Physiologically‐Based Pharmacokinetic Modeling of the PARP Inhibitor Niraparib

ABSTRACT

A physiologically‐based pharmacokinetic (PBPK) model of niraparib and its primary metabolite using a relevant virtual cancer population is reported here. A series of in vitro experiments using liver S9, microsomes, and hepatocytes with various inhibitors and recombinant supersomes demonstrated that niraparib is specifically metabolized by carboxylesterase 1 via amide hydrolysis to an acid metabolite (M1). Available virtual cancer populations, along with reference populations, were applied to modeling simulations using fixed trial designs with demographic and clinical chemistry parameters from patients receiving niraparib in clinical studies. Simulations of niraparib and its metabolite M1 were verified across numerous available clinical studies and repeat dose ranges in cancer patients within 2‐fold. The PBPK model was used to simulate exposures in moderately hepatic impaired, healthy Chinese and Japanese virtual populations as a surrogate of cancer comorbidity. The PBPK model confirmed minimal DDI liability with niraparib as a precipitant for most in vitro tested drug metabolizing enzymes and transporters. In vitro, niraparib lacks any CYP inhibition, induces CYP1A2 but not CYP3A4, and is not a CYP substrate, unlike some other PARPi's, which inhibit and induce numerous enzymes/transporters and are objects of CYP metabolism. At clinically relevant doses of niraparib ≥ 200 mg, a weak induction risk is predicted with sensitive CYP1A2 substrates, such as caffeine, and both niraparib and olaparib clinically increase serum creatinine in cancer patients, with up to a moderate inhibition risk predicted with MATE‐1/‐2K substrates, such as metformin, using a PBPK model of niraparib in the absence of a dedicated DDI study.

Study Highlights

• What is the current knowledge on the topic?

  • ○
    Clinical studies have shown that niraparib exhibits linear pharmacokinetics, a long half‐life, high volume of distribution, metabolism to an inactive primary metabolite (M1), and limited DDI liability.
• What question did this study address?

  • ○
    What is the specific metabolic route of niraparib? Can a PBPK model match the clinical profiles of niraparib and its metabolite M1 using a virtual cancer patient population?
• What does this study add to our knowledge?

  • ○
    In vitro studies demonstrate that niraparib is metabolized by CES1 to its primary metabolite M1 and subsequent acyl‐glucuronide and is not metabolized by CES2 or CYPs. PBPK modeling successfully captured both niraparib and M1 profiles across single and repeat‐dose ranges using CES1‐mediated clearance and a virtual cancer population and helped analyze precipitant DDI liabilities prospectively.
• How might this change drug discovery, development, and/or therapeutics?

  • ○
    Given the challenges of conducting clinical studies in cancer patients, this work provides the potential application of PBPK modeling to understanding mechanistic disposition, biopharmaceutic formulations, other populations, and interpatient variability of a CES1 substrate niraparib and its clinical precipitant liabilities.

1. Introduction

Niraparib (Zejula) is a potent and selective oral inhibitor of poly ADP‐ribose polymerases (PARPi), PARP‐1 and PARP‐2 [1]. PARPs are a family of enzymes responsible for DNA repair. Niraparib inhibits both PARP‐1 and PARP‐2 and demonstrated clinical activity in patients with ovarian cancer [2]. Niraparib was initially approved by the US Food and Drug Administration (FDA) in 2017 as maintenance treatment in adult patients with recurrent epithelial ovarian, fallopian tube, or primary peritoneal cancer after a complete or partial response to platinum‐based chemotherapy (PBCT) [3] and by the European Medicines Agency (EMA) in 2017 as maintenance treatment of adult patients with platinum‐sensitive relapsed high‐grade serious epithelial OC after a complete or partial response to PBCT [4]. In 2022, FDA approval of niraparib as second‐line maintenance therapy was amended to include only patients with deleterious or suspected deleterious germline BRCA (gBRCA) mutations [5]. In the recurrent setting, the recommended starting dose of niraparib is 200 mg taken once daily. However, for those patients who weigh ≥ 77 kg and have baseline platelet count ≥ 150,000/μL, the recommended starting dose of niraparib is 300 mg taken once daily [6].

Niraparib is available at 100, 200, and 300 mg dose strengths [7]. It is rapidly absorbed, reaching maximum plasma concentration within 5 h, with linear pharmacokinetics (30–400 mg) [8]. After that, the plasma concentration decreases in a biphasic pattern. Niraparib has a bioavailability of 73% and binds to 83% of plasma proteins. The mean volume of distribution is 1220 L. Niraparib is mainly metabolized by amide hydrolysis catalyzed by carboxylesterases (CES), forming an inactive acid metabolite (M1) that subsequently undergoes glucuronidation to acyl isomers (M10, M12). Niraparib shows a long terminal half‐life (approximately 2 days). The principal elimination routes of niraparib and its metabolites are the hepatic/biliary and renal pathways, with 47.5% of the drug excreted in the urine and 38.8% in the feces over 21 days [8, 9, 10].

