The PBPK models for healthy and diseased population in three ethnic groups were developed using the PK-Sim^® (Version 10.0, Bayer Technology Services, Leverkusen, Germany).

The modeling parameters that were optimized and identified as having potentially significant effects on PK variables of OSI (under the steady-state concentration-time curve: AUC_ss, steady-state peak concentration: C_ss,max and plasma C_trough) were chosen for the sensitivity analysis. These specific parameters include: 1) f_up, 2) Rbp, 3) GET, 4) Albumin level, 5) CYP CL_int,u, 6) CYP3A4 auto inhibition and induction parameters, 7) CYP enzyme expression, and 8) liver volume.

To evaluate the influences of the selected parameters on PK variables, each parameter’s value was altered by ±20% during the sensitivity analysis. The sensitivity coefficient (SC) was then calculated using the following Eq. 4 (Li et al., 2020): SC = ∆ Y / Y ÷ ∆ P / P (4)

Where ∆Y is the alteration of predicted PK variables; Y is the initial value of predicted PK variables; ∆P is the alteration of model parameters; P is initial value of assessed parameters. If the absolute value of SC is above 1.0, it indicates that this parameter has a significant influence on the predicted PK variables.

The developed PBPK model was utilized to simulate the plasma C_trough and EGFRm⁺ inhibition time profiles for three ethnic populations with NSCLC. The virtual population’s demographic characteristics and dosage regimes were based on relevant clinical papers (Table 2). Each simulation consisted of ten virtual subjects. The estimation of EGFRm⁺ (T790M and L858R) inhibition in plasma or lung time profiles was achieved through the utilization of the following Eqs 5–7: d O E m d t = k o n × C l u n g × E m f r e e − k o f f × O E m (5) d E m f r e e d t = E m 0 − E m f r e e × k t u r n o v e r − k o n × C l u n g × E m f r e e + k o f f × O E m (6) Inhibition  % = O E m E m f r e e + O E m × 100 (7)

Where OEm is the concentration of OSI-EGFRm⁺ complex formed. Em_free is the free concentration of EGFRm⁺. Em₀ is starting concentration of EGFRm⁺, and was set 0.299 μM based on the paper (Bartelink et al., 2022). C_lung, k_on, k_off, and k_turnover are the free OSI concentration in the lung, association, dissociation rate constant of OSI, and re-synthesis rate constant of EGFRm⁺, respectively. The k_on values of 0.91 (binding to T790M) and 0.44 (binding to L858R) μM⁻¹s⁻¹ were obtained from the paper (Zhai et al., 2020), equivalent to the ratio of K_inact/K_i. k_off was assumed to be 0 due to irreversible covalent binding to EGFR for OSI. k_turnover was obtained to be 0.025 h⁻¹ from the paper (Greig et al., 2015).

The diseased PBPK model was used to evaluate the influence of CPY3A4 variants, CYP1A2 CL_int,u, auto-inhibition/induction on CYP3A4, f_up, and albumin level on plasma C_trough and pulmonary EGFRm⁺ inhibition. 26 CPY3A4 variant CL_int,u values from a referenced paper (Gao et al., 2022) were incorporated into the PBPK model to assess the impact. The CYP1A2 CL_int,u was varied within the range of 0.52–5.2 μL/min/pmol. Additionally, Auto-inhibition and induction were set at 0 for K_i/E_max CYP3A4 and 2.5 μM (K_i CYP3A4)/10.8 (E_max CYP3A4), respectively. The f_up values were set within the range of 0.009–0.034, while the albumin levels were set within the range of 1.1–6.1 g/dL. The dosing regimen involved taking 80 mg MD for 14 consecutive days, with the assumption that steady-state is reached after 14 days of OSI intake. The virtual population’s demographic characteristics were set to Caucasian, as mentioned in Table 2.

In the analysis conducted using JMP software (JMP Pro 16.0.0, SAS Institute Inc., NC, United States), a one-way analysis of variance (ANOVA) was performed. The Tukey-Kramer test was used to compare multiple groups. A p-value of less than 0.05 and 0.01 was considered statistically significant, indicating a significant difference between the groups being compared.

