Firstly, the basic pyrotinib PBPK model was developed with the parameters both collected from preclinical and limited clinical data (SAD and MAD, NCT01937689) to describe the intracorporal process of the compound and then to support the DDI clinical trial design. Secondly, in order to improve the prediction performance of the model, and confirm the contribution of CYP3A to the metabolism of pyrotinib, the basic model was optimized and the mechanistic PBPK model was formed. Multi-dimensional validation was carried out on the mechanistic PBPK model, including not only SAD and MAD (NCT01937689) data, which were the same as the validation data of the basic model, but also the data of pyrotinib coadministration with capecitabine (NCT02361112) and pyrotinib coadministration with itraconazole (NCT04479891). Thirdly, untested scenarios including pyrotinib coadministration with CYP3A modulators, pyrotinib administration in hepatic dysfunction and elderly populations were simulated with the mechanistic pyrotinib PBPK model. Moreover, genotypes of CYP3A4*1G, CYP3A4*18A and CYP3A5*3, and the concentrations of 4β-OHC and CHO were also detected based on the blank hemocyte/plasma/serum samples from eighteen Chinese (NCT04479891). Meanwhile, based on the published protein abundance information of corresponding genotypes, the correlation between pyrotinib clearance and CYP3A5*3 allele was explored with the mechanistic model. Figure 1 displays the strategy diagram and Table 1 shows the simulation scenarios with the mechanistic PBPK model.

Since the basic pyrotinib PBPK model was originally developed to support clinical trial design, and the mechanistic PBPK model has been validated with multiple clinical data, only the final model parameters are presented here. The mechanistic pyrotinib PBPK model was developed using advanced dissolution, absorption and metabolism (ADAM) model in the SimCYP Population-Based Simulator (Version 19, SimCYP Limited, Sheffield, United Kingdom, Certara Company). The permeability of compound in Caco-2 Transwell model and dissolution profile in aqueous buffer were used to develop the absorption model. Full-PBPK model was used to describe the distribution characteristics of pyrotinib. The volume of distribution at steady state (V _ss) and partition coefficient of tissue to plasma (K _p) were predicted. Meanwhile, K _p scalar of 1.80 was used to match the observed concentration-time profiles. Elimination of pyrotinib in PBPK model was jointly depicted by intrinsic clearance of human recombinant CYP isoforms (CL _int), renal clearance (CL _R), additional systemic clearance (CL _{additional systemic}) and biliary clearance (CL _{int_Bile}) in SimCYP. The contribution of each CYP isoform to the overall hepatic metabolism was calculated according to the protein abundance of the specific CYP450 isoforms in human liver and its corresponding CL _int. Moreover, the contributions of kidney, unknown pathway and bile to the clearance of pyrotinib were obtained from the results of mass balance study in human (CTR20170528) and biliary intubation experiment in rats. Except for the CL _R (0.17 L/h) that were collected from clinical data, CL _{int_Bile} (0.4 μL/min/10⁶) and CL _{additional systemic} (4.5 L/h) were fitted according to the contribution percentage of pyrotinib clearance mentioned above and the PK profiles of pyrotinib in SAD clinical study. Physicochemical parameters, such as protein binding, B/P ratio and molecular weight (MW) were obtained from experimental results in vitro. LogP and pK _a were predicted based on the structure of pyrotinib. The detailed model parameters are summarized in Supplementary Table S1 of supplemental file.

The basic PBPK model was validated with limited clinical data, and the mechanistic PBPK model was fully validated with the clinical data that were obtained from the clinical studies of pyrotinib in healthy volunteers (SAD clinical study), the clinical studies of pyrotinib in breast cancer patients (MAD clinical study, NCT01937689), the efficacy study of pyrotinib coadministration with capecitabine (NCT02361112) and the DDI study of pyrotinib concomitant used with itraconazole (NCT04479891). The trial design in SimCYP Simulator was set to match population demographics (ethnicity, age and sex), dosage regimens and blood collection time points of each clinical study. The prediction performance of the mechanistic pyrotinib PBPK model was evaluated based on two criteria: 1) the observed values were within the 90% confidence interval (CI) of the predicted concentration-time profile; 2) the ratios of simulated area under the curve (AUC) and maximum concentration (C _max) values to those observed values were within a predefined boundary of 0.5 to 2.0-fold. All simulations were performed based on 10 trials with 10 subjects (n = 100).