In vitro, niraparib has been shown to inhibit CYP2C19 and CYP2D6 enzymes, with IC50 values of 91 and 73 μM, respectively, but not CYP1A2, CYP2B6, CYP2C8, CYP2C9, or CYP3A4 (IC50 > 100 μM), and to induce CYP1A2 (EC50 6.8 μM, Emax 6.3). It is also an in vitro inhibitor of transporters P‐gp, BCRP, OCT‐1, MATE‐1, and MATE‐2K, with IC50 values of 131, 5.8, 34.1, 0.18, and 0.14 μM, respectively. In vitro, M1 is not an inhibitor or inducer of any CYPs or transporters, and both niraparib and M1 are not inhibitors of any UGTs [internal data]. Conversely, the PARPi olaparib is predominantly metabolized by CYP3A4 and has potential to cause CYP3A4 reversible and time‐dependent inhibition (TDI) and induction, as shown by in vitro and clinical DDI studies [11]. Olaparib also inhibits CYP2C9 and CYP2C19, and it induces CYP2B6 and, to a lesser extent, CYP1A2. Additionally, olaparib is a substrate of P‐gp and inhibits multiple transporters such as P‐gp, OCT1, OAT1B1, and OAT3, with the most notable being inhibition of OCT2, MATE‐1, and MATE2‐K at clinically relevant concentrations, leading to a reversible and dose‐dependent increase in creatinine [12, 13, 14]. As a result, physiologically‐based pharmacokinetic (PBPK) modeling has successfully been applied to olaparib for predicting DDIs, understanding patient populations, and subsequent applications, such as optimal dosing to patients with hepatic/renal impairment [15, 16].

In a similar manner, the objectives of the work were to:

• utilize in vitro test systems demonstrating specific carboxylesterase clearance of niraparib and generation of its primary metabolite M1.

• utilize virtual cancer populations in Simcyp and niraparib‐patient derived clinical data to demonstrate the applicability of such populations in PBPK.

• develop a mechanistic PBPK model for niraparib and its metabolite M1 using Simcyp with physicochemical and physiological parameters, human in vitro data, and human in vivo data to simulate niraparib exposure in cancer patients.

• apply clinical data from niraparib patients with moderate hepatic impairment or with Chinese/Japanese ethnicity to simulate niraparib exposure.

• prospectively use PBPK to contextualize clinical relevance of any in vitro DDI liability identified.

2. Methods

2.1. In Vitro Phenotyping Test Systems

Stock solutions of niraparib (0.2, 2, and 10 mM), reference substrate p‐nitrophenylacetate (p‐NPA, 200 mM), and inhibitors bis‐nitrophenyl phosphate (BNPP, 2 mM), eserine (2 mM), and 1‐aminobenzontriazole (1‐ABT, 200 mM) were prepared in a solvent or water.

2.2. Reaction Phenotyping Using Human Liver S9 and Microsomes

Mixed donor human S9 fraction and human liver microsomes (HLM) (Corning Life Sciences, Woburn, MA, USA) were prepared with 0.5 M NaKPO₄ and 0.50 mM CaCl₂ buffer at pH 7.4 or with 0.5 M NaKPO₄ buffer at pH 7.4, respectively, and inhibitor stock solution to give a final concentration of 2 and 1 mg protein/mL, respectively. S9 was pre‐incubated for 5 min (37°C), and the reaction was initiated by addition of niraparib stock solution to give a final concentration of 10 μM, whereas microsomes were pre‐incubated with niraparib first, then initiated by addition of NADPH solution.

2.3. Reaction Phenotyping Using Hepatocytes in Suspension

Mixed gender cryopreserved human hepatocytes (HH) (BioreclamationIVT, Baltimore, MD, USA) were thawed and resuspended in Williams' E medium supplemented with L‐glutamine and 15 mM HEPES pH 7.4 (incubation medium). Incubations (n = 3/condition) were prepared in 24‐well plates, containing 0.15 million cells/mL, and pre‐incubated at 37°C in 95% humidified atmosphere containing 5% CO₂ for 15 min at 150 rpm in a humidified incubator. Selective inhibitors were then added for a further 15‐min incubation before adding niraparib or p‐NPA at target concentrations of 10 and 50 μM, or 1 mM, respectively, to give a final volume of 1.2 mL.

2.4. Reaction Phenotyping Using Plated Hepatocytes

Plateable cryopreserved human hepatocytes (HH) (ThermoFisher Scientific, single donor) were thawed in recovery medium, viability was checked using the trypan blue exclusion method, and the cells were resuspended in primary cell plating medium (UPCM). Hepatocytes (0.15 million cells in 0.20 mL) were seeded into each well of type I collagen‐coated flat‐bottom 48‐well plates and incubated at 37°C in a 95% humidified atmosphere containing 5% CO₂. After 4 h, the non‐adherent cells were removed by washing with incubation medium. Niraparib and p‐NPA stock were diluted in the culture medium, and incubations with human hepatocytes were carried out in supplemented Williams' E medium with a total incubation volume of 0.2 mL.

2.5. Reaction Phenotyping Using Supersomes and Recombinant Amidase

Baculovirus‐transfected insect cell systems with either single expressed human carboxylesterase (RhCES1b, 1c, and 2) or cytochrome P450 singly expressed enzyme CYP3A4 (Corning Life Sciences, Woburn, MA, USA) and recombinant amidase from Pseudomonas aeruginosa expressed in E. coli (Sigma, Saint Louis, MO, USA) were used to assess niraparib metabolism. Incubations were carried out with 0.1 mg recombinant protein (CES) or 100 pmol cytochrome P450 per mL incubate, both with cofactor mixture (containing 0.50 mg glucose‐6‐phosphate, 0.25 units of glucose‐6‐phosphate dehydrogenase, and 0.5 mg of MgCl₂·6H₂O, 0.125 mg NADP in 0.5 M NaKPO₄ buffer at pH 7.4), or recombinant amidase at 4.1 or 41 μg protein per mL phosphate buffer (pH 7.4). After a pre‐incubation of 5 min (37°C), reactions were initiated by the addition of stock solutions of niraparib or p‐NPA to give target concentrations of 1 and 10 μM, or 1 mM, respectively. The protein concentration was 0.020 mg/mL recombinant protein during testing the reference substrate p‐NPA, and the samples were not quenched.

