Physiologically‐based pharmacokinetic modeling predicts the drug interaction potential of GLS4 in co‐administered with ritonavir

Abstract

GLS4 is a first‐in‐class hepatitis B virus (HBV) capsid assembly modulator (class I) that is co‐administered with ritonavir to maintain the anticipated concentration required for the effective antiviral activity of GLS4. In this study, the first physiologically‐based pharmacokinetic (PBPK) model for GLS4/ritonavir was successfully developed. The predictive performance of the PBPK model was verified using data from 39 clinical studies, including single‐dose, multiple‐dose, food effects, and drug–drug interactions (DDI). The PBPK model accurately described the PK profiles of GLS4 and ritonavir, with predicted values closely aligning with observed data. Based on the verified GLS4/ritonavir model, it prospectively predicts the effect of hepatic impairment (HI) and DDI on its pharmacokinetics (PK). Notably, CYP3A4 inducers significantly influenced GLS4 exposure when co‐administered with ritonavir; co‐administered GLS4 and ritonavir significantly influenced the exposure of CYP3A4 substrates. Additionally, with the severity of HI increased, there was a corresponding increase in the exposure to GLS4 when co‐administered with ritonavir. The GLS4/ritonavir PBPK model can potentially be used as an alternative to clinical studies or guide the design of clinical trial protocols.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

GLS4 is a CYP3A4 inducer that primarily maintains its effective concentration through co‐administration with ritonavir, an inhibitor of CYP3A4.

• WHAT QUESTION DID THIS STUDY ADDRESS?

The study explored whether physiologically‐based pharmacokinetic (PBPK) models could predict the pharmacokinetics (PK) of GLS4 and its PK when co‐administered with ritonavir. It also explored the prediction of GLS4's PK effect scenarios of drug–drug interactions (DDI) not yet clinically tested and in populations with hepatic impairment (HI), based on the developed PBPK model.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

The PBPK model‐predicted changes in GLS4 exposure when GLS4/ritonavir is administered in the presence of CYP3A4 modulators or populations with HI, as well as the effects of GLS4/ritonavir on CYP3A4 substrates.

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

Serving as an alternative to clinical studies, the PBPK model can be a valuable tool for assessing the DDI potential of GLS4/ritonavir, guiding dosage adjustments in special populations, and supporting regulatory submissions.

INTRODUCTION

Hepatitis B virus (HBV) infection is a substantial global health threat. ¹ Over 296 million people were chronically infected with HBV, with over one‐third of the world's HBV infections and deaths due to HBV occurring in China, and about 10 million Chinese with chronic HBV are estimated to die in 2030. ² , ³ In China, Sunshine Lake Pharma Co., Ltd has made significant strides in combating this issue by developing GLS4, an HBV capsid assembly modulator designed to disrupt the assembly of the HBV nucleocapsid. ⁴ , ⁵ , ⁶ GLS4 has successfully advanced to phase III clinical trials. It is worth noting that GLS4 is a sensitive Cytochrome P450 (CYP) 3A substrate, whereas it has an inducible effect on the CYP3A4 enzyme. ⁷ , ⁸ The results of clinical trials demonstrated that the anticipated concentration necessary for effective antiviral activity cannot be attained through the exclusive administration of GLS4. When ritonavir is utilized as a booster in co‐administration with GLS4, it can facilitate the attainment of an effective trough concentration. ⁵ Therefore, GLS4 will be marketed as a co‐packaged product with ritonavir, akin to Paxlovid.

However, unlike Paxlovid, patients infected with HBV require long‐term administration of GLS4/ritonavir. GLS4, as a CYP3A4 inducer, and ritonavir, a CYP3A inhibitor, when combined as a co‐packaged product, may potentially affect other CYP3A substrate drugs, resulting in a risk of serious drug–drug interactions (DDI). Due to the necessity of long‐term medication, DDI is unavoidable, and the DDI resulting from the concurrent use of co‐administered GLS4 with ritonavir is presently poorly understood, and this data must be clearly defined before it goes to the market. Similarly, additional data is required to understand the effect of hepatic impairment (HI) on the pharmacokinetics (PK) of GLS4 in co‐administration with ritonavir.