Considering the selectivity, safety, quantitative predictability and maximum inhibition of inhibitor (Ke et al., 2014), itraconazole was selected as the strong inhibitor in pyrotinib clinical DDI study. To identify the optimal design of itraconazole concomitant with pyrotinib in clinical DDI study, a series of dosage regimens of itraconazole and pyrotinib, including dose, interval and duration, were simulated with the basic PBPK model. Combined with the safety of pyrotinib and itraconazole, the predicted AUC ratio (AUCR) and C _max ratio (C _max R) of pyrotinib coadministration with itraconazole versus pyrotinib given alone were used as the reference for the design of dosage regimens. All simulations were performed based on representative population (SimCYP, Version 18). The simulated scenarios of itraconazole combined with pyrotinib are shown in Supplementary Table S2 of supplemental file.

The mechanistic pyrotinib PBPK model was used to simulate the untested clinical DDI scenarios and estimate the effect of strong CYP3A inhibitors (clarithromycin, ketoconazole and itraconazole), moderate CY3A inhibitors (fluconazole, erythromycin, ciprofloxacin and diltiazem), mild CY3A inhibitors (fluvoxamine, fluoxetine and cyclosporine) and moderate CYP3A inducer (efavirenz) on the exposure of pyrotinib. “Chinese Healthy Volunteer” (20–50 years old, male:female = 1:1) and the default model of perpetrators within SimCYP were directly used for DDI simulations. To maximize the effect of perpetrators on pyrotinib, perpetrators were taken according to the clinical maximum dosage regimen, and a single dose of pyrotinib (400 mg, recommended dose in clinical treatment) was taken after the predicted plasma concentration of perpetrators reached the steady state. The detailed dosage regimens of untested DDI scenarios are shown in Table 1A.

The PK characteristics of pyrotinib in specific populations, including Chinese healthy volunteers, healthy volunteers (Caucasians), geriatrics and cirrhotic patients with Child-Pugh scores (CP) A, B and C (corresponding to mild, moderate and severe hepatic impairment) were simulated as well. As only 0.13% parent drug was excreted by kidney, the PK characteristic of renal impairment population was not estimated in this study. Except for Chinese healthy volunteers, other populations in SimCYP population library were Caucasians, so healthy volunteer (20–50 years old) was used as the baseline population. Geriatric population was divided into three subsets including 65–75, 75–85 and 85–95 age groups. For other specific populations, the simulated age range was set to 20–50 years old. Unless otherwise specified, physiological models within SimCYP database were directly used and all the simulations in this part included 10 trials with 10 subjects (n = 100) with the same proportion of male to female. Ultimately, the comparison of 400 mg pyrotinib system exposure in a specific population to that of baseline population (Healthy Volunteers of 20–50 years old, male: female = 1:1) was calculated in order to quantify the potential influence of internal factors on pyrotinib exposure. The detailed dosage regimens of simulations in specific populations are shown in Table 1B.

After overnight fasting, whole blood samples (before administration pyrotinib) were collected from 18 healthy volunteers participating in the DDI clinical study. The blood samples were immediately centrifuged at 2000×g, 4°C for 10 min. The hemocytes were collected for the detection of genotyping and the well-known single nucleotide polymorphisms (SNP) including CYP3A4*1G (rs2241480), CYP3A4*18A (rs28371759) and CYP3A5*3 (rs776746) were analyzed by polymerase chain reaction (PCR) and Sanger Sequencing. The content of serum CHO, one of the blood biochemical items, was detected during pre-enrollment screening (Day -2). The plasma samples were collected for the detection of 4β-OHC concentration by the method of LC-MS/MS. The detailed detection method of 4β-OHC concentration was as follows:

The ACQUITY UPLC^® I Class HPLC system (Waters, United States) connected to the API5500 Triple Quad mass spectrometer equipped with the Analyst 1.63 software (AB Sciex, Darmstadt, Germany) was used for the quantitative analysis. Chromatographic separation was performed on a XBridgeTM BEH C18 XP Column (4.6 × 75 mm, 2.5 μm, Waters). The mobile phases were (A) aqueous solution containing 0.1% formic acid (LC/MS grade, Fisher Scientific, United States) and (B) methanol (HPLC grade, Fisher Chemical, United States). Gradient elution was used as follows: 95% B at 0–4.9 min with a flow rate of 0.8 ml/min, 95%–60% B at 4.9–5.0 min with a flow rate of 0.8 ml/min, 60%–98% B at 5.0–5.3 min with a flow rate of 0.8 ml/min, 98% B at 5.3–5.9 min with a flow rate of 1.0 ml/min, 98%–95% B at 5.9–6.0min with a flow rate of 1.0 ml/min, 95% B at 6.0–8.0min with a flow rate of 0.8 ml/min. The injection volume was 5 μL. Mass spectrometric analysis was performed using positive ESI in multiple reaction-monitoring (MRM) mode. Ion transitions were m/z 385.3→109.1 and m/z 392.5→109.1 for 4β-OHC (Sigma-Aldrich/Avanti, Germany) and 4β-hydroxycholesterol-d7 (Internal standard, Sigma-Aldrich/Avanti, Germany), respectively. The analyte 4β-OHC displayed a good linearity at the range of 5–500 ng/ml. The developed method was validated according to the latest FDA guidelines for bioanalytical method validation.

Protein abundances of corresponding CYP3A4*1G, CYP3A4*18A and CYP3A5*3 genotypes in Chinese liver were retrieved from PubMed database with the keywords of “CYP3A4*1G/CYP3A4*18A/CYP3A5*3”, “protein abundance” and “Chinese volunteers”. Considering the inter-individual variation, protein abundance of individuals was obtained from literature (Xiao et al., 2019) by Plot Digitizer (Version 2.26, GetData, China). Variable coefficients were also calculated based on the observed values. The trial design in SimCYP Simulator was set to match the clinical study and the simulations were conducted based on 10 trials with 10 subjects (n = 100). In addition to changing the abundance of corresponding isoenzyme in the liver of Chinese physiological model, other parameters were directly used.

The statistical analyses about significance levels and correlation analyses were assessed with the GraphPad Prism Version 8.0 (GraphPad Software, San Diego, CA, United States), and p-value < 0.05 was used to estimate the significance.

Various dosage regimens (See Supplementary Table S2) of pyrotinib concomitant with strong inhibitor, itraconazole, were simulated with the basic PBPK model to explore the “worst-case” DDI scenario. The predicted DDI magnitude increased with the duration of itraconazole treatment, and the loading dose of itraconazole on the first day could also increase the inhibition of CYP3A4. Meanwhile, the increase of pyrotinib dose could not alter the value of AUCR. Since the predicted C _maxR was lower than two and the clinical dose of pyrotinib was 400 mg per day, the low-dose pyrotinib (80 mg, single dose) and itraconazole (200 mg, QD for 17 days), dosage regimen G in Supplementary Table S2 of supplemental file, were used in the DDI clinical trial to obtain the potential maximum DDI effect while ensuring the safety and tolerability of subjects (Liu et al., 2021).

The mechanistic pyrotinib PBPK model was developed and validated based on preclinical and clinical data. The final pyrotinib PBPK model parameters were presented in Supplementary Table S1. The observed and simulated geomean plasma concentration-time profiles for single or multiple dose(s) of pyrotinib were shown in Supplementary Figures S1–S3. Ratios (predicted value vs. observed value) of main pharmacokinetic parameters were approximate to 0.80–1.25 folds (See Supplementary Figure S4). All of them indicated that the model could well capture the absorption, distribution and elimination characteristics of pyrotinib in vivo. In addition, the mechanistic PBPK model was also validated by the data from clinical DDI study. The predicted and observed pyrotinib concentration-time profiles in the presence and absence of itraconazole were well matched and presented in Supplementary Figure S5, and the ratios of PK parameter between predicted values and observed values were within 0.80–1.25 folds. Meanwhile, the predicted metabolism fraction (f _m) of CYP3A4 to the total clearance of pyrotinib was consistent with the experimental results (Zhu et al., 2016), which illustrated that pyrotinib PBPK model was a mechanistic model and could be used to simulate preset scenarios without further modification. The predicted contribution percentage of each organ/CYP isoform was shown in Supplementary Figure S6.

