A physiologically‐based pharmacokinetic/pharmacodynamic modeling approach for drug–drug‐gene interaction evaluation of S‐warfarin with fluconazole

Abstract

Warfarin is a widely used anticoagulant, and its S‐enantiomer has higher potency compared to the R‐enantiomer. S‐warfarin is mainly metabolized by cytochrome P450 (CYP) 2C9, and its pharmacological target is vitamin K epoxide reductase complex subunit 1 (VKORC1). Both CYP2C9 and VKORC1 have genetic polymorphisms, leading to large variations in the pharmacokinetics (PKs) and pharmacodynamics (PDs) of warfarin in the population. This makes dosage management of warfarin difficult, especially in the case of drug–drug interactions (DDIs). This study provides a whole‐body physiologically‐based pharmacokinetic/PD (PBPK/PD) model of S‐warfarin for predicting the effects of drug–drug−gene interactions on S‐warfarin PKs and PDs. The PBPK/PD model of S‐warfarin was developed in PK‐Sim and MoBi. Drug‐dependent parameters were obtained from the literature or optimized. Of the 34 S‐warfarin plasma concentration‐time profiles used, 96% predicted plasma concentrations within twofold range compared to observed data. For S‐warfarin plasma concentration‐time profiles with CYP2C9 genotype, 364 of 386 predicted plasma concentration values (~94%) fell within the twofold of the observed values. This model was tested in DDI predictions with fluconazole as CYP2C9 perpetrators, with all predicted DDI area under the plasma concentration‐time curve to the last measurable timepoint (AUC_last) ratio within twofold of the observed values. The anticoagulant effect of S‐warfarin was described using an indirect response model, with all predicted international normalized ratio (INR) within twofold of the observed values. This model also incorporates a dose‐adjustment method that can be used for dose adjustment and predict INR when warfarin is used in combination with CYP2C9 perpetrators.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

The pharmacokinetics (PKs) and pharmacodynamics (PDs) of warfarin are influenced by genetic polymorphisms in cytochrome P450 (CYP) 2C9 and vitamin K epoxide reductase complex subunit 1 (VKORC1), making it difficult to manage dosage, especially in combination with CYP2C9 inhibitors.

• WHAT QUESTION DID THIS STUDY ADDRESS?

The PKs and PDs of S‐warfarin have been extensively studied using physiologically‐based PK/PD modeling. This model can be used to assess the effects of drug–drug−gene interactions on S‐warfarin PKs and PDs.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

The effect of CYP2C9 inhibitors on S‐warfarin PKs/PDs can be accurately predicted, and a method for dose adjustment was incorporated.

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

Provides a tool for dose adjustment in populations with different CYP2C9 and VKORC1 genotypes when warfarin is combined with CYP2C9 inhibitors.

INTRODUCTION

Warfarin is one of the most widely used oral anticoagulants in the world, having been synthesized in the 1940s and first used clinically in 1953. ¹ Despite the increasing prevalence of direct oral anticoagulants (DOACs), warfarin remains essential in the management of antithrombotic states. ² However, patients administered warfarin need frequent international normalized ratio (INR) monitoring to prevent bleeding events due to the narrower therapeutic window.

Clinically, warfarin is administered as a racemate, with equal amounts of the R‐/S‐enantiomers. Warfarin acts as an indirect anticoagulant by inhibiting vitamin K epoxide reductase complex subunit 1 (VKORC1). ³ VKORC1 reduces vitamin K to vitamin K hydroquinone (VKH2), which is required for the synthesis of important coagulation factors, such as II, VII, IX, and X. After oral administration, warfarin is rapidly absorbed with a bioavailability close to 100%. ⁴ Warfarin is extensively (~95%) converted to metabolites excreted in the urine, ⁵ with the major primary metabolic pathway being the oxidative pathway. ⁶ Conjugative pathways can further metabolize the oxidized products. ⁷

For S‐warfarin, more than 80% are metabolized by cytochrome P450 (CYP)2C9, and the product is 7‐ or 6‐hydroxy (OH)‐S‐warfarin. ⁸ In addition, CYP3A4 is thought to be the main enzyme involved in metabolizing warfarin to 10‐OH‐S‐warfarin. ⁹ Further metabolism of 7‐ or 6‐OH‐S‐warfarin by UDP‐glucuronosyltransferases (UGTs), with UGT1A6 for 6‐OH‐S‐warfarin and UGT1A1 for 7‐OH‐S‐warfarin. ⁷ However, 10‐OH‐S‐warfarin is probably reduced by carbonyl reductase 1 (CBR1). ¹⁰ Most studies suggested that S‐warfarin has three to five times higher anticoagulant activity than R‐warfarin. ¹¹ , ¹² , ¹³ In this study, we mainly consider the modeling of S‐warfarin.