According to Table 2, the ratios of predicted and observed geometric mean AUC and C_max for OSI were within the acceptable range of 0.5–2.0, indicating that the population PBPK model accurately predicts the PK variables of OSI at single-dose and steady-state in the three ethnic populations (Caucasian, Japanese, and Chinese). Moreover, it is notable that most of the predicted/observed ratios fell within the range of 0.8–1.25, further suggesting a high level of accuracy in the PBPK model’s predictions. This indicates that the model is capable of accurately capturing the PK behavior of OSI in individuals from different ethnic backgrounds, both in terms of OSI exposure (AUC, C_max). Overall, the simulations validate the predictive performance of the PBPK model for OSI in healthy and diseased populations, also providing confidence in its ability to estimate PK parameters across diverse ethnic populations.

Additionally, the predicted ratios of PK variables for OSI co-administered with ITR and RIF are provided in Supplementary Table S1. These predicted ratios were found to be consistent with the clinical data (Vishwanathan et al., 2018b), further confirming the accuracy of the CYP3A4 metabolic contribution in the PBPK model.

According to Table 3, the most sensitive parameters with a significant impact on the AUC_ss, C_ss,_max and plasma C_trough of OSI are f_up and albumin level. The albumin level has a significant impact on all three PK variables, in particular on plasma C_trough, with a SC value of −2.10. Similarly, f_up also has a significant impact on two PK variables. Additionally, CYP3A4 CL_int,u, CYP3A4 concentration, and liver volume also exhibit a relatively large influence on PK variables. However, the majority of the modeling parameters had only a slight impact on the PK variables of OSI, as revealed by the sensitivity analysis.

Table 4 displays the predicted plasma C_trough values for various dosing regimens in NSCLC patients of three ethnic groups, as determined by the PBPK model. The observed ratios ranged from 0.80 to 1.25, indicating a good agreement between the predicted and actual C_trough values. These simulations demonstrate that the diseased PBPK model is capable of accurately predicting steady-state OSI plasma C_trough in NSCLC patients.

Figure 1 illustrates the time profiles of plasma and pulmonary EGFRm⁺ (T790M and L858R mutants) inhibition in the three ethnic groups during a 14-day period of treatment with 80 mg MD of OSI. According to a study on the effect of EGFRm⁺ suppression (Food and Drug Administration FDA, 2015), an effective pharmacodynamics (PD) threshold for OSI was defined as more than 80% EGFRm⁺ inhibition. The time-course of plasma and pulmonary T790M and L858R inhibition is similar across the three ethnic groups. Notably, the minimal rates of pulmonary inhibition for both mutants exceed 80%, which is significantly higher than in plasma. The simulations of pulmonary 80% inhibition align well with a study conducted on NCI H1975 cells (Food and Drug Administration FDA, 2015). As a result, the prediction of pulmonary T790M and L858R inhibition, as target tissue, over time proves valuable for assessing clinical efficacy compared to plasma EGFRm⁺ inhibition.

Based on Figures 5A, B, the plasma C_trough and pulmonary EGFRm⁺ inhibition in three ethnic populations were evaluated for four dosing regimens of OSI. The optimal dosing regimens for each ethnic group were determined based on the geometric mean of C_trough and minimal EGFRm⁺ inhibition (Calculated using pulmonary C_trough), along with a 95% confidence interval (95% CI). The results show that the plasma C_trough values at the dose of 80 mg OD in all three ethnic NSCLC patients fall within the desired PK threshold range. This ensures both clinical efficacy (C_trough > 338 nmol/L) and safety (Ct_rough < 677 nmol/L). Additionally, the minimal EGFRm⁺ inhibition achieved at the doses of 80 mg and 160 mg OD in all three ethnic NSCLC patients exceeds the desired 80% inhibition. Moreover, it is observed that the minimal EGFRm⁺ inhibition approaches 100% at the dose of 80 mg OD compared to 160 mg OD. This suggests that higher dosing regimens may have limited increases in clinical efficacy, and result in more adverse event. Therefore, the recommended dose of 80 mg OD in NSCLC patients, as suggested by the PBPK model, aligns well with the clinical dose suggestion.

The K_lu,p parameter plays a crucial role in characterizing the lung distribution of OSI from the plasma. According to the Rodgers and Rowland approach, the predicted K_lu,p -value was 14.3. However, experimental analysis revealed a mean K_lu,p -value of 28.5 (Dickinson et al., 2016). The impact of the K_lu,p -value on OSI plasma and lung PK is detailed in Supplementary Table S3. The analysis revealed that the K_lu,p -value has a significant impact on lung PK but hardly impact on plasma PK. Consequently, in this model, a K_lu,p -value of 28.5 was ultimately utilized to enhance the accuracy of lung concentration prediction.