The stability of niraparib was measured through incubation with control matrices, and incubation times of 0, 30, and 120 min (supersomes), plus 180 and 240 min for HH suspension, and plated HH at 24, 48, and 72 h. The enzymatic reactions were stopped by quenching with acetonitrile/methanol (1/1). All samples were stored at −20°C and analyzed for niraparib by LC–MS, based on peak area. The concentrations of metabolites M1 and M10 were determined using a qualified LC–MS method. Reference p‐NPA was incubated up to 10 min in matrices, and product formation was detected by monitoring the absorbance increase per minute at 410 nm and normalized against blank control.

2.6. PBPK Modeling of Niraparib and Its Metabolite

Whole‐body PBPK modeling and simulations were performed using Simcyp (version 22; Certara, Sheffield, UK) and a modified virtual cancer patient population. Simulations were conducted using a fixed trial design, where appropriate elements (e.g., dosing regimen, number of subjects, age range, proportion of females, and clinical chemistry) matched the clinical study and with n, the number of simulated trials to give ≥ 100 subjects. All clinical data utilized are available from published studies (e.g., ascending multiple‐dose FTiH study [8], a food effect study [17], a human ADME study [10], and additionally with metabolite M1 from a hepatic impairment study [18] and repeat dose combination study [19]). Further studies evaluated niraparib in Chinese [20] and Japanese [21] cancer patients in the absence of comorbidity populations.

In brief, the model was developed using the physicochemical properties and measured in vitro and observed clinical data. The details for the model input parameters are shown in Table 1. The advanced dissolution, absorption, and metabolism (ADAM) model was used to predict the oral absorption based on in vitro permeability data from the LLC‐PK1 cell line to give an effective human permeability (Peff,man) of 2.49 × 10⁻⁴ cm/s with an absorption scalar based on observed human exposure. Niraparib is classified as BCS class II compound (high permeability, low solubility) based on the highest strength of 300 mg administered (as capsules or immediate release tablets) and measured pH‐independent solubility (0.7–1.1 mg/mL) across the physiological pH range. The distribution was predicted using method 2 (Rodgers and Rowland, no scalar), giving a predicted Vss value of 12.4 L/kg, demonstrating extensive tissue distribution consistent with the observed clinical PK data. Intestinal efflux or hepatic uptake transporters were not considered in the model due to the high permeability, high Fa observed, and clinical dose linearity (30–400 mg). Estimated in vivo oral clearance (CLpo, 14.2 L/h) was used in the PBPK model where only niraparib was reported, whereas, enzymatic clearance was used and optimized to CES1 (4.0 μL/min/mg protein) with predicted unbound cell value of 0.815 (Fu,inc) in a top‐down manner to simulate primary metabolite M1 profiles. Renal clearance of niraparib was determined using fraction unbound in plasma (fu,p) * glomerular filtration rate (GFR) (1.275 L/h) as equivalent to the average determined renal clearance (0.98–1.41 L/h) [8, 22].

TABLE 1.

PBPK Model Parameters for Niraparib and its Metabolite M1.

Compound	Niraparib		Metabolite M1
Parameter	Value	Source	Value	Source
Phys chem and blood binding
Molecular Weight (g/mol)	320.4	Measured	321.4	Predicted ADMETpredictor
Log P	2.46	Measured	1.896	Predicted ADMETpredictor
Compound type	Monoprotic base		Ampholyte
pKa1	9.95	Measured	4.63, 9.95	Predicted ADMETpredictor
Blood/Plasma ratio (B/P)	1.7	Measured	0.61	Simcyp Optimized
Fraction unbound (Fu)	0.17	Measured	0.127	Predicted ADMETpredictor
Absorption model	ADAM		NA
Permeability, Papp = (10−6 cm/s)	18 [28]	Measured [Verapamil as positive control]	4.39	Measured
Permeability, Peff = (10−4 cm/s)	2.49	SimCyp Predicted	0.256	SimCyp Predicted
Absorption Rate Scalar	2	SimCyp Optimized for best fit	NA
Formulation	IR	Capsule or Tablet as appropriate	NA
Solubility profile	1.04	Measured (pH 6.8)	NA
Intrinsic Solubility, S 0 (mg/mL)	0.0007	SimCyp Predicted/SIVA v4 Module 3	NA
Distribution model	Full PBPK		Full PBPK
Kp Method	2	SimCyp Predicted [Rodgers & Rowland]	2	SimCyp Predicted [Rodgers & Rowland]
Vss (L/kg)	12.4	SimCyp Predicted [Rodgers & Rowland]	0.352	SimCyp Predicted [Rodgers & Rowland]
Scalar	1	Default	1	Default
Elimination model	In vivo
CLpo (L/h)	14.2 (50%)	Measured (Observed CV%)	14.2 (30%)	Estimated to be equivalent to niraparib (Default CV%)
CLr (L/h)	1.275	Predicted Fu,p * GFR = 0.17*7.5 L/h = 1.275 L/h	0.953	Predicted Fu,p * GFR = 0.127*7.5 L/h = 0.953 L/h
	Enzymatic
CES 1 Clint (uL/min/mg prot)	4.0	Optimized based on observed parent & metabolite studies	NA
Fu_inc	0.815	SimCyp Predicted	NA
Interactions	Value (μM)
CYP1A2	6.76; 6.34‐fold	Measured in vitro EC50 and Emax for induction	NA
CYP2C19	91	Measured in vitro IC50 for inhibition	NA
CYP2D6	73	Measured in vitro IC50 for inhibition	NA
PgP	131	Measured in vitro IC50 for inhibition	NA
BCRP	5.8	Measured in vitro IC50 for inhibition	NA
OCT‐1	34.1	Measured in vitro IC50 for inhibition	NA
MATE‐1	0.18	Measured in vitro IC50 for inhibition	NA
MATE‐2K	0.14	Measured in vitro IC50 for inhibition	NA

