Impact of Intracellular Concentrations on Metabolic Drug-Drug Interaction Studies

Accurate prediction of drug-drug interactions (DDI) is a challenging task in drug discovery and development. It requires determination of enzyme inhibition in vitro which is highly system-dependent for many compounds. The aim of this study was to investigate whether the determination of intracellular unbound concentrations in primary human hepatocytes can be used to bridge discrepancies between results obtained using human liver microsomes and hepatocytes. Specifically, we investigated if Kpuu could reconcile differences in CYP enzyme inhibition values (Ki or IC50). Firstly, our methodology for determination of Kpuu was optimized for human hepatocytes, using four well-studied reference compounds. Secondly, the methodology was applied to a series of structurally related CYP2C9 inhibitors from a Roche discovery project. Lastly, the Kpuu values of three commonly used CYP3A4 inhibitors—ketoconazole, itraconazole, and posaconazole—were determined and compared to compound-specific hepatic enrichment factors obtained from physiologically based modeling of clinical DDI studies with these three compounds. Kpuu obtained in suspended human hepatocytes gave good predictions of system-dependent differences in vitro. The Kpuu was also in fair agreement with the compound-specific hepatic enrichment factors in DDI models and can therefore be used to improve estimations of enrichment factors in physiologically based pharmacokinetic modeling. Electronic supplementary material The online version of this article (10.1208/s12248-019-0344-8) contains supplementary material, which is available to authorized users.


INTRODUCTION
Accurate predictions of drug-drug interactions (DDIs) are a challenging task during drug development because the relevant inhibitor concentration is not directly accessible (1,2). Predictions of DDIs depend on in vitro parameters such as the half-maximal inhibitory concentration (IC 50 ) or the inhibition constant (K i ) (3). These in vitro values are then used in physiologically based pharmacokinetic (PBPK) models to simulate DDIs in vivo (4).
For predictions of metabolic DDIs with CYP enzymes, K i or IC 50 values for a given compound are typically determined in liver microsomes or in hepatocytes. There is a growing interest in using cryopreserved human hepatocytes (HH) instead of human liver microsomes (HLM) for preclinical DDI assessment. In the microsomal setting, the drug concentration available to interact with the enzyme is equal to the unbound drug concentration in the incubation medium (i.e., the nominal incubation concentration, C inc , corrected for fraction of unbound drug in the liver microsomes (f u,mic )) (5). However, the HH model may be a more physiologically relevant system than HLM, because it takes into account multiple processes that influence enzyme inhibition. In the hepatocyte setting, compounds need to reach the cell interior where the CYP enzymes are located, a process that may involve passive permeability and active transport. Subsequent to this, the intracellular unbound drug concentration (C u,cell ) may be further influenced by metabolic clearance, protein binding, and partitioning into organelles (e.g., lysosomal trapping) or cellular membranes (5)(6)(7)(8). These factors can all contribute to inconsistencies between the C u,cell and C inc , leading to differences in the K i,app or IC 50,app values measured in HLM and HH. Therefore, careful consideration must be given to which in vitro system is appropriate for testing enzyme inhibition.
Determination of C u,cell is especially challenging if active transport is involved and is therefore classified as a low-confidence parameter in PBPK modeling (9). The US Food and Drug Administration (FDA) and the European Medicines Agency (EMA) recommended the [I]/K i ratio (where [I] is the inhibitor concentration) as a parameter in static DDI predictions (10). However, the determination of [I] is not currently standardized. The surrogates commonly used are the average or maximum unbound concentration in blood or at the inlet to the liver, or the maximum total concentration in the circulation at steady state (10)(11)(12). In addition to the different methods to estimate [I], the determination of K i also varies. For example, K i,app -based on determinations in cryopreserved HH suspended in human plasma-has been proposed (13). Such systems aim to reduce in vitro to in vivo translational uncertainties by more closely mimicking the in vivo situation, building protein binding and intracellular bioavailability considerations into the in vitro system.
Strategies for more direct estimations of C u,cell recently emerged (14)(15)(16)(17)(18). These include the parallel determination of hepatocellular drug accumulation at steady state (Kp) and the unbound fraction of drug in the cell (f u,cell ) which are combined to calculate the unbound partition coefficient Kp uu . Kp uu is defined as the ratio between unbound drug concentration in the cell interior and the unbound drug concentration in the cell exterior at steady state and can also be derived from kinetic parameters (17).
We recently introduced the term intracellular bioavailability (F ic ), defined as the ratio between the intracellular unbound concentration, which is available to elicit effects inside the cell, and the extracellular concentration. In the present work, F ic is equal to Kp uu , as no protein was added to the incubation media, and therefore, the term Kp uu has been used for simplicity. F ic or Kp uu account for active and passive mechanisms and predict target engagement and phenotypic responses in cells (19). We have shown that this parameter reflects transporter effects in cell lines transfected with single transporters. Further, we have seen differences in the Kp uu values for freshly isolated, plated HH, and suspended HH. This is explained by the altered clearance in the two systems (7).
Given this strong influence of culture formats on Kp uu , we hypothesized that Kp uu could be used to reconcile the differences in measurement obtained from different humanrelevant experimental systems. We therefore first investigated a series of reference CYP inhibitors (saquinavir, nelfinavir, enoxacin, and clarithromycin) in HLM and HH, previously shown to have differences in K i,app in rat liver microsomes and rat hepatocytes (20,21). Second, we applied an optimized assay for a series of investigational compounds from a drug discovery program that were all identified as CYP2C9 inhibitors. Finally, Kp uu was determined for a series of commonly used CYP inhibitors of the azole antifungal family-ketoconazole, itraconazole, and posaconazole. These compounds were chosen because they require hepatic enrichment factors (referred to as Bhepatic uptake value^in the SimCYP software), to reconcile in vitro and in vivo K i values (11,22). We reasoned that these scaling factors represent Kp uu at the site of DDI.

