Calibrating the in vitro-in vivo correlation for OATP mediated drug-drug interactions with rosuvastatin using static and PBPK models

252 of 250 words max. Introduction: 451 of 750 words max. Discussion: 1594 of 1500 words max. Abbreviations: AUCR, AUC ratio; BCRP; breast cancer resistance protein; CCK8, cholecystokinin-8; DDI, drug-drug interaction; DMEM, Dulbecco’s modified eagle medium; E217βG, estradiol 17-β-D-glucuronide; HEK293, human embryonic kidney 293; IC50, halfmaximal inhibitory concentration; IVIVC, in vitro-in vivo correlation; NPV, negative predictive This article has not been copyedited and formatted. The final version may differ from this version. DMD Fast Forward. Published on October 9, 2020 as DOI: 10.1124/dmd.120.000149 at A PE T Jornals on M ay 1, 2021 dm d.aspurnals.org D ow nladed from


Significance statement
Using 22 drugs, we show that a static model for OATP1B1/1B3 inhibition can qualitatively predict potential for DDI using a cut-off based approach as in regulatory guidances. However, consideration of both OATP1B1/3 and gut BCRP inhibition provided a better prediction of the magnitude of the transporter-mediated DDI of these inhibitors with rosuvastatin. Based on these results, we have proposed an empirical mechanistic-static approach for a more reliable prediction of transporter-mediated DDI liability with rosuvastatin that drug development teams can leverage.

Introduction
Organic anion transporting polypeptides (OATP) 1B1/3, which are mainly expressed in the liver, are clinically important transporters for drug-drug interactions (DDI) (EMA, 2012;PMDA, 2019;FDA, 2020b). During drug development, the in vivo OATP1B-mediated DDI liability of an investigational drug is often assessed using in vitro inhibition assays, and the in vitro inhibition data is used to predict the potential to inhibit these transporters in vivo. Since inhibition can be substrate-dependent, careful choice of in vitro probes is an essential first step to minimize false negative (FN) as well as false positive (FP) predictions (Izumi et al., 2015;FDA, 2020b).
The science of in vitro to in vivo extrapolation for transporter related DDI encompasses several perspectives and considerations. For instance, for a qualitative prediction of whether an in vitro inhibitor of OATP1B1/3 is likely to inhibit these transporters in vivo, the regulatory guidances have recommended cutoff values for R (the predicted ratio of the area under the curve in the presence and absence of the investigational drug as an inhibitor) based on a basic model (EMA, 2012;PMDA, 2019;FDA, 2020b). These DDI guidances highlight the need to eliminate false negatives and propose an empirical qualitative cutoff value-based assessment to determine whether in vivo DDI is likely. From a clinical pharmacology perspective, a quantitative prediction is of high relevance to understand clinical impact of a DDI. To that end, mechanistic models have been applied for DDI prediction. 7 For success of either a cutoff based approach or a mechanistic model based prediction, appropriate in vitro data that give the most accurate predictions is vital to provide optimal guidance to the drug development teams as well as patients enrolled in the studies. We evaluated the predictive performance of RST versus E 2 17βG and CCK8 in our in vitro inhibition assay for qualitative R-value cutoff based prediction as well as quantitative prediction of DDI using 22 compounds. We further attempted to holistically approach transporter DDI prediction for these 22 compounds by using a combined static mechanistic model that takes into consideration in vitro inhibition of OATP1B1/3 and BCRP, based on the mechanistic information about RST that its clearance is mediated by both, OATP1B1/3 and gut BCRP. Lastly, we also used PBPK modeling to simulate the DDI between RST and three of the 22 studied drugs, rifampicin, asunaprevir, and velpatasvir, to evaluate the DDI potential of these drugs using the dynamic method to evaluate if PBPK would further refine the predictions beyond the other approaches.

Cell System and Transporter Inhibition Assay for OATP1B1/3
Transporter inhibition assay was conducted as follows.