In recent years, modeling and simulation (M&S) have accelerated novel drug discovery and development in the pharmaceutical industry. Among these approaches, physiologically‐based pharmacokinetic (PBPK) modeling is a methodology capable of harmonizing physiological and drug preclinical and clinical information to simulate the processes of drug absorption, distribution, metabolism, and excretion (ADME). PBPK is a tool for anticipating unknown clinical scenarios, with a growing trend in its applications in novel drug regulatory submissions. Major applications include simulating DDI, predicting PK in populations affected by organ impairments, forecasting PK within pediatrics, and more. This study developed a PBPK model for GLS4 in co‐administration with ritonavir based on in silico‐derived physicochemical properties, in vitro, and clinical study data. The developed PBPK model is intended to predict the DDI potential of GLS4 in co‐administration with ritonavir and the expected exposure in populations with HI and conducted dose adjustment recommendations during clinical drug administration if necessary.

METHODS

Software

The development and verification of the PBPK model were executed using the open‐source platform PK‐Sim® (version 11.0, part of the Open‐Systems‐Pharmacology suite). Published clinical trial data were digitized using GetData Graph Digitizer (version 2.26.0.20©, S. Fedorov). Plot figures to visualize data were using R (version 4.1.2, R Foundation for Statistical Computing, Vienna, Austria) and Rstudio (version 2023.09.0‐463, Posit, Inc., Boston, USA).

Virtual populations

Virtual East Asian and European healthy males with reverse transcription polymerase chain reaction (RT‐PCR) expression of cytochrome P450 3A4 (CYP3A4), CYP3A5, CYP2D6, and P‐glycoprotein (P‐gp) created by PK‐Sim were using for the development and verification of PBPK model of GLS4 and ritonavir. To better fit the PK profiles of ritonavir, the activities of CYP3A, CYP2D6, and P‐gp in the virtual population infected with human immunodeficiency virus (HIV) were adjusted to 0.49, 0.77, and 0.79‐fold its original values, respectively. ⁹

The severity of HI in the virtual population was judged by the Child–Pugh (CP) scoring system. ¹⁰ The values of different parameters, including organ blood flow, liver organ volume, glomerular filtration rate, hematocrit, and plasma proteins, have been adjusted for the healthy population based on the reported proportional changes in values for the cirrhotic population in the literature. ¹¹ , ¹² , ¹³ , ¹⁴ Information on the demographic data and administration protocols of all clinical studies can be found in Table S1, and Table S2 summarizes the physiological changes in healthy and HI virtual populations in the PBPK model.

PBPK model development

The overall of the PBPK model development process consisted of development, verification, and application (Figure 1). The PBPK model for GLS4 was developed to gather parameters from in silico and preclinical studies of the physicochemical properties of GLS4, as well as its absorption, distribution, metabolism, and excretion (ADME) processes. Part of the parameters for GLS4 were optimized using internal data sets from clinical studies on the PK of 120 mg GLS4 (single‐dose, multiple‐dose administration, and food effect).

FIGURE 1.

The overall PBPK model development, verification, and application process flow chart.

The intestinal permeability of GLS4 was based on in vitro permeability studies (Caco‐2, A‐B) and further optimized according to clinical observational PK data. The dissolution curve of GLS4 at pH = 6.8 was fitted with a Weibull cumulative distribution function to obtain the dissolution parameters “Dissolution time (50%)” and “Dissolution shape,” describing the fasting absorption process of GLS4. “Dissolution time (50%)” and “Dissolution shape” were fitted to postprandial clinical PK data to better describe the fed absorption process. Lipophilicity was optimized within the range obtained in silico based on measured values to find the best match for the observed clinical PK data of GLS4. The standard PK‐Sim 3‐compartment model (intracellular, interstitial, and vascular space) with a single permeation barrier between the interstitial and intracellular space in each organ was used to characterize the distribution of GLS4. The tissue‐to‐plasma partition coefficient of GLS4 was estimated using the Schmitt method. GLS4 is a sensitive substrate of CYP3A, with over 80% of native GLS4 cleared by its mediated metabolism, and GLS4 induces the activity of CYP3A4. The K _m, V _max, E _max, and EC₅₀ values for GLS4 were obtained through recombinant enzyme and induction experiments, and the k _cat was further optimized based on clinical continuous administration PK data to better describe the PK of GLS4 for both single and continuous dosing.

