Physiologically Based Pharmacokinetic Modeling of Rosuvastatin to Predict Transporter-Mediated Drug-Drug Interactions

Purpose To build a physiologically based pharmacokinetic (PBPK) model of the clinical OATP1B1/OATP1B3/BCRP victim drug rosuvastatin for the investigation and prediction of its transporter-mediated drug-drug interactions (DDIs). Methods The Rosuvastatin model was developed using the open-source PBPK software PK-Sim®, following a middle-out approach. 42 clinical studies (dosing range 0.002–80.0 mg), providing rosuvastatin plasma, urine and feces data, positron emission tomography (PET) measurements of tissue concentrations and 7 different rosuvastatin DDI studies with rifampicin, gemfibrozil and probenecid as the perpetrator drugs, were included to build and qualify the model. Results The carefully developed and thoroughly evaluated model adequately describes the analyzed clinical data, including blood, liver, feces and urine measurements. The processes implemented to describe the rosuvastatin pharmacokinetics and DDIs are active uptake by OATP2B1, OATP1B1/OATP1B3 and OAT3, active efflux by BCRP and Pgp, metabolism by CYP2C9 and passive glomerular filtration. The available clinical rifampicin, gemfibrozil and probenecid DDI studies were modeled using in vitro inhibition constants without adjustments. The good prediction of DDIs was demonstrated by simulated rosuvastatin plasma profiles, DDI AUClast ratios (AUClast during DDI/AUClast without co-administration) and DDI Cmax ratios (Cmax during DDI/Cmax without co-administration), with all simulated DDI ratios within 1.6-fold of the observed values. Conclusions A whole-body PBPK model of rosuvastatin was built and qualified for the prediction of rosuvastatin pharmacokinetics and transporter-mediated DDIs. The model is freely available in the Open Systems Pharmacology model repository, to support future investigations of rosuvastatin pharmacokinetics, rosuvastatin therapy and DDI studies during model-informed drug discovery and development (MID3). Supplementary Information The online version contains supplementary material available at 10.1007/s11095-021-03109-6.


.1 PBPK model building
PBPK model building was started with an extensive literature search to collect physicochemical parameters, information on absorption, distribution, metabolism and excretion (ADME) processes and clinical studies of intravenous and oral administration in single-and multiple-dose regimens. In addition to drug plasma concentration-time profiles, observed data on fraction excreted in urine or feces and tissue concentrations were integrated. The data of the clinical studies was digitized and divided into a training dataset for model building and a test dataset for model evaluation. The studies for the training dataset were selected to include intravenous and oral studies covering the whole published dosing range. If multiple studies of the same dose were available, studies with many participants, modern bioanalytical methods and frequent as well as late sampling were chosen for the training dataset. Model input parameters that could not be informed from literature were optimized by fitting the model simulations of all studies assigned to the training dataset simultaneously to their respective observed data.

Virtual individuals
Virtual mean individuals were generated for each study according to the published demographic information with corresponding age, weight, height, sex and ethnicity. If no information was provided, a default value was substituted (30 years of age, male, European, mean weight and height characteristics from the PK-Sim ® population database). Enzymes and transporters relevant to the pharmacokinetics of rosuvastatin were incorporated in agreement with current literature, utilizing the PK-Sim ® expression database [1] to define their relative expression in the different organs of the body. Details and references on the distribution and localization of the implemented metabolizing enzymes and drug transporters are provided in Table S7.0.1.

PBPK model evaluation
Model performance was evaluated with multiple methods. First, predicted plasma concentration-time profiles were compared visually with the data observed in the respective clinical studies. Second, the predicted plasma concentration values of all studies were plotted against their corresponding observed values in goodness-of-fit plots. In addition, model performance was evaluated by comparison of predicted to observed values of area under the plasma concentration-time curve from the time of drug administration to the last concentration measurement (AUC last ) and peak plasma concentrations (C max ).
As quantitative measures of the model performance, the mean relative deviation (MRD) of all predicted plasma concentrations (Equation S1) and the geometric mean fold error (GMFE) of all predicted AUC last and C max values (Equation S2) were calculated. MRD and GMFE values ≤ 2 characterize an adequate model performance.
(log 10 c predicted,i − log 10 c observed,i ) 2 (S1) where c predicted,i = predicted plasma concentration, c observed,i = corresponding observed plasma concentration, k = number of observed values.
GM F E = 10 x with x = 1 m m i=1 log 10 predicted PK parameter i observed PK parameter i where predicted PK parameter i = predicted AUC last or C max value, observed PK parameter i = corresponding observed AUC last or C max value, m = number of studies.
Furthermore, the physiological plausibility of the parameter estimates and the results of a sensitivity analysis were assessed.