R-value calculations using basic model
AUCR estimated by basic model was calculated using the following equation (FDA, 2020b): , ,

Equation 1
I u,in,max is the estimated maximum unbound plasma inhibitor concentration at the inlet to the liver IC 50 is the half-maximal inhibitory concentration f u,p is the unbound fraction in plasma; for highly bound drugs, the f u,p were rounded up to 0.01 for this part of the calculation. (Supplemental Table 1) C max is the maximal plasma concentration F a is the fraction absorbed. F a =1 was used as the worst-case estimate F g is the intestinal availability. F g = 1 was used as the worst-case estimate. k a is the absorption rate constant. ka = 0.1/min was used as the worst-case estimate.
Q h is the hepatic blood flow rate. Qh = 1500 mL/min. R B is the blood-to-plasma concentration ratio.
R B was assumed to be 1.

Static models for DDI predictions
The in vivo OATP1B1/3-mediated DDI magnitude with RST was quantitatively predicted using the following basic static model equation: AUCR predicted by static model = ( ) Equation 2 where f e,OATP1B is the fraction of systemic clearance of RST that was mediated by OATP1B1 and 1B3.
I u,in,max is the estimated maximum unbound plasma inhibitor concentration at the inlet to the liver IC 50 is the half maximal inhibitory concentration Since hepatic clearance contributes to 72% of the total body rosuvastatin clearance, and the relative contributions of OATP1B1, 1B3 and NTCP to the overall hepatic uptake of RST has been estimated to be 70%, 20% and 10%, respectively (Wang et al., 2017, Elsby 2012), the f e,OATP1B1 equals 0.504 and f e,OATP1B3 is approximately 0.144.
For those studied drugs that had the potential to inhibit intestinal BCRP in vivo, as determined by the criteria of I gut /IC 50 ≥ 10, where I gut = dose of studied drug/250 mL (FDA, 2020b), equation ) Equation 3 where f e,BCRP is the fraction excreted by BCRP in the gut, which is 0.5 (Elsby 2015) and f e,OATP1B is the fraction of systemic clearance of RST that was mediated by OATP1B1 and 1B3.

Model performance assessment for static and combined static model
In addition to the FDA recommended cutoff values of R ≥ 1.1, the cutoff recommended by EMA (R ≥ 1.04) was also used to calibrate the in vitro system (EMA, 2012;FDA, 2020b).
To qualitatively assess the predictive performance of RST and E 2 17βG /CCK8 using cutoff- This article has not been copyedited and formatted. The final version may differ from this version. To quantitatively assess the predictive performance of RST and E 2 17βG /CCK8, the root mean squared error (RMSE) was calculated using the following equation: Retrospective prediction of clinical DDI with PBPK modelling DDI simulation with RST using PBPK approach was performed for three of the 22 drugs. These three drugs were selected based on their OAT1B and intestinal BCRP inhibitory potencies: rifampicin is a well-known strong OATP1B inhibitor; velpatasvir is a moderate OATP1B inhibitor, and asunaprevir is a weak OATP1B inhibitor.

DDI simulations
Clinical DDIs between the three drugs (rifampicin, asunaprevir, and velpatasvir) and RST were predicted using the established models. The model parameters are provided in Supplemental  Fig. 2, and Supplemental Table 2). The RMSE of predictions using the combined static model when intestinal BCRP inhibition was also taken into account in addition to OATP1B1/3 inhibition were 0.767 and 0.812 for RST and E217βG/CCK8, respectively (Table   1, Fig. 2, and Supplemental Table 2).