In this study, the PBPK model was employed to simulate the impact of key parameters on OSI plasma C_trough and pulmonary EGFRm⁺ inhibition. The simulations revealed that CYP3A4 variants, f_up, and albumin levels significantly influence these PK and PD outcomes. Sensitivity analysis further emphasized that f_up and albumin level are the most sensitive parameters affecting the PK variables of OSI. Given the significance of three parameters, adjustments were made in the diseased PBPK model. As no differences in f_up were observed among the different ethnic groups, the same f_up value was utilized for the PBPK model across all three ethnic populations. However, hypoalbuminemia is commonly observed in NSCLC patients and those with hepatic impairment, leading to a decrease in albumin levels. This decrease can substantially alter the plasma protein binding of OSI, which has a high affinity for albumin. The simulations conducted in this study strongly demonstrated that albumin level has a substantial impact on the PK variables of OSI as well as EGFRm⁺ inhibition. These findings align well with multiple clinical observations where albumin levels in patients have shown a strong correlation with clinical efficacy (Brown et al., 2017; Yokota et al., 2022; Ishikawa et al., 2023).

In this study, despite CYP3A4 CL_int,u not showing high sensitivity in the sensitivity analysis, it still exhibits a large impact on OSI plasma C_trough as shown in Figure 3A. Additionally, the effect of CYP1A2 activity on OSI (a CYP1A2 substrate) plasma C_trough and pulmonary EGFRm⁺ inhibition was also examined, considering that smoking can induce CYP1A2 activity. The simulation demonstrated that significant changes occur in OSI plasma C_trough and pulmonary EGFRm⁺ inhibition when CYP1A2 activity is increased by 5- or 10-fold. On the other hand, OSI plasma C_trough and pulmonary EGFRm⁺ inhibition exhibited a decreasing trend but no significant change when CYP1A2 activity increased to 1.5-fold (approximately 1.55-fold higher in enzyme activity caused by smoking). This finding aligns with multiple clinical observations where there is no significant impact on the clinical efficacy between smokers and nonsmokers (Food and Drug Administration FDA, 2015; Kishikawa et al., 2020; Jin et al., 2022; Yokota et al., 2022), despite two clinical studies suggesting a potential impact (Rodier et al., 2022; Abu Hamdh and Nazzal, 2023).

It's worth noting that OSI inhibits its own metabolism through competitive inhibition against CYP3A4 and also enhances its own metabolism through induction of CYP3A4 expression. Although the sensitivity analysis did not show significant effects of auto-inhibition and induction parameters on PK variables of OSI (as seen in Table 3), statistically significant changes in plasma C_trough were observed with or without auto-inhibition and induction (Figure 3I). Hence, it is necessary that auto-inhibition/induction parameters were incorporated into the PBPK model of OSI.

Overall, this study highlights the crucial role of several factors in influencing the plasma C_trough and pulmonary EGFRm⁺ inhibition of OSI, particularly in diseased populations. The study supports the clinical relevance of f_up and albumin levels in predicting OSI efficacy. Additionally, the CYP3A4 and CYP1A2 activities also play relative important roles in determining OSI plasma C_trough and pulmonary EGFRm⁺ inhibition. The impact of smoking-induced CYP1A2 activity on OSI appears to be less significant compared to the influence of CYP3A4 activity. Additionally, the competitive inhibition and induction of CYP3A4 by OSI may contribute to changes in plasma C_trough, even though these effects were not prominently reflected in the sensitivity analysis.

In addition, the study revealed that T790M inhibition is less impacted by multiple factors compared to L858R inhibition. This disparity is primarily attributed to the larger k_on value of OSI against T790M than L858R (0.91 μM⁻¹s⁻¹ vs. 0.44 μM⁻¹s⁻¹). These key factors equally influence the plasma concentration of OSI, potentially resulting in smaller variations in pulmonary T790M inhibition due to the larger k_on value.