CYP2C9 is susceptible to a large number of genetic polymorphisms, CYP2C9 variant alleles *2 (Arg144Cys) or *3 (Ile359Leu), which have been shown to be strongly associated with reduced CYP2C9 catalytic activity. ¹⁴ The genetic polymorphism of VKORC1 (c.‐1639G>A, rs9923231) has been found to have a significant correlation with warfarin pharmacodynamics (PDs). ¹⁵ , ¹⁶ These two genetic polymorphisms were incorporated into the modeling process.

Fluconazole is a commonly used antifungal drug in clinical practice, which is a moderate inhibitor of CYP2C9 and CYP3A4. ¹⁷ , ¹⁸ The concomitant use of fluconazole is required in some patients taking warfarin, ¹⁹ , ²⁰ , ²¹ but the interaction between the two drugs makes it difficult to manage the dose. Several studies have described the effect of fluconazole on the anticoagulant effect of warfarin, ²² , ²³ proposing pharmacokinetic (PK)‐based methods of dose adjustment. ²⁴ However, there have been no clinical trials exploring the effects of fluconazole on the PKs and PDs of warfarin in people with different CYP2C9 and VKORC1 genotypes.

Physiology‐based pharmacokinetic (PBPK) modeling is a methodology capable of integrating physiological, chemical, drug‐dependent preclinical, and clinical information to simulate drug absorption, distribution, metabolism, and excretion, and ultimately untested clinical conditions. ²⁵ The PBPK model is recommended for predicting the effects of drug–drug‐interaction (DDI) and genetic polymorphisms on drug metabolizing enzymes and transporters. ²⁶ , ²⁷

In this study, we applied the PBPK/PD modeling approach to provide dose adjustments when fluconazole is combined with warfarin in populations with different CYP2C9 and VKORC1 genetic polymorphisms.

METHODS

Software

PBPK/PD modeling was performed with the open source PK‐Sim and MoBi modeling software (version 11.2, part of the Open Systems Pharmacology Suite, ²⁸ www.opensystems‐pharmacology.org). Published plasma concentration and INR profiles were digitized using GetData Graph Digitizer (version 2.22, S. Fedorov). Parameter optimization using Monte Carlo algorithm. PK parameters were calculated using noncompartmental model analysis with ncappc package (version 0.3.0 ²⁹ ;). Graphics were compiled with R (version 4.2.3, The R Foundation for Statistical Computing) and RStudio (version 2023.06.0, RStudio, Inc.).

PBPK model development

The PBPK model for S‐warfarin is built by middle‐out approaches. ³⁰ First, we conducted an extensive literature search to collect the physicochemical properties, and absorption, distribution, metabolism, and excretion processes parameters of S‐warfarin and 6‐, 7‐, 10‐OH‐S‐warfarin. Secondly, A study with CYP2C9 genotypes, S‐warfarin and 6‐, 7‐, 10‐OH‐S‐warfarin plasma concentration profiles were used to develop a parent‐metabolite model to avoid parameter non‐identification problems. ³¹ , ³² Third, we extracted the S‐warfarin plasma concentration profiles from the clinical studies literature. These clinical studies were screened on the basis of the following criteria: (1) intravenous or oral administration of warfarin; (2) healthy volunteers; and (3) availability of plasma concentration‐time data sets for S‐warfarin. The CYP2C9 genotype information was also extracted from the literature. S‐warfarin plasma profiles were categorized into internal training and external test data sets. The studies selected for the training data set contained a wide range of information (i.e., different dosing regimens, sampling times and frequencies, and genotypes). The test data set was used for model evaluation. If subjects in the literature were taking warfarin racemate, then the dose in the model was counted as half.