PBPK model sensitivity analysis
Sensitivity of the final model to single parameters (local sensitivity analysis) was calculated as relative change of AUC 0-24 using the Sensitivity Analysis tool implemented in PK-Sim ® [2]. Sensitivity analysis was performed applying a relative perturbation of 1000 % (variation range 10.0, maximum number of 9 steps). Parameters were included into the analysis if they were optimized, if they are associated with optimized parameters or if they might have a strong impact due to calculation methods used in the model.
Sensitivity to a parameter was calculated as the ratio of the relative change of the simulated AUC to the relative variation of the parameter around its value used in the final model according to Equation S3.
where S = sensitivity of the simulated AUC 0-24 to the examined model parameter value, ∆AU C = change of the simulated AUC 0-24 , AU C = simulated AUC 0-24 with the original parameter value, ∆p = change of the examined parameter value, p = original parameter value. A sensitivity of 0.5 signifies that a 100 % change of the examined parameter value causes a 50 % change of the simulated AUC 0-24 .

Competitive inhibition
Competitive inhibitors reversibly bind to the active site of an enzyme or transporter and compete with the substrate for binding. Competitive inhibition can be overcome by high substrate concentrations (concentration-dependency); therefore, the maximum reaction velocity (v max ) remains unaffected, while the Michaelis-Menten constant (K m ) is increased by the inhibition (K m,app , Equation S4). The reaction velocity (v) during co-administration of substrate and competitive inhibitor is described by Equation S5 [2]: where K m,app = Michaelis-Menten constant in the presence of the inhibitor, K m = Michaelis-Menten constant, [I] = free inhibitor concentration, K i = dissociation constant of the inhibitor-enzyme or the inhibitor-transporter complex, v = reaction velocity, v max = maximum reaction velocity, [S] = free substrate concentration.

Rosuvastatin population pharmacokinetic (PopPK) analysis 2.1 Background
Typical rosuvastatin plasma concentration-time profiles show an unusual shape with a slow absorption phase and late C max (t = 5.0 h). This delayed absorption has been described previously [3], but a mechanistic explanation could not be found in the literature. Therefore, rosuvastatin PBPK model building was supported by a population pharmacokinetic (PopPK) analysis to investigate and improve the description of the slow rosuvastatin absorption.

Objectives
The first objective of this analysis was to develop a PopPK model of rosuvastatin based on the digitized mean data from the only published intravenous study of rosuvastatin [4], and individual data from two oral studies [5,6]. The model should focus on the description of the absorption phase and late C max after oral rosuvastatin administration and support the PBPK model development.
In addition, during the DDIs with rifampicin and probenecid, a much faster rosuvastatin absorption and earlier C max (t = 1.5 h) were observed, but not during the DDI with gemfibrozil. The second objective of this analysis was to extend the developed rosuvastatin PopPK model to analyze and describe the differences in absorption, bioavailability and clearance of rosuvastatin during the DDIs with rifampicin, probenecid and gemfibrozil, adding data of three different DDI studies to the dataset [7,7,8].

Dataset
For rosuvastatin PopPK model development, the mean data from the only published intravenous study of rosuvastatin [4] and individual rosuvastatin plasma concentration-time profiles from two oral rosuvastatin studies [5,6], were used. For the DDI analysis, individual rosuvastatin plasma profiles before and during administration of rifampicin [7], individual rosuvastatin plasma profiles before and during administration of probenecid [7] and mean data of the only published study of the gemfibrozil-rosuvastatin DDI [8] were added to the dataset, see Table S2.3.1.

Model building and evaluation
Population pharmacokinetic analysis was performed using non-linear mixed-effects modeling techniques implemented in NONMEM (version 7.4.3). These allow estimation of population medians for pharmacokinetic model parameters with simultaneous quantification of interindividual variability (IIV). Model selection was based on the objective function value (OFV) provided by NONMEM, visual inspection of goodness-of-fit plots and the precision of parameter estimates. A nested model was considered superior to another when the OFV was reduced by 3.84 units (χ 2 -test statistic, p < 0.05, 1 degree of freedom).
The First-Order Conditional Estimation with Interaction (FOCE-I) method was applied and models were coded in the ADVAN6 subroutine. For the structural base model one-, two-and threecompartment models were tested with first-order and saturated elimination (Michaelis-Menten) kinetics. Subsequently, different absorption models, such as zero-order, first-order and mixed parallel zero-and first-order absorption processes as well as split doses were evaluated. Saturable processes on absorption rates were tested using Michaelis-Menten kinetics. Based on the structural base model, IIVs were modeled exponentially and evaluated univariately. IIVs were added to the model if they improved the model in a statistically significant manner and if the parameter estimates of the model remained stable.
After the rosuvastatin model was established, the DDI profiles were added to the dataset and all data were modeled together using the same model with the DDI effects implemented via covariates on the model parameters of the initial model. For the DDI arms, covariate factors on the bioavailability and clearance were needed to account for the effects during the different DDIs. In addition, it was tested whether the absorption was influenced by the DDI.