DDI simulations between RST and rifampicin, velpatasvir, and asunaprevir using PBPK
modeling PBPK models were constructed to retrospectively predict DDI of rifampicin, asunaprevir, and velpatasvir with RST. In order to sufficiently describe the observed AUCR, a scaling factor was applied to the experimental IC 50 before incorporating to the PBPK model. This scaling factor was found to be substrate-dependent; for IC 50 obtained from inhibitory assays with RST as substrate, a factor of 200 was applied and a factor of 100 was applied to those obtained from assay using E217βG/CCK8 as substrates (Supplemental Table 4). Similar scaling factors were applied in other reported studies (Chen et al., 2018;Yoshida et al., 2018). The PBPK models reasonably predicted the magnitude of DDI between RST and rifampicin and asunaprevir with the predicted AUCR < 1.5-fold difference compared to the observed AUCR (Fig. 3) (Prueksaritanont et al., 2014;Eley et al., 2015;Lai et al., 2016;Prueksaritanont et al., 2017; This article has not been copyedited and formatted. The final version may differ from this version.  Takehara et al., 2018). The DDI simulation between RST and velpatasvir was underpredicted with a 2-fold difference compared to the observed AUCR (Mogalian et al., 2016).

Discussion
In this study, we evaluated the suitability of AUCR (predicted AUC ratio due to DDI) cutoff values suggested by the regulatory agencies with the aim of calibrating our internal in vitro assay system for OATP1B1/1B3 inhibition, for qualitative as well as quantitative DDI predictions (EMA, 2012;PMDA, 2019;FDA, 2020b). Using either RST or E 2 17βG/ CCK8 as probe substrates, the in vitro DDI potential of 22 selected drugs, which have previously been studied in clinical DDI studies with RST, was evaluated using the basic model, and the predicted DDI magnitude was compared to the observed DDI in vivo. Substrate dependent inhibition of OATP1B1/1B3 is well documented (Izumi et al., 2015) and we corroborated the observation that the in vitro assays using E217βG/CCK8 as probes yield higher R-values (more potent IC 50 values) than those using RST as a probe substrate (Izumi et al., 2015).
For a qualitative analysis, a true or FP or FN prediction was made using the IC 50 values to predict magnitude of in vivo DDI for RST using static models (EMA, 2012;PMDA, 2019;FDA, 2020b To translate the inhibitory potency from in vitro inhibition assays to more accurate quantitative prediction, approaches with mechanistic understanding are needed, such as a static mechanistic model that combines inhibition of multiple clearance pathways or even a more comprehensive dynamic PBPK model. In order to utilize these models, a thorough understanding of the disposition pathway of the clinical probe substrate i.e. RST, is essential. For instance, while OATP1B1/3 are identified as the key players in the hepatic uptake of RST, the involvement of the BCRP also needs careful consideration because it governs the absorption of the RST from the intestine (Elsby et al., 2012;Hua et al., 2012;Elsby et al., 2016;Bae et al., 2018). Studies have found BCRP is responsible for as much as 50% of drug efflux from gut enterocyte. Similarly, sodium-taurocholate co-transporting polypeptide (NTCP) also participates in the transport of RST in the liver. Based on in vitro data, the relative contributions of OATP1B1, 1B3, and NTCP to the overall hepatic uptake of RST has been estimated to be 70%, 20%, and 10%, respectively (Wang et al., 2017). The hepatic clearance of RST was reported to be approximately 72% of the total plasma clearance (Martin et al., 2003a;Martin et al., 2003b). Hence, the fractional contribution of transporters towards the clearance of RST can be estimated to be 0.504, 0.144, and 0.072 for OATP1B1, 1B3, and NTCP, respectively. Other studies report somewhat higher contribution of OATP1B3, OATP2B1 and NTCP at 10-35% toward RST clearance (Ho et al., 2006;Bi et al., 2013;Wang et al., 2017;Zhang et al., 2019). Given the difficulties in accurate estimation of contribution by each transporter in RST clearance, in this study, we considered only the roles of BCRP and OATP1B1/3 in RST disposition, since together, these account for a majority of RST clearance. In addition, our focus was to determine the most reliable evaluation of in vitro data requested by regulatory agencies in order to better predict RST DDI. This allowed for a reasonable prediction of the AUCR while maintaining a simple model. Generally, there is also a concern about accurate DDI prediction for highly protein bound drugs (>99%) for OATP1B1/1B3. In our assessments, almost all the drugs reported in the University of Washington DDI database, including the ones used in our study, which had both, a DDI with rosuvastatin (in vivo AUCR>1.25) and were highly protein bound (>99% bound), also inhibited BCRP in vitro (list provided in Supplemental Table 6). Since prediction of gut BCRP inhibition takes into account the total gut concentrations of the inhibitor, the impact of protein binding value used in overall predictions for these drugs (i.e. actual <1% value versus rounded up to 1% unbound as recommended in the regulatory guidances) did not significantly impact the predictions.