Systemically relevant physiological parameters (organ volumes, blood flow, hematocrit, etc.) were provided to PK‐Sim along with a small molecule model. ³³ Drug‐specific physicochemical properties (molecular weight, lipophilicity [logP], solubility, and fraction unbound, etc.) were obtained from the literature. Organ‐plasma partition coefficients were determined using the Rodgers and Rowland method based on the literature and the optimal overlap between the observed and predicted concentration‐time data sets. ³⁴

According to the literature, ¹⁴ , ³⁵ , ³⁶ CYP enzymes metabolism, renal plasma clearance, organic anion transporter 2 (OAT2)‐mediated drug distribution and VKORC1‐mediated target‐mediated drug disposition (TMDD) were included in the models for S‐warfarin. To simplify the model, we assume that S‐warfarin is metabolized by CYP2C9 and CYP3A4. Figure 1 summarizes the simplified metabolic process. The renal plasma clearance of 6, 7‐OH‐S‐warfarin was also included. Metabolism of S‐warfarin and 6‐, 7‐, 10‐OH‐S‐warfarin is described using Michaelis–Menten kinetics. The Michaelis–Menten constant (K _m ) for CYP2C9, CYP3A4, UGT1A6, UGT1A1, and CBR1 is extracted from the literature. ⁷ , ⁹ , ¹⁰ , ³⁷ The renal plasma clearance for S‐warfarin and metabolites was defined using first‐order kinetic and extracted from literature. ¹⁴ OAT2‐mediated drug transport from the interstitial space to the intracellular in the liver is described using Michaelis–Menten equation too, the K _m extracted from literature. ³⁵ VKORC1 protein was added for binding to S‐warfarin, and the dissociation rate constant was obtained from literature. ³⁸ In the modeling, the turnover number (k _cat) for all enzymes metabolism, maximum rate of reaction for OAT2, and dissociation constant for TMDD were identified in the CYP2C9*1/*1 population. Next, only the k _cat values of the CYP2C9 and CYP3A4 enzymes were altered in different genotype populations.

FIGURE 1.

Metabolic pathway for S‐warfarin. CBR1, Carbonyl Reductase 1; CYP, cytochrome P450; UGT UDP‐glucuronosyltransferases.

Virtual individuals and virtual population characteristics

The virtual individuals in PBPK model were created based on data from healthy individuals, using gender, ethnicity, and the mean values for age, weight, and height reported in each clinical study. If demographic information was not provided, the following default values were used instead: male subjects, European, 30 years old, weight 73 kg, height 176 cm (characteristics from the PK‐Sim demographic database ³⁹ ).

A virtual population of 100 individuals was created in the PBPK model with race, proportion of female subjects, and the ranges of age, weight, and height set according to clinical trial demographics. If the literature lacked subject information, a virtual population containing 100 male subjects aged 20–50 years was created, and the virtual populations were all created according to the PK‐Sim default method. ³⁹

PBPK model evaluation

Model‐based simulations were created for comparison with observed concentration‐time data sets of S‐warfarin in different CYP2C9 genotype groups. For clinical trials that did not report information on the CYP2C9 genotype, the population was assumed to be CYP2C9*1/*1 because this genotype is the most common 2C9 polymorphism in more than 90% of American, East Asian, and 79% of European populations. ⁴⁰ To compare the variability of observed and simulated PK data sets, 68% population prediction intervals are plotted if the observed concentration‐time data set is reported as the mean (±SD; approximately the mean ± SD under the assumption of a normal distribution), whereas 95% population prediction intervals are depicted if all individual concentration‐time data sets are available. An intuitive criterion for good model performance is that the 95% population prediction interval should cover the observed individual plasma concentration‐time data sets or the observed total plasma concentration‐time data set should be within the 68% population prediction interval. ⁴¹ The goodness‐of‐fit plot was used to compare predicted versus measured concentrations, predicted maximum concentration (C _max) versus measured C _max, and predicted area under the plasma concentration‐time curve (AUC) from the time of dosing to the time of the last concentration measurement (AUC_last) versus measured AUC_last to assess the goodness of fit of the model. The average fold error (AFE) was used to compare predicted versus measured C _max and AUC_last, a twofold error is considered an acceptable error range for model evaluation. ⁴² Equation 1 of the AFE is as follows:

$$\mathrm{AFE}=\frac{\text{Predicted value}}{\text{Observed value}}$$ (1)

DDI model development and evaluation

The published fluconazole PBPK model was integrated with the S‐warfarin model to evaluate the inhibitory effect of fluconazole on S‐warfarin metabolism. ⁴³ Inhibition of CYP2C9 and CYP3A4 by fluconazole was established using a noncompetitive inhibition process. Competitive inhibition constant (K _i ) values were selected based on data from CYP2C9*1/*1 subjects in combination with fluconazole and remained unchanged in other CYP2C9 genotypes.