In a similar manner, the M1 model was developed using physicochemical properties calculated from the structure using ADMET predictor v10 (Simulations Plus, Inc. Lancaster, CA) and in vitro study data. Blood to plasma ratio was optimized as a sensitive parameter to match the observed profile. A full PBPK model was employed using method 2, giving a low predicted Vss value of 0.352 L/kg, consistent with an acid. M1 clearance was estimated to be equivalent to niraparib, demonstrated by the parallel elimination phase, indicating formation rate‐dependent clearance.

PBPK models were validated and verified (V/V) with clinical PK data before being applied to ethnic or hepatic impaired cancer populations, or prospective DDI assessment. The predictive performance was evaluated by overlaying the observed data with the simulated mean, 5th and 95th percentile concentration‐time profiles. Accuracy was determined by predicted‐to‐observed (P/O) ratios for PK parameters after single and multiple doses within a predefined 2‐fold acceptance criteria.

3. Results

3.1. In Vitro

In vitro incubations characterized the disappearance of niraparib with or without formation of metabolites. Comparable results were observed between plated human hepatocytes incubated at 10 and 50 μM niraparib for up to 72 h. Incubations with inhibitors 1‐ABT and eserine were comparable to controls, where the average remaining percentage of niraparib across both concentrations was 15%, 14%, and 14%, respectively. In contrast, BNPP incubations resulted in no loss of niraparib compared to time zero (Figure 1a).

FIGURE 1.

In vitro clearance of niraparib and formation of metabolites using cellular and sub‐cellular systems with specific inhibitors. Bar charts show in vitro incubations of niraparib in plated human hepatocytes and percentage of niraparib remaining up to 72 h at two different concentrations, 10 and 50 μM (a), and the normalized percentage of the total appearance of the primary metabolite M1 from incubations at 50 μM with human liver S9 (b), and human liver microsomes (c) up to 120 min, and additional appearance of the subsequent glucuronide M10 (relative to M1) with suspended human hepatocytes up to 120 min (d) and plated human hepatocytes up to 72 h (e), in the absence of any inhibitor, control (red), and in the presence of inhibitors, 1‐ABT (orange), eserine (green), or BNPP (blue), with supersome CES1b, CES1c, CES2, Amidase, and CYP3A4 incubations at 30 and 120 min identifying the enzymatic pathways responsible for metabolizing niraparib by forming M1 metabolite (f).

Formation of M1 in control human liver S9 and microsomes was 0.144 and 0.208 nmol/mL, respectively, compared to 0.103 and 0.149 nmol/mL in eserine‐treated S9 and microsomes following incubation of niraparib at 10 and 50 μM. CYP3A4 inhibition assessed in microsomes using 1‐ABT was 0.287 nmol/mL, whilst M1 was not detected in BNPP incubates in either matrix up to 120 min (Figure 1b,c).

Human hepatocytes (suspension and plated) controls formed 5.09 and 10.6 nmol/mL of M1 after 240 min and 72 h, respectively. The viability of the hepatocytes before use in the present study was 80.3% and 88.6%, respectively. Incubations with the inhibitors 1‐ABT and eserine generated 4.25 and 4.15 nmol/mL in suspension after 240 min and 9.56 and 9.65 nmol/mL in plated after 72 h, respectively. Corresponding downstream glucuronide M10 was observed in control, 1‐ABT, and eserine incubations at 0.46, 0.28, and 0.31 nmol/mL in suspension and at 2.07, 1.51, and 2.01 nmol/mL in plated hepatocytes, respectively; M10 was not observed in BNPP incubates (Figure 1d,e). In supersomes containing CES1b, CES1c, CES2, rCYP3A4, or reductase/amidase, formation of M1 was only observed in CES1b and CES1c (36 and 40 pmol/mL, respectively) (Figure 1f).

3.2. PBPK Model Validation and Verification

Niraparib PBPK model was validated and verified using 11 single and repeat dose levels (30–400 mg) from the FTiH study and the 300 mg dose level in the human ADME and food effect studies. Data from HEPATIC (control cohort) impairment study and IOLite study were used for M1 assessment. The model captured the plasma concentration‐time profiles as shown by representative plots in Figure 2. Predicted versus observed plots for Cmax and AUC mean values ± standard deviation are shown in Figures 3 and 4 for the model V/V and extended verification/application, respectively.