Parameters of ritonavir related to the physicochemical properties and processes of ADME were obtained through an extensive literature search, and the PBPK model for ritonavir was developed with slight revisions based on published PBPK models. ¹⁵ , ¹⁶ The physicochemical and PK parameters of GLS4 and ritonavir implemented in the PBPK model are summarized in Table S3.

PBPK model verification

The PBPK models were verified by comparing predicted plasma concentration–time profiles with observed data. A visual assessment was conducted between the predicted and observed concentration–time profiles, as well as through goodness‐of‐fit plots for the predicted versus observed data. The model performance was quantitatively verified by calculating the predicted/observed ratio (R _pre/obs) and the geometric mean fold errors (GMFE) of the predicted and observed PK parameters (AUC, C _max). The GMFE within a twofold range indicates a successful model prediction.

First, the predictive performance of the PBPK models for GLS4 and ritonavir should be verified separately. After successfully verifying the PBPK models for both drugs, the prediction performance of GLS4 in combination with ritonavir was further verified.

PBPK model application

The PBPK model was applied to predict DDI and HI‐induced changes in exposures of GLS4 after multiple once‐daily doses of GLS4/ritonavir (120 mg/100 mg) for 9 days. Each prediction consisted of 100 virtual Asian standard subjects (50% women; 19–40 years of age) each in the following groups: (1) Healthy control (healthy subjects); (2) Predicting the effect of CYP3A inhibitors on GLS4 (healthy subjects); (3) Predicting the effect of CYP3A inducers on GLS4 (healthy subjects); (4) Predicting the effect of GLS4 versus ritonavir on CYP3A4 substrates (healthy subjects); (5) Predicting the effect of mild HI on GLS4 (CP‐A subjects); (6) Predicting the effect of moderate HI on GLS4 (CP‐B subjects); and (7) Predicting the effect of severe HI on GLS4 (CP‐C subjects). The model of rifampicin ¹⁷ (a strong CYP3A4 inducer), efavirenz ¹⁸ (a moderate CYP3A4 inducer), clarithromycin ¹⁹ (a strong CYP3A inhibitor), and midazolam ²⁰ (a CYP3A4 substrate) utilized in the PBPK model application were all sourced from the OSP‐PBPK‐Model‐Library. The corresponding qualification reports for these models can be accessed on the Open‐Systems‐Pharmacology (accessible via: github.com/Open‐Systems‐Pharmacology/OSP‐Qualification‐Reports).

RESULTS

PBPK model verification

For the development and verification of the PBPK model, PK of 19 GLS4 studies ⁵ (oral administration) and 20 ritonavir studies ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ (oral administration) information were used. Through verification, the PBPK models of GLS4 and ritonavir demonstrated good descriptive (internal data set) and predictive (external data set) performance.

The verification results for the concentration–time profiles and correlation of PK parameters are shown in Figures 2, 3, 4. Within the dose range of 2.5–340 mg, except for the single‐dose groups of 2.5 and 7.5 mg, the ratio of predicted to observed values for AUC and C _max for other dose groups was within a twofold range, with GMFE values for the 2.5–340 mg dose range being 1.34 and 1.30, respectively (Table S4). By comparing the predicted values of the GLS4 PBPK model with observed concentration–time curves and PK parameters, this study demonstrated that the established GLS4 PBPK model showed good predictive performance in predicting oral doses of 15–340 mg GLS4 across various dosing regimens (single dose, multiple doses, fasting, and postprandial), as depicted in Figure 2. The PBPK model for ritonavir showed good predictive performance for both healthy populations and HIV patients receiving oral doses of 100–500 mg of ritonavir, with the ratio of predicted to observed values for AUC and C _max within a twofold range. The GMFE values for the oral dose range of 100–500 mg of ritonavir were 1.17 and 1.18, respectively (Table S5); moreover, the ritonavir PBPK model also demonstrated good predictive performance in predicting DDI when ritonavir was co‐administered with different CYP3A4 and P‐gp substrates, with the ratio of predicted to observed values of AUC and C _max within a twofold range, and GMFE values of 1.24 and 1.18, respectively (Table S6). Following the successful validation of the predictive performance of the PBPK models for GLS4 and ritonavir, the PK predictions for the co‐administration of GLS4 with ritonavir were also validated, showing that the PBPK model exhibited good predictive performance, especially at steady state (Figure 3, Table S4). The PBPK model effectively described GLS4's auto‐induction effect (Figure 2) and ritonavir's inhibitory effect on GLS4 (Figure 3).

