572 (simulated)	90.5 (observed) 90.5 (simulated)
GMR (omeprazole/control)	0.89 (observed) 0.87 (simulated)	0.88 (observed) 0.82 (simulated)
Model application – Simulation of higher dose
Food effect	600	Fasted	3879	879
Fed (high-fat meal)	5754	1078
GMR (Fed/Fasted)	1.48	1.23
Gastric pH change	600	Control	6419	1223
Ranitidine	2675	441
GMR (ranitidine/control)	0.42	0.36
Omeprazole	3705	644
GMR (omeprazole/control)	0.58	0.53
Whole versus crushed tablets	Fasted	600	Whole	3877	880
Crushed	4363	1003
GMR (Crushed/Whole)	1.13	1.14
Fed	Whole	6433	1226
Crushed	6495	1239
GMR (Crushed/Whole)	1.01	1.01

Plasma exposure of tipifarnib in cancer patients was higher than in healthy subjects. In healthy subjects, dose-normalized AUC_0–inf following a single oral dose of 50 mg was 9.5 × 10⁻⁶ ng/mL*h/mg, whereas the dose-normalized AUC_tau following 200 mg b.i.d. dosing in cancer patients was 17.9 × 10⁻⁶ ng/mL*h/mg. Plasma PK profiles in patients with age and gender matching to the clinical study simulated using the default cancer population model underpredicted the observed data. By reducing the UGT1A4 abundance, the observed data were recovered (Figure 2). To simulate the effect of hepatic impairment on the plasma PK parameters in cancer patients, serum HSA and serum creatinine data were used to generate virtual populations. As shown in Table 4 and Figure 3c, the impact of mild and moderate hepatic impairment on the plasma exposure of tipifarnib following multiple dosing for 5 days was recovered reasonably. Simulated/observed ratios were generally within 1.25-fold of the observed data.

TABLE 4 Comparison of the model-predicted and observed effect of hepatic impairment (HI) on the plasma exposure of tipifarnib and accumulation during multiple dosing in cancer patients with or without moderate hepatic impairment. Application of the model to predict the effect of mild, moderate, and severe HI and renal impairment (RI) on tipifarnib PK at a dose of 600 mg b.i.d.

	Tipifarnib dose (mg)	Parameter ratio	Day 1	Day 5
Model development and verification
HI	Mild HI vs. Normal	50 b.i.d.	C max	1.46 (observed) 1.81 (simulated)	1.90 (observed) 1.83 (simulated)
AUC0–12h	2.01 (observed) 1.74 (simulated)	2.20 (observed) 1.76 (simulated)
Moderate HI vs. Normal	200 b.i.d.	C max	1.09 (observed) 1.23 (simulated)	1.17 (observed) 1.44 (simulated)
AUC0–12h	1.13 (observed) 1.46 (simulated)	1.56 (observed) 1.71 (simulated)
Accumulation (Day 5/Day 1)
Mild HI	50 b.i.d.	C max	1.19 (observed) 1.25 (simulated)
AUC0–12h	1.33 (observed) 1.34 (simulated)
Moderate HI	200 b.i.d.	C max	1.34 (observed) 1.37 (simulated)
AUC0–12h	1.53 (observed) 1.49 (simulated)
Normal	50 b.i.d.	C max	0.96 (observed) 1.25 (simulated)
AUC0–12h	1.22 (observed) 1.33 (simulated)
Normal	200 b.i.d.	C max	1.25 (observed) 1.23 (simulated)
AUC0–12h	1.19 (observed) 1.15 (simulated)
Model application – Simulation of higher dose in non-cancer subjects
HI	Mild HI vs. Normal	600 b.i.d.	C max	1.30	1.43
AUC0–12h	1.48	1.64
Moderate HI vs. Normal	C max	1.73	2.09
AUC0–12h	2.19	2.67
Severe HI vs. Normal	C max	2.28	3.35
AUC0–12h	3.37	5.14
RI	Mild RI vs. Normal	600 b.i.d.	C max	1.02	1.02
AUC0–12h	1.01	1.01
Moderate RI vs. Normal	C max	1.23	1.26
AUC0–12h	1.28	1.32
Severe RI vs. Normal	C max	1.22	1.23
AUC0–12h	1.28	1.29