FIGURE 2.

Representative simulated PBPK and observed oral PK profiles of niraparib. Plots show plasma concentration (y‐axis) against time, hours (x‐axis) with individual subject data points given (open circles, ), observed mean (solid red line, —) and predicted mean (solid black line, —) and 5:95 percentile (dashed black line, —) from the single ascending first‐time in patient study (Sandhu et al. 2013) following a single dose of niraparib at 30 mg (a) and 300 mg (b); from normal hepatic control cancer patients (Akce et al. 2021) following a single 300 mg dose of niraparib (c) and the resulting M1 metabolite (d); from cancer subjects receiving 21 days repeat dose of 300 mg niraparib (e) and the resulting M1 metabolite (f) (Yap et al. 2022).

FIGURE 3.

Predicted versus observed Cmax and AUC parameters from analyzed clinical studies used in the development, verification and validation of the niraparib PBPK model. Plots show each niraparib clinical study simulated with the predicted values (y‐axis) against the observed values (x‐axis) for the Mean ± Standard Deviation (SD) of Cmax (a) and AUC (b). Diagonal lines represent line of unity (solid line), 0.8 to 1.25‐fold (dotted lines····), 2‐fold (dashed lines ‐ ‐), and 3‐fold (longer dashed lines — —). Clinical studies are represented by blue circles (, ) for Sandhu et al. (2013); purple circle () for Akce et al. (2021); and orange triangles (, ) for Yap et al. (2022), with solid and open symbols indicating Day 1 and Day 21; yellow diamonds (, ) for van Andel et al. (2018) with solid and open symbols indicating Part 1 and Part 2 of the hADME; green squares (, ) for Moore et al. (2018) with solid and open symbols indicating Fasted and Fed cohorts; and in each instance SD shown by solid black error bars for both predicted (vertical) and observed (horizontal), respectively.

FIGURE 4.

Predicted versus observed Cmax and AUC parameters from analyzed clinical ethnic and hepatic studies used in the extended verification and application of the niraparib PBPK model. Plots show each niraparib clinical study simulated with the predicted values (y‐axis) against the observed values (x‐axis) for the Mean ± Standard Deviation (SD) of Cmax (a) and AUC (b). Diagonal lines represent line of unity (solid line), 0.8‐ to 1.25‐fold (dotted lines····), 2‐fold (dashed lines ‐ ‐), and 3‐fold (longer dashed lines — —). Clinical studies are represented by red symbols (, , , ) for Zhang et al. (2020); purple symbols (, ) for Akce et al. (2021); and black symbols (, , , ) for Yonemori et al. (2021), with solid and open symbols indicating Day 1 and Day 21, and where circles represent the simulation as a virtual cancer population and squares as the virtual special populations, either as ethnic populations (Chinese or Japanese) or as moderate Hepatic Impaired population; and in each instance SD shown by solid black error bars for both predicted (vertical) and observed (horizontal), respectively.

All pharmacokinetic parameters were within the 2‐fold acceptance criteria for predicted‐observed mean ratios across all studies, except on Day 21 using the Japanese population (Table S1).

3.3. Drug–Drug Interaction Assessment

Results of the PBPK simulations for the DDI assessment are given as geometric mean ratio (GMR) and 90% confidence interval for AUC and Cmax, comparing substrate exposure with and without repeated doses of niraparib at 300 mg, and where any risk was identified, lower doses of 100 and 200 mg were assessed, respectively (Table 2).

TABLE 2.

Predicted drug–drug interaction profile of niraparib as a precipitant following repeat oral dosing at 100, 200, or 300 mg with clinical probe substrates using PBPK modeling.

Enzyme/Transporter	Clinical probe substrate	Precipitant niraparib dose	Geometric mean ratio (GMR) a
			AUC	(90% CI)	Cmax	(90% CI)
CYP1A2 b	Caffeine	100 mg	0.85	0.84–0.87	0.97	0.96–0.97
		200 mg	0.75	0.74–0.77	0.94	0.93–0.95
		300 mg	0.68	0.66–0.70	0.92	0.91–0.93
	Theophylline	100 mg	0.88	0.87–0.89	0.99	0.98–0.99
		200 mg	0.79	0.77–0.81	0.97	0.97–0.98
		300 mg	0.73	0.71–0.75	0.96	0.96–0.97
	Phenacetin	100 mg	0.87	0.86–0.88	0.92	0.91–0.93
		200 mg	0.78	0.76–0.80	0.85	0.84–0.87
		300 mg	0.71	0.69–0.73	0.80	0.78–0.82
	Clozapine	100 mg	0.90	0.89–0.91	0.96	0.95–0.96
		200 mg	0.82	0.81–0.84	0.92	0.91–0.93
		300 mg	0.77	0.75–0.79	0.89	0.88–0.91
CYP2C19	Mephenytoin	300 mg	1.02	1.02–1.02	1.01	1.01–1.01
	Tolbutamide	300 mg	1.00	1.00–1.00	1.00	1.00–1.00
CYP2D6	Dextromethorphan	300 mg	1.04	1.03–1.03	1.03	1.03–1.03
	Nebivolol	300 mg	1.02	1.02–1.02	1.03	1.03–1.03
P‐gp	Dabigatran Etexilate	300 mg	1.01	1.01–1.01	1.02	1.02–1.02
	Digoxin	300 mg	1.00	1.00–1.00	1.00	1.00–1.00
BCRP	Rosuvastatin	300 mg	1.08	1.08–1.09	1.18	1.17–1.20
	Sulfasalazine	300 mg	1.18	1.16–1.20	1.19	1.17–1.21
MATE‐1/MATE‐2K and OCT1 c	Metformin	100 mg	1.71	1.67–1.76	1.40	1.38–1.43
		200 mg	2.52	2.41–2.63	1.71	1.66–1.77
		300 mg	2.63	2.51–2.76	1.75	1.70–1.81