FIGURE 2.

Comparison between the observed and predicted pharmacokinetic profiles of GLS4. SD, single dose. The model predictions are shown as red solid lines and the corresponding observed data as black dots (arithmetic mean).

FIGURE 3.

Comparison between the observed and predicted pharmacokinetic profiles of GLS4 in co‐administration with ritonavir. The model predictions are shown as red solid lines, the 95% prediction intervals as a blue region, and the corresponding observed data as black dots (arithmetic mean).

FIGURE 4.

Correlation of predicted and observed GLS4 AUC and C _max values of all studies. SD, single dose. The solid line marks the line of identity. SD, Single Dose; RTV, ritonavir. The dotted lines indicate 0.8‐ to 1.25‐fold, and the dashed lines indicate 0.5‐ to 2‐fold acceptance limits.

PBPK model application

The predictive results concerning the effects of strong CYP3A4 modulators and HI on the PK of GLS4, as well as the DDI between GLS4/ritonavir co‐administered with CYP3A4 substrates, were detailed in Tables 1, 2, 3.

TABLE 1.

The model‐predicted arithmetic means of C _max and AUC values for GLS4 after once‐daily doses of GLS4/ritonavir (120 mg/100 mg) co‐administered with Perpetrators for 9 days in healthy subjects.

PK Parm	Perpetrators
	–	Rifampicin a	Ratio	Efavirenz b	Ratio	Clarithromycin c	Ratio
AUC0–24 (ng/mL h)	8079.3	806.9	0.10	2382.4	0.30	11,485.8	1.42
Cmax (ng/mL)	1348.7	392.1	0.29	631.7	0.47	1776.2	1.31

^a 600 mg, PO once daily for 9 days.

^b 400 mg, PO once daily for 9 days.

^c 500 mg, PO every 12 h for 9 days.

TABLE 2.

The model‐predicted arithmetic means of C _max and AUC values for midazolam co‐administered with GLS4/ritonavir (120 mg/100 mg) or ritonavir (100 mg) alone in healthy subjects.

PK Parm	Perpetrators
	–	RTV a	Ratio	RTV + GLS4 b	Ratio
AUC0–24 (ng/mL h)	23.3	300.2	12.9	259.8	11.2
C max (ng/mL)	8.70	26.72	3.07	23.85	2.74

^a 100 mg, PO once daily for 9 days.

^b 120 mg/100 mg, PO once daily for 9 days; RTV, ritonavir; midazolam (2 mg), PO (single dose) on day 9.

TABLE 3.

The model‐predicted arithmetic means of C _max and AUC values for GLS4 after once‐daily doses of GLS4/ritonavir (120 mg/100 mg) for 9 days in subjects with hepatic impairment compared with matched healthy controls.

PK Parm	Virtual subjects
	Normal	CP‐A	Ratio	CP‐B	Ratio	CP‐C	Ratio
AUC0–24 (ng/mL h)	8079.3	10,745.9	1.33	14,212.6	1.76	17,720.7	2.19
C max (ng/mL)	1348.7	1883.3	1.40	1570.9	1.16	1258.6	0.93

Abbreviation: CP, Child–Pugh.

PBPK predictions indicated a significant effect of CYP3A4 inducers on exposure to GLS4 co‐administered with ritonavir. Specifically, as shown in Table 1, rifampicin reduced the AUC and C _max of GLS4 to 0.10‐ and 0.29‐fold, respectively, and efavirenz reduced the AUC and C _max of GLS4 to 0.30‐ and 0.47‐fold, respectively. Clarithromycin increased the AUC and C _max of GLS4 to 1.42‐ and 1.25‐fold, respectively, which were within a safe concentration. Additionally, the co‐administration of GLS4 and ritonavir exhibited a significant effect on CYP3A4 substrates, with the AUC and C _max of midazolam increasing to 11.2‐ and 2.74‐fold, respectively. In contrast, the administration of ritonavir alone increased to 12.9‐ and 3.07‐fold, respectively (Table 2).