The DDI modeling performance was evaluated by comparison of predicted versus observed S‐warfarin PK data sets, and by calculation of DDI AUC ratios according to Equation 2.

$$\begin{array}{c}\mathrm{DDI}\mathrm{AUC}\text{ratio}=\frac{{\mathrm{AUC}}_{\text{treatment}}}{{\mathrm{AUC}}_{\text{reference}}}\end{array}$$ (2)

where AUC treatment is the AUC of S‐warfarin with fluconazole co‐treatment, and AUC reference is the AUC for warfarin administration alone.

PBPK/PD model development and evaluation

To obtain the INR values, the PBPK model was imported into MoBi, and a set of differential equations were added to define the PD model. ⁴⁴ The indirect inhibitory effect of warfarin on coagulation factors is described by the maximum effect model as Equation 3.

$$\begin{array}{c}\mathrm{E}=\frac{I_{\max }·C_{\text{Swar}}}{{\mathit{IC}}_{50}+C_{\text{Swar}}}\end{array}$$ (3)

where C _Swar is S‐warfarin concentration in plasma, I _max is the maximum possible inhibition and IC ₅₀ is the plasma concentration of warfarin for which half of I _max is achieved.

INR values for subjects with different VKORC1 genotypes were extracted from literature, ⁴⁵ and VKORC1–1639 genotypes were categorized as AA or G(G/G and A/G). The IC ₅₀ values were optimized, assuming that different VKORC1 genotypes have different IC ₅₀ values.

Model‐based simulations were created for comparison with observed INR data sets in different VKORC1 genotype groups. For clinical trials that did not report information on the VKORC1 genotype, the population was assumed to be VKORC1G because this genotype is the most common VKORC1 polymorphism in Whites and African Americans. ⁴⁶ The 68% population intervals and goodness‐of‐fit plots were plotted to compare predicted INR values with observed values.

PBPK/PD model application

After evaluating the performance of the model, the dose of warfarin required to achieve a steady‐state INR of two for different combinations of CYP2C9 and VKORC1 was determined using Monte Carlo simulation methods in a 30‐year European man. Three scenarios were simulated to assess the effect of fluconazole and warfarin on INR values. Warfarin was administered alone for 45 days. On days 11–24 or 11–45, respectively, 400 mg of fluconazole was administered in combination with warfarin. To identify warfarin dose adjustments in different scenarios, the formula proposed by Kunze et al. ²⁴ was incorporated into the PD model. The formula is presented as Equation 4:

$$\begin{array}{c}{D}_{\mathrm{new}}=D_{\mathrm{old}}\left[\frac{f_m}{1+\frac{\left[I\right]}{K_i}}+1-f_m\right]\end{array}$$ (4)

where D _old is the dose that brings the steady‐state INR to 2 when warfarin is used alone, f _m is the fraction of CYP2C9 enzyme metabolism, K _i is competitive inhibition constant, [I] is the plasma concentration of fluconazole, D _new is the dose that brings the steady‐state INR to two when warfarin is used with fluconazole. The f _m value is assumed to be 0.87, which is the average value from the original literature and is close to the model simulation results. ²⁴

After obtaining the dose‐adjustment scheme given by the formula, the simulation with the new dose scheme was rerun to assess whether the INR value reached the target value 2.

RESULTS

PBPK model evaluation

In all 34 S‐warfarin concentration‐time from clinical trials, there were 22 single‐dose dosing data, six multidose dosing data, and six data in combination with fluconazole (see Tables S1–S5). Administration protocols included oral and intravenous routes, with individual doses ranging from 0.375 to 0.75 mg/kg and 3 to 30 mg. The participants were all healthy volunteers, with an age range of 18–65 years, and a body weight of 43.1–103.9 kg.