^a Inhibition Category: Strong risk ≥ 5‐fold (red); Moderate risk ≥ 2‐ to < 5‐fold (orange); Weak/mild risk ≥ 1.25‐fold to < 2‐fold (yellow); No risk < 1.25‐fold (green); Induction Category: Strong risk ≥ 0.2‐fold (red); Moderate risk > 0.2‐ to ≤ 0.5‐fold (orange); Weak/mild risk > 0.5‐fold to ≤ 0.8‐fold (yellow); No risk > 0.8‐fold (green).

^b Induction predicted using EC50 6.76 μM and Emax 6.34 parameters determined for the most sensitive in vitro human hepatocyte donor (DMQ).

^c OCT1 inhibition unlikely to contribute or be clinically relevant as niraparib Cmax is approximately 1400 ng/mL in plasma following repeat dose of 300 mg on Day 21 (Sandhu 2013), which equates to < 1 μM free Cmax, and approximately 45‐fold lower than IC50 of 34.1 μM.

The PBPK model predicts no DDI risk (> 0.8‐ to < 1.25‐fold) with sensitive substrates for CYP2C19 (e.g., mephenytoin or tolbutamide) or CYP2D6 (e.g., dextromethorphan or nebivolol). The PBPK model predicts no induction risk at 100 mg repeat dosing of niraparib with specific CYP1A2 enzyme substrates (e.g., caffeine, theophylline, phenacetin, or clozapine). However, a weak DDI risk (0.74‐ to 0.77‐fold and 0.66‐ to 0.70‐fold) is predicted with caffeine at 200 mg and 300 mg repeat dosing of niraparib. The PBPK model predicts no DDI risk with transporter substrates for P‐gp (e.g., dabigatran etexilate or digoxin) or BCRP (e.g., rosuvastatin or sulfasalazine) with niraparib up to 300 mg repeat dosing. The PBPK model predicts a weak DDI risk (1.7‐ to 1.8‐fold) with clinical MATE‐1/MATE‐2‐K probe substrate metformin following 100 mg repeat dosing of niraparib which increases to a moderate DDI risk (2.4‐ to 2.8‐fold) at 200 mg and 300 mg. Additionally, simulations were conducted considering CES1 phenotype and CES1 inhibition with simvastatin; where a PM phenotype impacted AUC_(0–24), AUC_inf, half‐life and clearance by up to 8‐fold compared to observed, more than 2‐times greater difference to EM; whilst no DDI inhibition was observed with simvastatin (Table S1 and Appendix S1).

4. Discussion

The current work successfully demonstrates specific carboxylesterase (CES) clearance of niraparib to its primary metabolite using multiple in vitro test systems and integration of this metabolism into a PBPK model using a derived cancer population to describe multiple clinical studies.

More specifically, niraparib metabolism is mediated by CES1 isoforms 1b and 1c in in vitro experiments. CES1 is predominantly located in the liver hepatocytes and, to a notably lesser extent, in intestinal enterocytes [23]. It was observed that in vitro niraparib disappearance and metabolite formation rates in the systems were low, corresponding to the observed low human clearance, long half‐life, and limited metabolism to primary metabolite M1 identified in the [14C]‐radiolabelled study. In vitro data in hepatocytes demonstrated subsequent downstream production of the phase II acyl glucuronide, M10, as observed in humans [9]. Cytochrome P450 (CYPs), CES2, and amidase were shown not to be involved. CES1 can play a crucial role in the metabolism of many ester drugs, such as bioactivation of prodrugs [enalapril, imidapril (ACE inhibitors), oseltamivir, tenofovir (antiviral agents), capecitabine, irinotecan (anticancer agents)] or clearance by inactivation (clopidogrel, methylphenidate, nintendanib, remdesivir) [24].

In the current work, a middle‐out approach was employed in estimating the CES1 clearance of niraparib, utilizing both in vitro and clinical information. A bottom‐up approach was not utilized due to the very low turnover of niraparib and added uncertainty. Challenges of in vitro‐to‐in vivo extrapolation (IVIVE) for non‐CYP enzymes are well documented and an area of future opportunities [25]. A recent review of PBPK modeling for drugs cleared by non‐CYP enzymes identified twenty‐one articles focused on drugs metabolized by CES; thirteen focused on prodrugs activation [26]. In the review, clopidogrel was the most studied CES substrate, and oseltamivir was the most studied CES‐activated prodrug; a middle‐out approach was used in 17 out of 21 articles. One such article developed a CES1 bottom‐up PBPK model for eight pyrethroids [27]. The authors identified that the metabolism of these compounds was only by hepatic CES1 and CYP enzymes and generated CL_int values in HLM and cytosol in vitro systems. IVIVE was facilitated by considering an empirical free concentration adjustment factor to adjust the estimated in vivo hepatic CL_int to obtain the actual CL_{int_vivo}.