As the severity of HI increased, there was a corresponding increase in the exposure to GLS4 co‐administered with ritonavir. Compared with healthy controls with normal hepatic function, subjects with mild, moderate, and severe HI exhibited C _max of GLS4, all within a 1.5‐fold range, while the AUC of GLS4 increased by factors of 1.33‐, 1.76‐, and 2.19‐fold, respectively (Table 3).

DISCUSSION

The current study successfully developed a PBPK model for GLS4/ritonavir based on the physicochemical properties, metabolism, and interaction parameters of GLS4 from preclinical research and ritonavir parameters search from multiple databases. The GMFE between the model‐predicted values and actual observed values from clinical studies for GLS4 and ritonavir were all within a twofold range, demonstrating a robust predictive performance of the PBPK model across different dosing regimens as well as DDI.

GLS4 is mainly metabolized by CYP3A4, while it is also a CYP3A4 inducer. Therefore, continuous administration of GLS4 alone is insufficient to reach therapeutic concentrations. ritonavir, owing to its inhibitory effect on CYP3A4, serves as a booster co‐administered with GLS4, facilitating the maintenance of GLS4 exposure at therapeutic concentrations. The metabolic enzyme CYP3A4 is primarily located in the hepatic and intestine, playing a crucial role in human drug metabolism. ²⁹ Given the complex interactions between GLS4 and ritonavir on CYP3A4, there are potential risks of severe DDI. The inductive performance of the GLS4 PBPK model has been verified through multi‐dose observed data (Figure 2, Table S4), while the inhibitory performance of ritonavir was verified using data from reported clinical studies of DDIs between ritonavir and CYP3A4 substrates (Table S6). The PBPK models for GLS4 and ritonavir showed reliable inductive and inhibitory performance toward the CYP3A4, respectively. The PBPK simulations of GLS4/ritonavir co‐administered predicted values were also consistent with the observed data (Figure 3, Table S4). According to clinical study data, when CYP3A4 was inhibited by ritonavir, the metabolite M2 generated by metabolism through the secondary metabolic pathway increased, but its steady‐state AUC was less than 3.5% of GLS4, suggesting that secondary metabolism was limited. And none of the metabolites of GLS4 exhibit pharmacological activity. ⁵ Therefore, the secondary metabolic pathway of GLS4 was not included in the PBPK model. Based on the prediction results and the mechanisms of induction and inhibition, GLS4 co‐administered with ritonavir still had a specific inducing effect on metabolic enzymes. However, ritonavir's inhibitory effect is stronger than GLS4's inducing effect, leading to overall inhibition when both drugs are co‐administered, although this inhibition is less than when ritonavir was taken alone (Figure S1). When GLS4 was co‐administered with ritonavir at a steady state, the enzyme activity stabilized in a low activity range. Meanwhile, clinical studies have shown that the exposure of GLS4 in combination with ritonavir after 28 days of GSL4 administration is comparable to the exposure of GLS4 after 9 days of combining the two drugs. ⁶ Thus, the strategy of GLS4 in co‐administered with ritonavir was successful, and ritonavir effectively limited the CYP3A4‐induced effects of GLS4.