Chemicals
Compounds were retrieved from the in-house stock at the Roche laboratories at their highest available purity and dissolved at 10 mM in DMSO or, if lower, at their highest solubility. DMSO stocks were kept at − 20°C.

Compound Selection
The literature was screened for a validation set of compounds with reported discrepancies between cellular and microsomal IC 50 or Ki values. Enoxacin, clarithromycin, saquinavir, and nelfinavir were identified as suitable candidates (20). The method was then applied to an internal compound set from Roche (hereafter RO compounds) consisting of a structurally related series known to inhibit CYP2C9, and for which information on IC 50 in HLM and HH was available. As a third compound set, three members of the azole antifungal family-ketoconazole, itraconazole, and posaconazole-were identified as compounds with reported in vitro K i values that differ from the K i values reported in models predicting in vivo data.

Cell Culture
A pooled HH batch (BioreclamationIVT, LiverPool 10donor HH, product no. X008001, Lot RBR) was used in the CYP inhibition assay and for determination drug uptake in suspended HH. After thawing, the cells were suspended in InVitroGro CP medium at the specified concentration in each assay (product no. Z99029, BioreclamationIVT) and used immediately for experiments.

CYP Inhibition Assay in HLM and HH
For determination of CYP2C9 inhibition, 180 μL of HLM 0.2 mg/mL (BD Gentest, Cat no. 452117, pool from 150 donors of mixed gender) plus substrate (5 μM diclofenac for CYP2C9 and 5 μM midazolam for CYP3A4) was added to a deep 96-well plate together with the test compound or reference inhibitor (concentration range from 0.008 μM to 1 μM). After the addition of 20 μL of 10 mM NADPH, the plate was incubated for 5 min. The reaction was stopped by adding 200 μL of acetonitrile containing an internal standard. After centrifugation of 10 min at 6200×g, the concentration of the substrate in the samples was determined by LC-MS (Sciex API4000, see S6 for LC-MS parameters).
For HH, 50 μL of a suspension of cryopreserved HH (3 × 10 6 cells/mL) were added to a 96-well round bottom plate and pre-incubated for 20 min in Williams E medium (without FBS). Fifty microliters of a × 3 concentrated test compoundsolution and 50 μL of × 3 concentrated substrate (diclofenac for CYP2C9, midazolam for CYP3A4 and tacrine for CYP1A2) were added and the plate was incubated on a shaker at 37°C, 5% CO 2 , and 900 rpm. Incubations were stopped after 5 min by transferring 100 μL of the suspension to a fresh plate and adding 100 μL of acetonitrile containing the internal standard and then analyzed as above.
Where available, IC 50 values were compared to values found in the literature (Table S1).