Reliability of a bottom‐up approach is further confounded by marked differences in expression and activity of CES1 between individuals, contributing to interindividual variability in the pharmacokinetics (PK) of drugs metabolized by CES1. Both genetic and nongenetic factors contribute to CES1 variability [24]. Polymorphisms of the CES1 gene have been reported to affect the metabolism of dabigatran etexilate, methylphenidate, oseltamivir, enalapril, imidapril, and clopidogrel, whereas variants of the CES2 gene have been found to affect aspirin and irinotecan and may result in the observed clinical variability [28, 29]. PBPK modeling has been used to predict oseltamivir and methylphenidate exposure affected by interplay among CES1 pharmacogenetics, drug–drug interactions, and sex [30, 31]. Xiao et al. further demonstrated the CES‐exposure relationship using a plasma protein biomarker for the hepatic metabolism of methylphenidate and CES1 pharmacogenetics for enalapril [32, 33]. An assessment of poor metaboliser (PM) phenotypes would significantly impact niraparib clearance, but these predictions did not correlate with observed niraparib and M1 clinical data across two studies; whereas using the default extensive metaboliser (EM) phenotype did.

Drug interactions with CES are less common. It is known that alcohol is a specific inhibitor of CES1, impacting the metabolism of oseltamivir and clopidogrel but not the CES2‐mediated hydrolysis of aspirin [34, 35]. Specific in vitro inhibition of CES1 has been identified for several marketed drugs (e.g., simvastatin, nitrendipine, and telmisartan) [36]. Yet the extent of clinical DDI translation involving CES1 is limited, as the interaction study between simvastatin and clopidogrel was inconclusive (most likely a result of compensatory pathways), and the impact of dexamethasone on oseltamivir was only minor [37, 38]. The niraparib PBPK model explored a potential DDI with simvastatin acting as a perpetrator on CES1 and predicted no interaction. The model demonstrated that clinically relevant concentrations of simvastatin are not achieved and thus most likely why no CES1 DDI has been observed. A lack of specific CES substrates and inhibitors verified by clinical studies remains. Therefore, DDIs with niraparib as a victim are unlikely based upon current knowledge and its already low clearance; however, assessment of pharmacogenomic variants could be warranted.

The intended target population for niraparib is cancer patients [39]. The predictive performance of PBPK modeling in cancer populations has been reviewed in recent years [40, 41]. The Simcyp cancer‐solid population has been previously verified by assessing exposure variability of several drugs (e.g., axitinib, saquinavir, midazolam, and olaparib) [15, 42, 43]. Due to the similar demographic and clinical chemistry data in patients receiving niraparib and application to another PARPi, olaparib, the previously published Schwenger et al. cancer population was used as the basis for the virtual population [44]. The default formula for renal clearance within the Simcyp cancer population was used to capture the wider range of observed serum creatinine values [45]. In future, it may be possible to further refine the virtual cancer patient populations for niraparib through collation of more clinical study data [46].

In this work, a robust PBPK model was qualified and verified to successfully predict niraparib and its metabolite M1 pharmacokinetic profiles and parameters across multiple clinical studies in adult cancer patients with predicted/observed ratios were within 2‐fold acceptance, as shown in Figures 3 and 4. The PBPK model captured the linearity observed across the niraparib dose range 30 to 400 mg and repeat dosing up to 21 days (1 treatment cycle) [8] and moderate‐high degree of intra‐ and inter‐individual variability (40%–100% CV) has been reported in a bioequivalence study [7]. Ideally, a human ADME study would be the cornerstone to developing a PBPK model. However, > 2‐fold differences were reported in parameters (Cmax, AUClast, AUCinf, CL/F and Vd/F) from the human ADME study, where two cohorts of six patients were orally administered unlabelled and labeled 300 mg niraparib. This high variability was also observed following intravenous microdosing of niraparib particularly for CL and Vd where CV was 74% and 91% [9, 10, 22]. For radiolabelled niraparib, half‐life (28 h) did not match the bioanalytical non‐labeled (96 h), whilst the intravenous and oral clearance aligned (5.9 L/h vs. 8.4 L/h), closely matching the calculate absolute bioavailability of 73%. Yet the volume of distribution did not match even accounting for bioavailability (194 L vs. 1220 L). Van Andel et al. suggested that this disconnect is a result of longer relative sampling/limit of quantification of the different routes (96 h vs. 504 h). Across numerous studies, reported apparent clearance and half‐life is protracted and apparent volume of distribution very high. Therefore, in this instance the intravenous microdose does not accurately reflect the true profile of niraparib, resulting in a PBPK model that does not capture any oral profile. As a result, the microdose was not used to develop the model. Additionally, analytical review of the conditions used in the analysis of the primary metabolite M1 precluded use of the hADME study, due to potential acyl‐glucuronide (M10) degradation, in the M1 model development [22]. Subsequently, improved processing and bioanalytical conditions enabled validated quantification of M1 parameters and profiles following single and repeat dosing of niraparib at 200 or 300 mg [18, 19, 21].