In certain clinical studies (Brown et al., 2017; Boosman et al., 2022; Hashino et al., 2023), when patients’ plasma C_trough levels for OSI were divided into only two groups, the relationship between exposure and efficacy was not observed. This is likely because clinical efficacy requires a minimal plasma C_trough, while excessive C_trough levels can lead to adverse events and result in shorter progression-free survival (PFS). However, in a specific clinical study (Abu Hamdh and Nazzal, 2023), when patients’ plasma C_trough levels were divided into four groups, it was found that a C_trough level above 328 nmol/L was associated with longer PFS, whereas a C_trough level below 677 nmol/L could avoid some adverse events and also induce longer PFS. Therefore, for OSI, defining the range of 328 nmol/L to 677 nmol/L for efficacy and safety PK seems appropriate. However, the exact target inhibition level for clinical efficacy is not yet fully defined. In some relevant papers, thresholds such as >90% inhibition for soluble epoxide hydrolase (Lee et al., 2019), >70% inhibition for α-Glucosidase (Wang et al., 2019), and >75% inhibition for ALK (Yamazaki, 2013) have been established. Considering these references, in this study, a duration of >80% pulmonary EGFRm⁺ inhibition was selected as the efficacy PD threshold. This choice aligns with a study on NCI H1975 cells (Food and Drug Administration FDA, 2015).

In NSCLC patients, the exposure and CL/F of OSI are much lower compared to healthy subjects. The f_up, albumin level, and metabolizing enzyme concentration may contribute to a significant increase in exposure among patients compared to healthy individuals (Figure 3). There is significant variability observed in OSI plasma concentration among individuals, with some patients showing more than 50% and even up to 80% variability (Planchard et al., 2016). Additionally, plasma exposure to AZ5104 is reported to be 10%–23% higher in Caucasian individuals compared to Asians (Brown et al., 2017). To account for these variabilities, the PBPK model analyzes key factors that influence the PK of OSI (as shown in Figures 2–4). The higher CL/F observed in Caucasians compared to Japanese and Chinese individuals is likely a main contribution of differences in healthy subjects and NSCLC patients (Figure 2). The observed higher CL/F in Caucasians may be linked to the greater abundance of CYP enzymes and larger liver volume in this population. Sensitivity analysis has shown that CYP3A4 concentration and liver volume have a relatively significant impact on AUC. (Table 3). The higher CL/F in healthy subjects may result from the difference in f_up, CYP enzyme abundance and albumin level in the two populations. The inter-patient PK variability can be attributed to variations in %CV of albumin levels and metabolism enzyme variants among different patients (Figure 4). The %CV of albumin levels is approximately 20% CV in Caucasians and 30% CV in Japanese and Chinese populations, leading to a 50% PK variability among patients (Figure 4). Moreover, The PK variability in inter-ethnic group may be attributed to metabolism difference of CYP enzyme in the three inter-ethnic groups.

In this study, the optimal dosage regimens of OSI for each ethnic population were determined based on the geometric mean and 95% CI of plasma C_trough and pulmonary EGFRm⁺ inhibition (Figure 5). These values were assessed to ensure they fell within the range of efficacy and safety PK/PD thresholds. This approach builds upon strategies proposed in previous studies (Johnson et al., 2019) and utilized in another research paper (Einolf et al., 2017). The PBPK model for patient population suggested that a dosage of 80 mg OD for OSI can achieve desirable values for both efficacy and safety based on plasma C_trough. Additionally, pulmonary EGFRm⁺ inhibition was observed to be above 80% at both 80 mg and 160 mg OD dosages. Furthermore, the 95% CI values for T790M inhibition (a specific mutation) in the three ethnic groups were found to be above 80%, indicating a favorable therapeutic response.

The present model has several limitations that should be acknowledged. Firstly, the time-course of pulmonary EGFRm⁺ inhibition by OSI could not be directly verified using human clinical study data, which limits the ability to fully validate the model’s predictions in this regard. Secondly, the two active metabolites of OSI have not been incorporated into the current PBPK approach, which may lead to incomplete representation of the potential EGFRm⁺ inhibition. This could affect the accuracy of the model’s predictions. Additionally, it is worth noting that the mean value of albumin levels was used in the PBPK model due to the lack of experimentally determined albumin data in multiple PK studies. However, individual variations in albumin levels can exist, and not accounting for this variability may introduce some errors or inaccuracies in the predictions of plasma concentrations and PK parameters.