The modeling process is shown in Figure S1. The input parameters used in the PBPK model for S‐warfarin and 6‐, 7‐, 10‐OH‐S‐warfarin are listed in Table S4. A generic whole body 18 compartment PBPK model implemented in PK‐Sim was used. Each organ compartment comprises four inner subcompartments, that is, blood, plasma, interstitial, and intracellular spaces. ⁴⁷ The k _cat values were optimized using plasma concentration profiles of different CYP2C9 genotypes, and for CYP2C9*1/*1, *1/*3, *2/*3, and *3/*3, the k _cat values metabolized to 6‐OH‐S‐warfarin and 7‐OH‐S‐warfarin were reduced sequentially, respectively.

Figures 2 and 3 show the population profile predicted by the model for a single dose and multidose compared to the corresponding observed data. Simulated plasma profiles of S‐warfarin for bolus administrations as well as orally are in close concordance with observed data. For different CYP2C9 genotypes, the data are also well predicted.

FIGURE 2.

S‐warfarin predicted and observed plasma concentration–time profiles after single dose. (a–k) In different CYP2C9 genotypes, (l‐t) in unknow CYP2C9 genotypes. Population simulations (n = 100) are shown as lines with shaded areas (geometric mean and 95% population prediction intervals or geometric standard deviation). Observed data are shown as circles ± standard deviation if available. Predicted and observed areas under the plasma concentration–time curve from the first to the last data point (AUC_last) are compared in Tables S1 and S2. CYP, cytochrome P450; iv, intravenously; Plasma conc, S‐warfarin plasma concentration; po, oral; S‐War, S‐warfarin.

FIGURE 3.

S‐warfarin predicted and observed plasma concentration–time profiles after multiple doses. Population simulations (n = 100) are shown as lines with shaded areas (geometric mean and geometric standard deviation). Observed data are shown as circles ± standard deviation if available. Predicted and observed areas under the plasma concentration–time curve from the first to the last data point (AUC_last) are compared in Table S2. D, day of treatment; po oral; Plasma conc, S‐warfarin plasma concentration; qd, once daily; S‐War, S‐warfarin.

The goodness‐of‐fit plots of predicted versus observed plasma concentrations, as well as predicted versus observed AUC_last values and predicted versus observed C _max values are shown in Figure 4. For warfarin taken alone, 534 out of 561 predicted plasma concentration values (~95%) fell within the twofold acceptance criterion, and all 27 predicted AUC_last and C _max values fell within the twofold acceptance criterion.

FIGURE 4.

Goodness‐of‐fit plot of the PBPK model of S‐warfarin. Predicted versus observed plasma concentration (a), AUC (b), and C _max (c) of S‐warfarin from all clinical studies, predicted versus observed AUC ratio (d) from DDI clinical studies. The black solid lines mark the lines of identity and the prediction acceptance limits proposed by Guest et al. ⁵⁰ Black dotted lines indicate 1.25‐fold, black dashed lines indicate twofold deviation. Different colors represent different clinical trials. AUC, area under the concentration–time curve; C _max, maximum concentration; Conc, S‐warfarin plasma concentration; DDI, drug–drug interaction; PBPK, physiologically‐based pharmacokinetic.

Figure S2 compares the predicted and observed values of the simulated population profile for the plasma concentrations of metabolites. For CYP2C9*1/*1, *1/*3, and *2/*3, the predicted and observed values of 6‐, 7‐, 10‐OH‐S‐warfarin are close. Figure S3 shows the simulated fraction metabolized of 6, 7‐OH‐S‐warfarin in different CYP2C9 genotypes. Due to the different elimination rates for the different CYP2C9 genotypes, simulations were performed for different periods. For 6‐ and 7‐OH‐S‐warfarin, the metabolic fractions were 0.28, 0.17, 0.16, and 0.17, and 0.66, 0.76, 0.70, and 0.69 at 2C9*1/*1, *1/*3, *2/*3, and *3/*3, respectively. The total fraction metabolized for CYP2C9 is 0.94, 0.93, 0.86, and 0.86 at CYP2C9*1/*1, *1/*3, *2/*3, and *3/*3. Figure S4 shows a goodness‐of‐fit plot of the metabolite's plasma concentration, with 99 out of 129 predicted values (~77%) fell within the twofold acceptance criterion. In summary, this ensures the metabolic fractions of the different enzymes are within reasonable limits.