Model applications

The model was applied to predict the effect of fluconazole and efavirenz, a moderate CYP3A4 inhibitor and inducer, respectively (Table 2, Figure 3a,b). A weak DDI was predicted in healthy subjects for both agents. In cancer patients, as UGT1A4 was assumed to be downregulated, the simulated fm_CYP3A4 in cancer patients was higher than that in healthy subjects. As a result, the predicted DDI with fluconazole and itraconazole was slightly stronger than in healthy subjects. The effect of tipifarnib on the plasma exposure of CYP2B6, CYP2C8, CYP2C9, and CYP2C19 substrates using the in vitro _Ki values of tipifarnib were not significant, with AUC and C_max ratios <1.25. The simulated AUC_0–inf and C_max ratios for midazolam were also <1.25. As tipifarnib is an inhibitor and inducer of CYP3A4, the effect of either induction or inhibition was simulated as the worst case. A weak inhibition (1.25 < AUC_0–inf GMR <2) or weak induction (0.5 < AUC_0–inf GMR <0.8) was predicted.

The effect of food intake and gastric pH change by ARA administration were predicted for higher dose levels (600 mg) than those studied in the clinical studies (study R115777-BEL-16). A positive food effect with an AUC_0–inf increase of 1.48-fold was predicted (Table 3). Simulations by mimicking the gastric pH profiles after administration of ranitidine and omeprazole predicted reduced plasma exposure to tipifarnib. The effect of ARAs was greater on the plasma exposure of tipifarnib following a 600 mg dose compared with a 50 mg dose (Table 3).

The estimated disintegration rate constant (Kd1) for crushed tablets based on the dissolution data was 1.39 h⁻¹, which was much faster than for whole tablets (0.210–0.257 h⁻¹). In the fasted state, simulated plasma exposure was slightly higher after administration of crushed tablets (Table 3). However, in the fed state, the simulated exposures after administration of crushed and whole tablets were comparable. The sensitivity analysis around the tablet disintegration rate constant on the absorption of tipifarnib showed that oral absorption of tipifarnib is not affected by the disintegration rate of tablets when Kd1 is higher than 0.2 h⁻¹ (Figure S5).

The model was applied to predict the impact of hepatic and renal impairment on the plasma exposure of tipifarnib after oral administration of 600 mg dose (Table 4). The effect relative to the healthy subjects was predicted. Average ages of the generated populations with normal hepatic function, mild, moderate, and severe hepatic impairment were 59.4 years. Simulated fu increased with the severity of the impairment (0.0082 (normal), 0.010 (mild HI), 0.012 (moderate HI), and 0.016 (severe HI)). The simulated AUC_0–12h on Day 5 in subjects with mild, moderate, and severe hepatic impairment was 1.64, 2.67, and 5.14-fold higher than that in subjects with normal hepatic function, respectively (Figure 3c). The average ages of the generated populations with normal renal function, mild, moderate, and severe renal impairment were 71.8–72.8 years. Simulated fu slightly increased with the severity of the impairment from 0.0082 (normal) to 0.0092 (severe RI). The simulated AUC_0–12h on Day 5 in subjects with mild, moderate, and severe renal impairment was 1.01. 1.32, and 1.29-fold higher than in subjects with normal renal function, respectively.

