Estimation of fm of victim drugs by human microsomes
In this study, we measured changes in intrinsic clearance of 33 compounds in HLM, with and without ketoconazole, using the substrate depletion method to evaluate the fraction metabolized by CYP3A4 for these drugs. Figure S4 illustrates a moderate positive correlation between in vitro Clint and Clliver without inhibitors, indicating the reliability of this method. In addition to the above in vitro method, we also used in vivo and in silico prediction methods to determinefm . We calculated in vivofm using a PBPK-DDI model, by matching the simulated DDI PK data of the test article to the observed DDI PK data. We predicted in silico fm using computational methods based on the drug’s molecular structure, using ADMET Predictor module. The findings of our study are presented in Table 2, which includes reported in vitro findings, phenotyping study conclusions, and tested in vivo , in vitro , and in silicofm values, compared across three different approaches. The in vitro fm results were consistent with the reported in vitro fmor phenotyping study conclusions. However, weak correlations were observed between the in vivo fm and in silico fm values, as well as the in vitroresults (showed in the Figure S5). As previously mentioned, drugs with higher Clliver are more likely to cause DDIs and are more sensitive to changes in metabolic parameters such asfm . To investigate this further, we compared thein vitro fm minus in vivofm values between drugs with Clliver above and below 15 L h-1, as well as the predicted in silico fm minus in vivo fm values. Figure 2 indicate that drugs with a Clliver higher than 15 L h-1exhibit smaller variations of predicted in silicofm minus in viv o fmand in vitro fm minus in vivofm compared to drugs with a Clliver lower than 15 L h-1. Furthermore, when comparing drugs with a Clliver higher than 15 L h-1, the difference between the in vitro fm and in vivo fm was smaller than that between in silico and in vivofm . These results suggest that in vitro fm measurements may provide more accurate predictions of DDI, particularly for drugs with higher Clliver.

Predictive performance of the PBPK-DDI model

The PBPK-DDI model was used to predict DDI results by incorporating different fm s, such as in silico, 100%, andin vitro fm . The results were presented in Figures 4, 5, and 6, showing predicted AUCR, CmaxR, and TmaxR values. Notably, the predicted TmaxR was found to be accurate within 2 times the observed TmaxR when using in silico or in vitro fm methods, but not with 100%fm . The PBPK model utilizing in vitro fm data outperforms the other two values offm in predicting CmaxR and AUCR, as demonstrated by the close proximity of the predicted values to the unit line (y=x) compared to the 100% fm and in silico fm integrated PBPK-DDI model. As a summary, in vitro fm was the most precise method, as it accurately predicted the CmaxR of 31 drugs within 2 times of the measured results, and the AUCR of 28 drugs within 2 times of the measured results. In contrast, the use of 100%fm and in silico fm led to lower prediction accuracy. Only 24 drugs had their CmaxR predicted within 2 times of the measured results when using 100%fm or in silico fm , and 26 drugs for CmaxR when using in silicofm . Similarly, only 24 and 25 drugs for AUCR were predicted within 2 times of the measured CmaxR when using 100% fm or in silicofm , respectively. The findings demonstrate that incorporating in vitro fm data into PBPK models significantly enhances the accuracy of predicting the extent of DDIs for the parameters studied. While in silico fm shows some promise, its impact on prediction is limited.