DDI model evaluation

The fluconazole model was evaluated using individual fluconazole plasma concentration data from a DDI study, ²³ and the results are shown in Figure S5, where all predicted values fell within the twofold. The final competitive inhibition constant values selected for CYP2C9 and CYP3A4 were 19.0 and 19.3 μM, respectively, consistent with those reported in the literature. ⁴⁸ , ⁴⁹ The PBPK model was used to predict DDI scenarios for warfarin administration with fluconazole in different CYP2C9 genotypes. The simulated plasma concentration–time profiles are compared to the corresponding profiles observed from a clinical DDI study in Figure 5. For warfarin taken with fluconazole, 230 out of 235 predicted plasma concentration values (~98%) fell within the twofold acceptance criterion and all six predicted AUC_last and C _max values fell within the twofold acceptance criterion, as shows in Figure 4. The correlation of predicted and observed DDI AUC_last ratios of all analyzed clinical DDI studies is shown in Figure 4d, further demonstrating the good DDI performance with all predicted DDI ratios within the limits proposed by Guest et al. ⁵⁰

FIGURE 5.

S‐warfarin predicted and observed plasma concentration–time profiles after single dose with and without fluconazole. (a–e) In different CYP2C9 genotypes. Population simulations (n = 100) are shown as lines with shaded areas (geometric mean and 95% population prediction intervals (a–d) or geometric standard deviation (e, f). Observed data is shown as circles. CYP, cytochrome P450; FCZ, fluconazole; po, oral; Plasma conc, S‐warfarin plasma concentration; S‐War, S‐warfarin; TAD, time after dose.

PD model evaluation

A total of eight clinical trials provided INR values, four of which had VKORC1 genotype used to determine IC ₅₀ values. The IC ₅₀ values corresponding to the VKORC1AA and VKORC1G genotypes were 0.027 and 0.043 μM. All parameters of PD model are shown in the Supplementary Methods.

The effect–time profiles of INR predicted by the model in comparison to observed measurements are shown in Figure 6. Changes of INR were well‐predicted for both multiple dosing of warfarin and in combination with fluconazole. Figure S6 shows a plot of the goodness‐of‐fit of the INR, with all values within the twofold range, and 80 out of 91 predicted INR values (~88%) fell within the 1.25‐fold range. This demonstrates that the PD model adequately predicts the PD properties of S‐warfarin.

FIGURE 6.

Predicted and observed INR–time profiles. (a–d) In different VKORC1 genotypes and after multiple doses, (e) after single dose, (f) after single dose and with fluconazole, (g, h) after multiple doses. Population simulations (n = 100) are shown as lines with shaded areas (geometric mean and geometric standard deviation). Observed data is shown as circles ± standard deviation if available. D, day of treatment; FCZ, fluconazole; INR, international normalized ratio; po, oral; qd, once daily; VKORC1, Vitamin K epoxide reductase complex subunit 1.

Dose selection and adjustment

The dose required for a steady‐state INR value of 2 for different combinations of CYP2C9 and VKORC1 genotypes was determined. Table S5 shows a comparison of the determined dose with the label‐recommended dose. The simulated plasma concentration‐time profiles and INR values are shown in Figure 7, Figure S7. The effect of fluconazole co‐administration on days 11–24 was simulated with maximum INR values of 3.69, 3.6, 3.46, 3.53, 3.13, 3.15, 2.94, and 2.85, as shown in Figure 7a–h, most exceed the 2–3 INR range. In the fluconazole co‐administration on days 11–45 scenario, all INR values exceed the range of 2–3 as show in Figure S7a‐h. After incorporating Equation 4 into the PD model, the simulation was rerun using the recommended warfarin dose. In all groups, INR values were adjusted to levels close to those of warfarin administered alone. The adjusted dose is displayed in Figures 7 and S7. All dose‐adjustment regimens had a similarity in the ratio of warfarin increase or decrease, as shown in Figures S8 and S9.

FIGURE 7.

Simulated S‐warfarin plasma concentration–time and INR–time profiles. (a–h) In different CYP2C9 and VKORC1 genotypes. Different dosing regimens are shown as lines with color. S‐War/INR, S‐warfarin plasma concentration or INR value after warfarin administered alone. S‐War/INR with FCZ D11‐24, S‐warfarin plasma concentration or INR value with fluconazole administered on days 11–24. S‐War/INR adjusted, S‐warfarin plasma concentration or INR value after warfarin dose adjusted with fluconazole administered on days 11–24. Black dashed lines indicate the start and stop of administering fluconazole. Column and text indicate adjusted dose. CYP, cytochrome P450; FCZ, fluconazole; INR, international normalized ratio; Plasma conc, S‐warfarin plasma concentration; S‐War, S‐warfarin; VKORC1, Vitamin K epoxide reductase complex subunit 1.