Rosuvastatin PopPK model
The pharmacokinetics of rosuvastatin were best described by a two-compartment model with firstorder elimination (CL) from the central compartment. To describe the absorption phase and shape of the rosuvastatin plasma concentration-time profiles appropriately, the total rosuvastatin dose was split into a first (Dose 1) and a second dose (Dose 2), where the fraction of the second dose (VF2) was estimated and the fraction of the first dose was calculated as (1-VF2). Both doses were absorbed with the same absorption rate constant (Ka) and the same bioavailability, but the absorption of the second dose was delayed by a lag time (ALAG2). A schematic representation of the model is illustrated in Figure S2.4.1.
Parameter estimates of the final model are presented in Table S2  The final rosuvastatin PopPK model was then applied to investigate the rosuvastatin absorption phase during the different DDIs. Adding the data of the DDI studies to the PopPK dataset and using covariate factors on the bioavailability and clearance, the effects of the different DDIs could be well described. For the rifampicin and the probenecid DDIs, the second absorption process was no longer needed and a single absorption compartment described the data best. For gemfibrozil, two absorption compartments were still necessary. Therefore, only the rosuvastatin administration protocols for rosuvastatin monotherapy and during gemfibrozil co-administration were split as described above; during rifampicin and probenecid co-treatment the total rosuvastatin dose was released immediately. The parameter estimates between the model without and with DDI were comparable (Table S2.4.1).     [7] before and during rifampicin co-administration and the individuals 3301 -3306 of the study by Wiebe et al. 2020 [7] before and during probenecid co-administration   (Table S3.2.1). The results of the PopPK analysis (absorption phase best described by a split dose approach with lag time for the second dose) were integrated into the PBPK simulations, using the estimated PopPK median split dose parameters in all oral rosuvastatin administration protocols in PK-Sim ® , which greatly improved the results of the PBPK parameter identification. The final model applies active rosuvastatin transport by OATP2B1, OATP1B1/1B3, OAT3, Pgp and BCRP, as well as metabolism by CYP2C9 (Table S3.3.1). Details on the implementation of these drug transporters and metabolic enzymes in the different organs are provided in the system-dependent parameter table (Table S7.0.1)).
The good model performance is demonstrated in semilogarithmic ( Figure S3 Table S3.5.1. The correlation of predicted to observed AUC last and C max values is shown in Figure S3

Rosuvastatin clinical studies
The clinical studies used for rosuvastatin model development and evaluation are summarized in Table S3.2.1.

Rosuvastatin drug-dependent parameters
The drug-dependent parameters of the final rosuvastatin model are summarized in Table S3.3.1. The associated system-dependent parameters are listed in Table S7.0.1.

Sensitivity analysis
Sensitivity of the rosuvastatin model to single parameter values (local sensitivity analysis) was calculated as the relative change of the predicted AUC 0-24 ( Figure S3.5.3) of a 40 mg single dose of rosuvastatin administered as tablet in the fasted state (highest recommended dose). Sensitivity analysis was carried out using a relative parameter perturbation of 1000 % (variation range 10.0, maximum number of 9 steps). Parameters were included into the analysis if they were optimized (luminal intestinal permeability, basolateral intestinal permeability, OATP2B1 kcat, OATP1B1/1B3 kcat, OAT3 kcat, Pgp kcat, BCRP kcat, CYP2C9 CLspec), if they are associated with optimized parameters (OATP2B1 Km, OATP1B1/1B3 Km, OAT3 Km, Pgp Km, BCRP Km), or if they might have a strong impact due to calculation methods used in the model (solubility, lipophilicity, fraction unbound, blood/plasma concentration ratio, GFR fraction).