The mechanistic PBPK model was used to evaluate the effects of perpetrators (CYP3A inhibitors and inducers) on pyrotinib exposure in “Chinese healthy volunteer” aged 20–50 years with the same proportion of male and female. In the presence of strong CYP3A inhibitors itraconazole, ketoconazole and clarithromycin, the AUC _0–168 h of pyrotinib (400 mg, clinical dosage) was increased by 7.4-fold, 7.5-fold and 3.4-fold, respectively, compared with that of 400 mg pyrotinib alone. Although clinical DDI study indicated that itraconazole could increase the systemic exposure of pyrotinib (80 mg) by 11.8-fold, this was obtained by comparing the AUC _0–168 h of pyrotinib in the DDI clinical study with the AUC _0–24 h of pyrotinib (80 mg) administrated alone. Actually, under the same comparison criteria (AUC _0–24 h), pyrotinib exposure could be increased by 6.36-fold in the presence of itraconazole. Therefore, there was no contradiction between the predicted and observed values of pyrotinib exposure in the presence of itraconazole. The pyrotinib exposure increased by 7.1-fold, 2.5-fold, 2.6-fold and 2.1-fold when combined with erythromycin, diltiazem, fluconazole and ciprofloxacin, respectively. However, erythromycin was an exception as its effect on pyrotinib exposure was similar to itraconazole. In fact, erythromycin was a time-dependent CYP 3A inhibitor, which had a potent impact on CYP3A isoenzyme (Yadav et al., 2018). The mild CYP3A inhibitors might also produce significant effects on the pyrotinib disposition due to that the predicted systemic exposure increased by 1.58-fold and 2.07-fold when pyrotinib combined with fluvoxamine and fluoxetine, respectively. CYP3A moderate inducer, efavirenz, appeared to significantly reduce the pyrotinib exposure (76%). Detailed results were shown in Figure 2.

Since pyrotinib was mainly metabolized by the liver, the exposure of pyrotinib in the liver impairment and geriatrics was estimated. Compared with the baseline population, the liver impairment patients with CP score A, B and C had the increasing exposure along with the progression of disease, and the ratios of the main PK parameters (AUC _{0-96   h} and C _max) ranged from 1.62 to 5.42 and 1.24 to 3.09, respectively. The exposure of pyrotinib in geriatric population (aged 65–95 years) increased with age, and the detailed AUC _{0-96   h} increments were 79, 105 and 131% corresponding to geriatric with 65–75, 75–85 and 85–95 years old, respectively. Results were exhibited in Figure 3.

In the limited samples, CYP3A4*18A (rs28371759) and CYP3A5*3 (rs776746) exhibited high mutation frequency, while CYP3A4*1G (rs2241480) showed a monomorphic genotype. The clearance of pyrotinib in individuals with different genotypes was statistically significant (Figure 4A), and only the CYP3A5*3 gene mutation had the notable influence on 4β-OHC concentration or the molar ratio of 4β-OHC to CHO (Figures 4B,C). Moreover, a genotype-independent linear correlation was found between exposure/clearance of pyrotinib and 4β-OHC concentration, and the correlation coefficient (r) was about 0.535, p < 0.05 (Figure 5A). Actually, there was a better genotype-independent linear correlation between the exposure/clearance of pyrotinib and 4β-OHC/CHO (r > 0.6, p < 0.05; Figure 5B). Therefore, the CYP3A basal expression level indeed had an impact on system exposure and plasma 4β-OHC concentration, and the ratio of 4β-OHC to CHO could be used as an indicator for pyrotinib exposure.

There were few published data about the protein abundance of CYP3A4*18A in the Chinese, and the median protein abundance values of corresponding CYP3A5*1/*1, CYP3A5*1/*3 and CYP3A5*3/*3 in Chinese population were obtained from reference (Xiao et al., 2019). The collected data were separately put into the liver module of Chinese population database to predict the CL/F of pyrotinib at different genotypes. Due to the small sample size (n = 18), only CYP3A5*1/*3 and CYP3A5*3/*3 genotypes were detected among those hemocyte samples. Thus, the predicted concentration-time profiles were validated according to the limited data. All of the observations were within the 90% confidence interval of the predicted values, which exhibited good prediction performance under corresponding protein abundance values. However, there was no significant difference in CL/F among different genotypes, and the p-value was 0.85. Validation results were shown in Figure 6.