DISCUSSION

A PBPK model was developed for tipifarnib based on available physicochemical, in vitro (including CYP and UGT elimination and interaction parameters), and clinical data. The model was refined such as reduction in UGT-mediated metabolism in cancer patients and verified using data from several clinical studies including single and multiple-dose escalation, food effect, DDI (CYP3A4 and ARA), and mild/moderate hepatic impairment. Specifically, the DDI model was verified using the clinical victim DDI data with itraconazole and rifampicin to refine estimates of the role of CYP3A4 in tipifarnib metabolism. In addition to being a potent CYP inducer, rifampicin can also induce UGT.²¹ However, modeling of the available clinical DDI data with itraconazole provides confidence in the estimation of fm_CYP3A4 for tipifarnib based on the metabolic stability in the presence and absence of CYP and UGT cofactors. To accurately model the absorption of tipifarnib, an ADAM model was built by incorporating physicochemical and biopharmaceutic properties of tipifarnib, as described in Table 1, along with gastrointestinal physiology such as changes in gastric pH and bile acids concentration after food intake and ARA administration. The model was verified using clinical food effect data with high-fat meal at 50 mg and 100 mg, and ARA (H2RA – ranitidine and PPI – omeprazole) co-administration at 50 mg. The effect of organ impairment was modeled using clinical data at 50 mg and 100 mg in mild and moderate hepatic impairment. The AUC and C_max ratios for mild HI/normal were observed to be higher than that of moderate HI/normal, possibly due to lower plasma unbound fraction (fu) in moderate HI in the clinical study conducted in cancer patients (study R115777-NED-3). To recover the observed fu in each population, serum albumin levels were used to generate the virtual population. In addition, the PBPK model accounted for the effect of tipifarnib PK in cancer patients. The final model adequately described the observed tipifarnib PK under several clinical conditions. The observed tipifarnib plasma concentration profiles were generally well captured by the simulations in healthy subjects and cancer patients across various clinical studies at 50–300 mg doses under q.d. and b.i.d. regimens, as shown in Figure 2. The AUC and C_max values were recovered well, with simulated vs. observed ratio <1.5-fold for all the scenarios modeled which was well within the two-fold of observed data that is generally considered to be acceptable.²² The model was subsequently applied to simulate DDI liability as a victim and perpetrator of several CYP enzymes, food effect, effect of ARAs, formulation change, and PK in cancer patients with hepatic and renal impairment.

Clinical DDI data in healthy subjects provided strong quantitative confirmation of tipifarnib as a substrate of CYP3A4 and guided the development of the PBPK model. Typically, in oncology drug development, it is not feasible to test all clinically relevant scenarios in healthy subjects due to ethical and safety considerations. Therefore, limited clinical experimentation in healthy subjects followed by a PBPK-based approach to cover other dosing scenarios of interest is an ideal strategy. It is also known that tumor-related inflammation can cause down-regulation of hepatic CYP3A4 in cancer patients.^23,24 However, CYP3A4 abundance (pmol/mg microsomal protein) in the cancer population model assumes the same values as in the non-cancer north European Caucasian population model, based on the meta-analysis data showing inconclusive difference.²⁴ As cancer patients are generally older and the content of hepatic microsomal protein level decreases with age, plasma exposure of various oncology drugs that are eliminated by CYP3A4-metablism have been recovered by age-matched simulations. The default Sim-cancer population does not incorporate any change to the expression of UGT enzymes. Evidence of a significant down-regulation of hepatic UGT1A4 levels in patients with hepatocellular carcinoma has been published.¹⁸ Although there is some indication that UGT1A1 is differentially down and up-regulated in different cancer types,²⁵ the reduction in tipifarnib clearance observed in cancer patients in study R115777-BEL-15 was accounted for in the PBPK model by a 25% reduction in UGT1A4 expression levels (liver, intestine, and kidney) in the Sim-cancer population. Therefore, the PBPK model was applied to predict the effects of moderate CYP3A4 inhibitor and inducer on tipifarnib PK in healthy subjects and cancer patients. The simulations demonstrated weak interaction with moderate CYP3A4 modulators in both populations (Figure 3a,b). A similar PBPK approach was adopted to inform tipifarnib's potential as a perpetrator of CYP-based DDI using sensitive substrates, which also demonstrated weak interaction. This multipronged approach of utilizing available clinical data and supplementing with PBPK informed DDI potential, enabled decision-making on co-administration, and guided dosing recommendations for tipifarnib.