DDI modeling
The rifampicin-rosuvastatin DDI was modeled using a previously established whole-body PBPK model of rifampicin [47]. The drug-dependent parameters of this model are reproduced in Table S4.2.1.
The rifampicin-rosuvastatin interaction was simulated as competitive inhibition of OATP2B1, Pgp, BCRP, OATP1B1/1B3 and CYP2C9 by rifampicin. The parameters to model these inhibitions were obtained from literature [11,[48][49][50] or in-house measurements (rifampicin Pgp IC 50 ), and were included in the rifampicin drug-dependent parameter Table S4.2.1. To account for the impact of rifampicin on the absorption of rosuvastatin, the rosuvastatin dose during the rifampicin-rosuvastatin DDI was modeled as a single dose without a lag time, as indicated by the PopPK analysis.
Details on the predicted clinical DDI studies are given in Table S4.3.1. Model predictions of rosuvastatin plasma concentration-time profiles before and during rifampicin co-administration, compared to observed data, are shown in Figures S4.4.1 and S4.4.2. Predicted compared to observed rosuvastatin fraction excreted in urine before and during rifampicin co-administration are shown in Figure S4

Rifampicin-rosuvastatin clinical DDI studies
The clinical studies used to evaluate the rifampicin-rosuvastatin DDI model performance are summarized in Table S4.3.1.

DDI modeling
The gemfibrozil-rosuvastatin DDI was modeled using a previously established whole-body parentmetabolite PBPK model of gemfibrozil with its metabolite gemfibrozil 1-O-β-glucuronide [64]. The drug-dependent parameters of this model are reproduced in Table S5.2.1.
The gemfibrozil-rosuvastatin interaction was simulated as competitive inhibition of OATP1B1/1B3, OAT3 and CYP2C9 by gemfibrozil and of OATP1B1/1B3 and OAT3 by gemfibrozil glucuronide. The parameters to model these inhibitions were obtained from literature [65][66][67], and were included in the gemfibrozil and gemfibrozil 1-O-β-glucuronide drug-dependent parameter Table S5.2.1. The rosuvastatin dose during the gemfibrozil-rosuvastatin DDI was modeled using the split dose approach to describe the still slow absorption of rosuvastatin during this DDI, as indicated by the clinically observed data and the PopPK analysis.
Details on the predicted clinical DDI study are given in Table S5.3.1. Model predictions of rosuvastatin plasma concentration-time profiles before and during gemfibrozil co-administration, compared to observed data, are shown in Figure S5.4.1. Gemfibrozil was administered twice daily for seven days; rosuvastatin was administered as a single dose together with the gemfibrozil morning dose on day 4 (72 h). The strong competitive inhibition of rosuvastatin uptake from the blood into liver and kidney via OATP1B1/1B3 and OAT3 leads to short increases of the simulated rosuvastatin plasma concentrations after every further administration of gemfibrozil (84 h, 96 h, 108 h). The correlation of predicted to observed DDI AUC last ratios and DDI C max ratios is shown in Figure S5.5.1. Table S5.5.1 lists the corresponding predicted and observed DDI AUC last ratios, DDI C max ratios, as well as GMFE values.

Gemfibrozil and gemfibrozil 1-O-β-glucuronide drug-dependent parameters
The drug-dependent parameters of the gemfibrozil parent-metabolite model are summarized in Table S5.2.1. The associated system-dependent parameters are listed in Table S7.0.1.

Gemfibrozil-rosuvastatin clinical DDI studies
The clinical studies used to evaluate the gemfibrozil-rosuvastatin DDI model performance are summarized in Table S5.3.1. 6 Probenecid-rosuvastatin drug-drug interaction (DDI) 6

.1 DDI modeling
The probenecid-rosuvastatin DDI was modeled using a previously established whole-body PBPK model of probenecid [80]. The drug-dependent parameters of this model are reproduced in Table S6.2.1.
The probenecid-rosuvastatin interaction was simulated as competitive inhibition of OATP1B1/1B3 and OAT3 by probenecid. The parameters to model these inhibitions were obtained from literature [81,82], and were included in the probenecid drug-dependent parameter Table S6.2.1. To account for the impact of probenecid on the absorption of rosuvastatin, the rosuvastatin dose during the probenecid-rosuvastatin DDI was modeled as a single dose without a lag time, as indicated by the PopPK analysis.
Details on the predicted clinical DDI study are given in Table S6.3.1. Model predictions of rosuvastatin plasma concentration-time profiles before and during probenecid co-administration, compared to observed data, are shown in Figure S6.4.1. Predicted compared to observed rosuvastatin fraction excreted in urine before and during probenecid co-administration are shown in Figure S6 Table S6.5.1 lists the corresponding predicted and observed DDI AUC last ratios, DDI C max ratios, as well as GMFE values.

Probenecid drug-dependent parameters
The drug-dependent parameters of the probenecid model are summarized in Table S6.2.1. The associated system-dependent parameters are listed in Table S7.0.1.

Probenecid-rosuvastatin clinical DDI studies
The clinical studies used to evaluate the probenecid-rosuvastatin DDI model performance are summarized in Table S6.3.1.   [63]. Details on administration protocols, study population and literature reference are listed in Table S6.3.1