Currently there are limited publicly available virtual populations of both ethnicity and co‐morbidities to be assessed in PBPK models. A Chinese cancer population has been described and applied to investigate exposure differences observed in patients receiving gefitinib and imatinib with respect to CYP2D6 polymorphism and altered physiology, respectively [47, 48]. Hence, here in the absence of clinically relevant population descriptors, Chinese or Japanese cancer patients receiving niraparib were assessed either as their virtual ethnic population or cancer population, which assumes that CES1 expression is represented correctly for niraparib metabolism in each population. As shown in Figure 4, simulations using the virtual Chinese population and the cancer population were within the clinically observed ranges [20], whereas the virtual Japanese population slightly over‐predicted exposures [18, 21]. This suggests that ethnicity may play a role, as genotype frequencies of CES1 (rs2244613 and rs8192935) differ significantly between healthy Chinese and Caucasians, and CES1 rs8192935 was associated with the peak concentration of dabigatran, but with no gender differences [49]. In contrast, a published clopidogrel model was first developed to capture the clopidogrel metabolic pathways (CES1 and CYPs) for a European population and was extrapolated to a Japanese population [50]. Given niraparib undergoes hepatic and renal elimination, its pharmacokinetics can be impacted in case of impairment of these organs. Accordingly, niraparib AUCinf was increased by 56% in patients with moderate hepatic impairment compared with subjects with normal hepatic function, which was associated with a lower recommended dose [18]. This clinical observation was confirmed with the PBPK model using a hepatically impaired population, but it is unclear whether the impact of moderate hepatic impairment is due to changes in CES1 expression/activity in combination with pathophysiological liver changes. However, whilst the predictions using special population (ethnic or hepatic impaired) were consistently higher than the virtual cancer population, both were generally in good agreement (within 2‐fold) as shown in Figure 4.

Of the four most prominent PARP inhibitors approved in the United States and Europe (niraparib, olaparib, rucaparib, and talazoparib), only olaparib has a published PBPK model which has been applied to bridging formulations, DDIs, and patient populations [15]. Additionally, it has been used for optimizing dosing regimen and predictions when olaparib is co‐administered with CYP3A4 modulators and patients with hepatic/renal impairment [16]. Whilst olaparib has several CYP DDI liabilities, these are well characterized using in vitro, in vivo, clinical studies, and PBPK. Additionally, olaparib is also an inhibitor of numerous transporters, including P‐gp, OCT1, OAT1B1, OAT3, OCT2, MATE‐1, and MATE2‐K [14]. A recent nonclinical study has shown that olaparib significantly changed the pharmacokinetics of the MATE‐1/MATE‐2K probe substrate metformin in rats and increased the AUC‐ratio by 1.74‐fold, which may be of clinical importance [51]. A similar DDI profile is observed for rucaparib [52]. Whilst another PARPi, talazoparib appears to have no discernible perpetrator interactions but is a potential object of P‐gp and BCRP inhibition [53].

Niraparib is not a CYP substrate and thus is not liable to CYP‐mediated interactions. CES‐mediated interactions are generally less of a clinical concern but remain a developing area of research. For niraparib, following repeat doses up to the clinical maximum dose of 300 mg daily, PBPK modeling predicts no inhibitory DDI risk with CYP enzyme substrates (CYP2C19 or CYP2D6) or transporter substrates (P‐gp or BCRP) which were previously identified in in vitro experiments. However, niraparib has a predicted weak induction risk following repeat dosing of niraparib at 200 and 300 mg with clinical CYP1A2 probe substrates, such as caffeine. Additionally, like olaparib, there is a weak DDI risk (< 2‐fold) with clinical MATE‐1/MATE‐2‐K probe substrate metformin following 100 mg repeat dosing of niraparib which increases to a moderate DDI risk (≥ 2‐ to 3‐fold) at 200 mg and 300 mg repeat dosing of niraparib. Therefore, caution is recommended when niraparib is administered concomitantly with MATE‐1/MATE2‐K substrates such as metformin. These prospective simulations, in the absence of a dedicated clinical DDI study, are consistent with the clinical observations of increased serum creatinine and glycaemia following niraparib or olaparib administration [54, 55].

The current work demonstrates the utility of prospective PBPK modeling in the assessment of niraparib model‐informed clinical precipitant DDI liabilities, which is important given the challenges of conducting clinical DDI studies in cancer patients. The outcome of this PBPK modeling work provides potential future application to understanding the mechanistic disposition of niraparib, interpatient variability as a substrate of a CES1, and exploring niraparib pharmacokinetics in different biopharmaceutic formulations and populations (e.g., pediatric).

Author Contributions

G.J.L., R.C.J., A.S.K. and K.S.T. wrote the manuscript. G.J.L. and K.S.T. designed the research. G.J.L. performed the research. G.J.L, R.C.J., A.S.K. and K.S.T. analyzed the data.

Funding

The authors have nothing to report.

Conflicts of Interest

G.J.L., R.C.J., A.S.K. and K.S.T. are employees of GSK and own stocks/shares in GSK.

Supporting information

Table S1: A Summary Table of Predicted and Observed PK Parameters for all Clinical Studies used in the PBPK model Validation and Verification of Niraparib.

Appendix S1: Additional Assessment of CES1 Phenotype and CES1 Inhibition by Simvastatin.

Lewis G. J., Jewell R. C., Krishnatry A. S., and Taskar K. S., “Physiologically‐Based Pharmacokinetic Modeling of the PARP Inhibitor Niraparib,” CPT: Pharmacometrics & Systems Pharmacology 15, no. 1 (2026): e70182, 10.1002/psp4.70182.