Using DDI guidance cutoff values, the two static models were compared for their qualitative predictive performance. For a quantitative analysis, the magnitude of the predicted DDI using static model of OATP1B1/1B3 inhibition was modified to incorporate the gut BCRP inhibition and the predictions with the combined static model were comparatively better than with the static model using OATP1B1/1B3 inhibition alone.
From a qualitative point of view, as described before, despite the substrate-dependent differences in the IC 50 potencies, for either RST or E217βG/CCK8 as in vitro probe substrates, the PPV and NPV metrics were fairly comparable when only OATP1B1 and OATP1B3 were considered as clearance mechanisms for RST. Although PPV and NPV were high, there were four FNs and one FP. However, it is noteworthy that except for velpatasvir, which has a complex disposition, all the other FNs appeared to cause only a mild DDI with <1.6-fold increase in RST AUC observed in vivo (Mogalian et al., 2017). Therefore, this calibration analysis suggests that the cutoff of 1.1 is reasonable and adequate for the in vitro systems tested here to capture clinically meaningful DDI of ≥ 1.6-fold. From a quantitative point of view, the static model that only considers OATP1B1/1B3, slightly under predicted the magnitude of the interaction ( Fig. 2A). Inclusion of BCRP inhibition in both static mechanistic model and PBPK for DDI prediction is crucial for some inhibitors that may inhibit BCRP in addition to OATP1B1/1B3. We used the criteria recommended by regulatory agencies to estimate whether an investigational drug has a potential to inhibit gut BCRP efflux of RST: I gut /IC 50 ≥ 10, where I gut = dose of inhibitor/250 mL (EMA, 2012;PMDA, 2019;FDA, 2020b). With the combined static model, which incorporated both inhibition of OATP1B1/1B3 and BCRP when relevant, the quantitative prediction improved as compared to OATP1B1/1B3 alone (Fig. 2B). The success of this approach for a broad range of compounds studied here corroborates the understanding that gut BCRP plays a significant role in RST disposition (Elsby et al., 2012;Elsby et al., 2016). The qualitative metric NPV improved with this model with no FNs, but the PPV decreased slightly due to an increase in FPs (Supplemental Table 2). Two potential reasons could be the overestimation of the in vitro inhibitory potential in vesicular transport assay because of the lack of protein binding as well as easier access to transporter in inside-out vesicles; and over-prediction of the drug concentration in the gut enterocytes leading to over-estimation of DDI at the enterocyte level in this empirical, static approach. As such, the in vitro to in vivo translation of gut BCRP inhibition from in vitro assays may need further tuning to reduce the FP cases.
To overcome the limitation of empirical estimation of static gut and liver concentrations, PBPK is closer to the observed AUCR, than using OATP1B1/1B3 alone (Jones et al., 2020). This example illustrates the advantage of using combined static model to provide a more accurate quantitative prediction. Such an approach can be employed to systematically investigate and then predict DDI involving multiple transporters with other probes of interest such as pitavastatin or pravastatin.
Overall, our results highlighted that: (a) from a qualitative perspective, the current R-value cutoff criteria recommended by FDA and PMDA (R > 1.1) appears to be reasonable for the in vitro

Funding
There was no external funding for this work.
This article has not been copyedited and formatted. The final version may differ from this version.   DDI simulations between RST and rifampicin, velpatasvir and asunaprevir using PBPK modeling. For rifampicin, the error bars represent the range of AUCRs reported from three studies (Prueksaritanont et al., 2014;Lai et al., 2016;Shen et al., 2017). For asuanprevir and This article has not been copyedited and formatted. The final version may differ from this version.