DISCUSSION

A systemic PBPK/PD model of S‐warfarin was successfully developed. The plasma concentrations of S‐warfarin simulated based on the model agreed with observations from intravenous and oral clinical studies. The model was also suitable for predicting S‐warfarin plasma concentrations in different CYP2C9 genotypes and INR values in different VKORC1 genotypes. It can also be used to assess PK and PD changes when warfarin is combined with fluconazole, based on which dosage adjustments can be made.

The PBPK/PD model for the S‐enantiomer was developed in this study, mainly because the S‐enantiomer is usually considered to have higher potency, and there is some controversy about the role of R‐enantiomer. ¹³ , ⁵¹ , ⁵² , ⁵³ This study's combination of S‐warfarin plasma concentrations with an indirect response model predicted INR observations well. In the literature, S‐warfarin concentration has also been used as the only exposure predictor of the INR response. ⁵⁴ , ⁵⁵ , ⁵⁶ Therefore, using only S‐warfarin concentrations in combination with the PD model to predict INR values is reasonable.

The intestinal permeability from the literature was used to increase the rate of absorption. ⁵⁷ Using literature values improves the simulated C _max values to be closer to the measured values. The metabolism of S‐warfarin is simplified to three hydroxyl compounds through two enzymes, CYP2C9 and CYP3A4. This simplification is justified because these three metabolites have higher plasma concentrations then the others. ¹⁴ In addition, the plasma concentrations of S‐warfarin in different CYP2C9 genotypes were well‐predicted by modifying the K _cat value of the CYP2C9 enzyme. However, the K _cat of the CYP3A4 enzyme was also modified in the CYP2C9*3/*3 genotype. Further research is required to determine whether metabolism through enzymes other than CYP2C9 and CYP3A4 should be considered in this genotype.

As early as 1994, Professor Levy proposed the TMDD for warfarin and subsequently simulated blood concentration profiles in rats and humans using the PBPK model. ³⁶ , ⁵⁸ Two recently published PBPK models for warfarin consider the effect of TMDD on warfarin, and both improve the model compared to models that do not incorporate TMDD. ³⁸ , ⁵⁹ In this study, the model was greatly improved by adding the TMDD process. This suggests that TMDD is a non‐negligible influence on the PKs of warfarin. Most TMDD modeling is complex and has a range of simplified models under different assumptions. ⁶⁰ Here, we do not use a complex modeling approach to implement the TMDD process, which has the advantage of reducing the complexity of the model.

The fraction of S‐warfarin metabolism via CYP2C9 is considered to be above 85% in the literature, ²³ which is similar to the present study (86%–94%). Accurate prediction of the fraction of enzyme metabolism is essential for applying the model to DDI. The prediction of plasma concentrations of S‐warfarin in combination with fluconazole at different CYP2C9 genotypes was well‐fitted to the observations. This also confirms that the model describes the PKs of warfarin under different CYP2C9 genotypes.

In 1968, Nagashima et al. ⁶¹ developed a PK/PD model relating prothrombin complex activity to warfarin concentration. Through a series of developments, it became one of four indirect response models. ⁶² This approach of treating coagulation factors as complexes is widely used in warfarin PK/PD modeling. ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ A series of Quantitative Systems Pharmacology (QSP) models of the coagulation cascade were also created. ⁶⁷ Simplification of the QSP model is necessary to reduce complexity. The PD model used in this study is simplified from one of the QSP models. ⁴⁴ , ⁶⁸ In contrast to modeling coagulation factors as complexes, this simplified model retains the three coagulation factors and includes the inhibition of vitamin K reduction by warfarin. This mechanism‐based modeling approach is more consistent with the anticoagulation mechanism of warfarin.