Based the developed PBPK model, we conducted a prospective prediction of the effects of DDI and HI on the PK of GLS4. Ritonavir, a time‐dependent inhibitor (TDI) of CYP3A4, forms stable complexes with the enzyme, cumulatively inhibiting its activity over time. On the other hand, CYP3A4 inducers like rifampin, efavirenz, and GLS4 increase the synthesis and decrease the degradation rate of enzyme, enhancing its expression. The activity of the enzyme, when ritonavir is used with an inducer, depends on the strength of both. Clinical studies have shown that strong inducers can reverse the inhibitory effect of ritonavir, significantly reducing the exposure of CYP3A4 substrates. As PBPK predictions, CYP3A4 inducers significantly affected the exposure of GLS4 (Table 1). The co‐administration of GLS4 and ritonavir mitigated the inhibitory effects of ritonavir on CYP3A4 (Table 2, Figure S1), yet it still significantly affected the exposure of CYP3A4 substrates. Moreover, even with ritonavir's significant inhibition of CYP3A4, the addition of clarithromycin can further increase the exposure of GLS4 (Table 1). According to PBPK model predictions of enzyme activity, the steady‐state liver CYP3A4 activity after co‐administration of three drugs (GLS4, ritonavir, and clarithromycin) is between 3.2% and 8.2% of normal, which is lower than the CYP3A4 activity after GLS4 co‐administered ritonavir (1.5%–12.5%) (Figure S2). HI had a significantly increasing AUC of GLS4, correlating with the severity of HI. Moreover, the effect of HI on the C _max of GLS4 remained within 1.5‐fold across all CP classes, displaying no apparent correlation with the severity of HI. Although the above‐predicted data are not supported by relevant clinical data validation, the greatest significance stage of using the PBPK model is when such data are not available. ³⁰

PBPK models, based on physiological and drug‐related parameters, simulate the metabolic processes of drugs under various administration protocols in different subjects. ³¹ Researchers can predict potential pharmacokinetic differences between different patient populations and the effects of organ impairments, understand the potential for DDI, and select suitable dosing protocols for clinical trials by adjusting the corresponding parameters. ³² For GLS4/ritonavir, there is currently a lack of understanding regarding the effects of PK in special populations and DDI. In this study, we developed a PBPK model to elucidate the effects on PK when GLS4 co‐administered with ritonavir in special populations and the potential of DDI, guiding the design of subsequent rational clinical trial protocols and further verifying and refining the PBPK model. For the developed PBPK model to support dosage recommendations in drug product labeling for scenarios not tested by clinical trials.

Concurrently, clinical studies data of GLS4 have demonstrated a certain variability in exposure detected in subjects from different groups administered with the same dosage. This variability has, to a certain extent, increased the GMFE of the GLS4 PBPK model (Table S4). We hypothesize that the different genotypes of CYP3A4 may be one of the factors contributing to this phenomenon. However, without supportive data, we have not conducted a gene–drug interaction model for GLS4 in this study, and this will be one of our future research directions for GLS4.

In summary, this study developed the first PBPK model for GLS4 and successfully verified the predictive performance for the PK of GLS4, both with and without co‐administration of ritonavir. The developed PBPK model for GLS4/ritonavir enables prospective predicting of DDI with co‐drugs and exposure changes in special populations. It provides recommendations for dose adjustments, supports product labels, and has the potential to serve as an alternative to dedicated clinical trials. The PBPK model is an indispensable and powerful tool in drug development.

AUTHOR CONTRIBUTIONS

All authors wrote the manuscript; Yimin Cui, Zhaoqian Liu, Xia Zhao, Zexu Sun, and Nan Zhao designed the research; Zexu Sun, Nan Zhao, Ran Xie, Bo Jia, Junyu Xu, Lin Luo, Yulei Zhuang, Yuyu Peng, Xinchang Liu, and Yingjun Zhang performed the research; Yimin Cui, Zhaoqian Liu, Xia Zhao, Zexu Sun, and Nan Zhao analyzed the data.

FUNDING INFORMATION

This research was funded by the Science and Technology Research Project of Heilongjiang (No. 2022ZXJ02C02), Enterprise Key Laboratory of Anti‐viral Drug Development of Guangdong Province (No. 2020B1212070003) and State Key Laboratory of Anti‐Infective Drug Development.

CONFLICT OF INTEREST STATEMENT

Lin Luo, Yulei Zhuang, Yuyu Peng, Xinchang Liu and Yingjun Zhang are employees of Sunshine Lake Pharma Co., Ltd. All other authors declared no competing interests for this work.

Supporting information

Data S1

Sun Z, Zhao N, Xie R, et al. Physiologically‐based pharmacokinetic modeling predicts the drug interaction potential of GLS4 in co‐administered with ritonavir. CPT Pharmacometrics Syst Pharmacol. 2024;13:1503‐1512. doi: 10.1002/psp4.13184