The effect of food and ARAs on tipifarnib PK was investigated clinically at lower doses, due to potential safety concerns, such as hematologic AEs, in healthy subjects.⁴ The PBPK model was subsequently used to investigate the effect of food and ARAs at the clinically relevant dose. These simulations predicted ~20–50% increase in exposure with high-fat meal, which was in line with the positive food effect observed at the lower doses. However, the effect of ARAs (ranitidine and omeprazole) was predicted to be much higher at 600 mg than that observed at 50 mg, which is expected given the significantly higher solubility limitation at the higher dose (dose/solubility ratio at pH 6.0 of 790 and 66 at 600 mg and 50 mg, respectively). The PBPK model-based prediction highlighted the risk of potentially underdosing the patients when co-administered with ARAs and hence the need to stagger ARA administration with respect to the dosing time of tipifarnib. The model was also used to project the effect of crushing the tablet, which demonstrated minimal impact on tipifarnib PK in both fasted and fed states. If validated clinically, this finding would provide an alternative dosing option for head and neck cancer patients, as dysphagia is a common symptom in this cancer type and makes swallowing whole tablets difficult.²⁶

Generally, patients with moderate or severe hepatic and renal impairment (HI and RI) are excluded from trials in oncology patients, and dosing recommendations are derived from a combination of dedicated studies in otherwise healthy subjects and modeling. PBPK modeling can potentially play an important role with further understanding of pathophysiological changes with HI or RI, and its impact on exposure. The utility of PBPK modeling in predicting the effect of organ impairment on exposure was systematically evaluated for 29 compounds by the IQ Consortium.¹⁶ Typically, the predictions were within two-fold of the observed ratios. However, in moderate and severe insufficiencies, the exposures were generally overpredicted, which could be due to a lack of clear understanding of all the pathophysiological changes causing hepatic or renal insufficiency. In the case of tipifarnib, mild and moderate HI and RI clinical data were available at low doses (50 and 200 mg b.i.d.) only, due to unacceptable safety risks for otherwise healthy subjects in the case of a significant increase in tipifarnib exposure. PBPK modeling was used to supplement the available clinical data by predicting exposures in all categories of HI and RI at the efficacious dose of 600 mg b.i.d. Modeling predicted a significant increase in exposure at 600 mg b.i.d. dose in subjects with severe HI, which could require dose adjustment in cancer patients with severe HI (Figure 3c). However, RI was not predicted to significantly impact tipifarnib exposure in any of the categories.

In summary, the tipifarnib PBPK model was built and verified using several clinical datasets. The verified model was then applied to inform DDI liability of tipifarnib, food effect, effect of ARAs, formulation change effects, and PK in HI and RI patients at the efficacious dose. This example showcases the totality of data approach in clinical pharmacology and biopharmaceutic risk assessment by combining PBPK modeling and clinical data.

AUTHOR CONTRIBUTIONS

All authors wrote the manuscript and designed the research. N.O. and H.B. conducted the research. All authors analyzed the data.

FUNDING INFORMATION

No funding was received for this work.

CONFLICT OF INTEREST STATEMENT

Amitava Mitra is an employee of Kura Oncology Inc. All other authors deny any competing interests for this work.

References

End DW, Smets G, Todd AV, et al. Characterization of the antitumor effects of the selective farnesyl protein transferase inhibitor R115777 in vivo and in vitro. Cancer Res. 2001;61(1):131‐137.

Gilardi M, Wang Z, Proietto M, et al. Tipifarnib as a precision therapy for HRAS‐mutant head and neck squamous cell carcinomas. Mol Cancer Ther. 2020;19(9):1784‐1796.

Ho AL, Brana I, Haddad R, et al. Tipifarnib in head and neck squamous cell carcinoma with HRAS mutations. J Clin Oncol. 2021;39:1856‐1864.

Ho AL, Adkins DR, Hannah GJ, et al. LBA47 ‐ a phase II study evaluating tipifarnib in mHRAS, recurrent or metastatic (R/M) head and neck squamous cell carcinoma (HNSCC) (AIM‐HN study). Eur Soc Med Oncol Annual Conf. 2023;34:S1286‐S1287.

Smith AE, Chan S, Wang Z, et al. Tipifarnib potentiates the antitumor effects of PI3Kα inhibition in PIK3CA‐ and HRAS‐dysregulated HNSCC via convergent inhibition of mTOR activity. Cancer Res. 2023;83(19):3252‐3263.

Combination trial of tipifarnib and alpelisib in adult recurrent/metastatic head and neck squamous cell carcinoma (R/M HNSCC). https://clinicaltrials.gov/study/NCT04997902

Tipifarnib investigator's brochure. Edition 17. 2022.