PBPK modeling of warfarin has previously been applied to investigate specific scenarios, such as studying bioequivalence, ⁵⁷ determining the role of OAT2, ³⁵ predicting in vivo target occupancy, ³⁸ and DDI. ⁵⁹ , ⁶⁶ , ⁶⁹ Two studies of PBPK/PD modeling are applied to dose management when warfarin and sorafenib or nirmatrelvir/ritonavir are administered in combination. ⁶⁶ , ⁶⁹ Compared to our model, the TMDD process was incorporated, the fraction of CYP2C9 enzyme metabolism was considered, and a different PD model was used. In another PBPK article, the authors modeled PBPK and successfully recovered plasma concentrations of warfarin in combination with fluconazole under different CYP2C9 genotypes. ⁵⁹ In contrast, we added the PD model, which can be used to assess changes in INR and make dosage adjustments under combination.

The PBPK/PD model was finally applied in DDI simulations, and to give dose adjustment recommendations. When warfarin is used in combination with fluconazole, the most reasonable strategy is to adjust the warfarin dose. Reducing the dose of warfarin can balance the decrease in clearance, ensuring that warfarin plasma concentrations remain unaffected and fluconazole's antifungal effects are maintained. Although fluconazole inhibits both CYP2C9 and 3A4, f _m in Equation 4 represents the fraction of metabolism via CYP2C9. This is because the fraction of CYP3A4 metabolism is low compared to 2C9. Although the primary purpose of this model is not to give dosing recommendations for different genotypes, the dose determined is consistent with that given on the label. This also indicates that the model successfully simulates the PK/PD relationship. The effect on INR values was simulated when fluconazole was used in combination with warfarin for 2 weeks and 35 days. The effect of genotype was not considered in the original literature presenting Equation 4, ²⁴ and, here, we simulate the use of the method to adjust the dose under different genotypes. Although Equation 4 is a dosage adjustment method based on PK principles, the given dosage adjustment regimen was effective in controlling the increase of INR. Because the parameters f _m, and K _i in Equation 4 were not changed in the different scenarios and the fluconazole concentrations [I] given based on the model were equal, the dosage adjustment regimens were similar in different genotype combinations. This implies that adjustment schemes based on PK principles may be effective in different genotypes. However, clinical studies are needed to investigate this hypothesis further.

Although the presented model performed well concerning both single and multiple doses and in most CYP2C9 and VKORC1 genotype groups, it has several limitations. First, in the CYP*3/*3 genotype, the predicted values of metabolites do not match well with the observed values. Nevertheless, S‐warfarin plasma concentrations in this genotype were well‐predicted when co‐administered with fluconazole. Second, the f _m values in Equation 4 were not adjusted for different genotypes as it is population average. In addition, it is not certain that f _m would change significantly in different population. Nevertheless, the INR is reasonably adjusted in the simulation by the dose calculated in Equation 4. The third limitation was that no specific dosage adjustment ratio was given. This is because the clinical doses of fluconazole and warfarin are highly variable between individuals. However, this study provides a modeling approach that can be used to simulate different dosage regimens.

CONCLUSIONS

An S‐warfarin PBPK/PD model was successfully developed to describe the PKs of S‐warfarin after intravenous and oral administrations, along with the effect of S‐warfarin on INR value. The model adequately describes the PKs and PDs of S‐warfarin, takes into account the effects of genetic polymorphisms, and successfully predicts DDI with CYP2C9 inhibitor. The S‐warfarin PBPK/PD model can be applied in DDI predictions as a CYP2C9 victim drug to evaluate the effects of CYP2C9 inhibitors. The incorporated dosage adjustment methods are also able to give recommendations for dosage adjustment when used in conjunction with CYP2C9 inhibitors.

AUTHOR CONTRIBUTIONS

K.G. wrote the manuscript. H.T.X. and H.S. designed the research. K.G., C.Z.S., and X.H.W. performed the research. K.G., X.W.W., W.X.S., W.H.W., and T.C. analyzed the data.

FUNDING INFORMATION

This study was supported by Funding of “Climbing Peak” Training Program for Innovative Technology team of Yijishan Hospital, Wannan Medical College (KPF2019016).

CONFLICT OF INTEREST STATEMENT

The authors declared no competing interests for this work.

Supporting information

Data S1.

Data S2.

Figures S1–S9.

Tables S1–S5.

Geng K, Shen C, Wang X, et al. A physiologically‐based pharmacokinetic/pharmacodynamic modeling approach for drug–drug‐gene interaction evaluation of S‐warfarin with fluconazole. CPT Pharmacometrics Syst Pharmacol. 2024;13:853‐869. doi: 10.1002/psp4.13123