Jean D, Naik K, Milligan L, et al. Development of best practices in physiologically based pharmacokinetic modeling to support clinical pharmacology regulatory decision‐making—a workshop summary. CPT Pharmacometrics Syst Pharmacol. 2021;10(11):1271‐1275.

Venkatakrishnan K, Rostami‐Hodjegan A. Come dance with me: transformative changes in the science and practice of drug‐drug interactions. Clin Pharmacol Ther. 2019;105(6):1272‐1278.

US FDA Guidance for Industry. Clinical drug interaction studies – Cytochrome P450 enzyme and transporter mediated drug interactions. https://www.fda.gov/media/134581/download

Zhang X, Yang Y, Grimstein M, et al. Application of PBPK modeling and simulation for regulatory decision making and its impact on US prescribing information: an update on the 2018‐2019 submissions to the US FDA's Office of Clinical Pharmacology. J Clin Pharmacol. 2020;60(Suppl 1):S160‐S178.

Riedmaier AE, DeMent K, Huckle J, et al. Use of physiologically based pharmacokinetic (PBPK) modeling for predicting drug‐food interactions: an industry perspective. AAPS J. 2020;22(6):123.

US FDA Guidance for Industry. Assessing the effects of food on drugs in INDs and NDAs—Clinical pharmacology considerations. https://www.fda.gov/media/121313/download

Mitra A, Suarez‐Sharp S, Pepin XJH, et al. Applications of physiologically based biopharmaceutics modeling (PBBM) to support drug product quality: a workshop summary report. J Pharm Sci. 2021;110(2):594‐609.

US FDA Guidance for Industry. The use of physiologically based pharmacokinetic analyses—Biopharmaceutics applications for oral drug product development, manufacturing changes, and controls. https://www.fda.gov/media/142500/download

Heimbach T, Chen Y, Chen J, et al. Physiologically‐based pharmacokinetic modeling in renal and hepatic impairment populations: a pharmaceutical industry perspective. Clin Pharmacol Ther. 2021;110(2):297‐310.

Yeo KR, Berglund EG. An integrated approach for assessing the impact of renal impairment on pharmacokinetics of drugs in development: pivotal role of PBPK modelling. Clin Pharmacol Ther. 2021;110(5):1168‐1171.

Strassburg CP, Manns MP, Tukey RH. Differential down‐regulation of the UDP‐glucuronosyltransferase 1A locus is an early event in human liver and biliary cancer. Cancer Res. 1997;57(14):2979‐2985.

Smout AJ, Jonkman JH, Peeters PA, et al. Effect of an evening dose of regular and effervescent formulations of ranitidine or cimetidine on intragastric pH in healthy volunteers. Aliment Pharmacol Ther. 1995;9:51‐56.

Tolman KG, Sanders SW, Buchi KN, Karol MD, Jennings DE, Ringham GL. The effects of oral doses of lansoprazole and omeprazole on gastric pH. J Clin Gastroenterol. 1997;24:65‐70.

Chen J, Raymond K. Roles of rifampicin in drug‐drug interactions: underlying molecular mechanisms involving the nuclear pregnane X receptor. Ann Clin Microbiol Antimicrob. 2006;5:3.

Wagner C, Pan Y, Hsu V, et al. Predicting the effect of cytochrome p450 inhibitors on substrate drugs: analysis of physiologically based pharmacokinetic modeling submissions to the US food and drug administration. Clin Pharmacokinet. 2015;54:117‐127.

Coutant DE, Kulanthaivel P, Turner PK, et al. Understanding disease‐drug interactions in cancer patients: implications for dosing within the therapeutic window. Clin Pharmacol Ther. 2015;98(1):76‐86.

Cheeti S, Budha NR, Rajan S, Dresser MJ, Jin JY. A physiologically based pharmacokinetic (PBPK) approach to evaluate pharmacokinetics in patients with cancer. Biopharm Drug Dispos. 2013;34(3):141‐154.

Hu DG, Marri S, Mckenzie PI, Hullin JA, et al. The expression profiles and deregulation of UDP‐glycosyltransferase (UGT) genes in human cancers and their association with clinical outcomes. Cancer. 2021;13:4491.

Manikantan K, Khode S, Sayed SI, et al. Dysphagia in head and neck cancer. Cancer Treat Rev. 2009;35(8):724‐732.