Intracellular Unbound Drug Accumulation Ratio (Kp uu )
Kp uu was determined as previously described (7,18) using Eq. (1): where Kp is the steady-state cellular uptake, and f u,cell the fraction of unbound compound in the cell, determined as described below. The fraction unbound of the compound in the medium (f u,medium ) was equal to 1 as no serum proteins or other drug binding components were added to the incubation media.

Steady-State Cellular Uptake (Kp)
The Kp in suspended HH was measured similarly to previous reports with some modifications (7,18). Briefly, cells were rinsed twice with pre-warmed containing 10 mM HEPES. One hundred microliters of cell suspension (5 × 10 6 cells/mL) was added to a conical 96-well plate, and 100 μL of 1 μM drug solutions in Hank's buffered salt solution (HBSS) buffered with 10 mM HEPES was added (resulting in a 0.5 μM final concentration). All incubations were performed in triplicates on up to three independent occasions. Full-time curves were established at 1, 3, 7.5, 15, 30, and 45 min to determine Kp at equilibrium. For single time point measurements, the incubation time was matched to the CYP inhibition assays. A sample of the medium was removed after centrifugation in a pre-cooled centrifuge at 4°C (100×g, 5 min), and ten-fold diluted with a mixture of acetonitrile and water (60:40) containing the internal standard for determination of C medium . The remaining supernatant was removed by aspiration and cells were washed twice with icecold buffer. The intracellular compound was extracted using the acetonitrile to water to internal standard mixture for determination of the amount of drug in the cells (A cell ). Protein content (P cell ) was quantified using the BCA assay in representative wells to establish the cellular volume (V cell ), assuming 6.5 μL/mg protein (23). Finally, Kp was calculated using Eq. (2) Determination of f u,cell and f u,mic f u,cell was measured in cassette mode as previously described, but with minor modifications (24). Briefly, frozen cell pellets were thawed on ice and diluted to 10 million cells/ mL in HBSS containing 10 mM HEPES. The suspension was homogenized using a mini bead beater (Precellys, EQ02520-300-RD000.0, bead no. VK01) for 2 cycles of 10 s with an intermittent time of 30 s. Compounds were added to the cell homogenate at a final concentration of 0.5 μM and the spiked homogenate was transferred to a dialysis chamber (Rapid Equilibrium Dialysis Device, Thermo Fisher Scientific). HBSS buffered with 10 mM HEPES was placed in the receptor compartment. Samples of spiked cell homogenate were placed at 4 and 37°C for the duration of the experiment for recovery and stability calculations. Incubation time was 4 h at 37°C and 900 rpm for all compounds except itraconazole which reached equilibrium only after 24 h (Fig.  S4). The unbound fraction in the cell homogenate (f u,hom ) was determined according to Eq. (3): and the fraction of unbound compound in the cell (f u,cell ) was calculated by correcting for homogenate dilution according to Eq. (4): where the dilution constant D was calculated using Eq. (5), and assuming the V cell to be equal to 6.5 μL/mg protein (23). P hom is the protein concentration of the cell homogenate quantified using the BCA assay.
The lower detection limit was reached for determination of f u,cell below 0.01%, and the binding was assumed to be linear at the concentrations used, based on previous results (18).
For determination of the fraction unbound to microsomes (f u,mic ), HLM at 0.2 mg/mL (equivalent concentration to CYP inhibition assay; BD Gentest, Cat no. 452117, pool from 150 donors of mixed gender) were used instead of cell homogenate and calculated in analogy to f u,hom (Eq. 3).

Determination of Molecular Properties
Molecular properties of literature compounds were determined using ADMET predictor (Simulations Plus, version 7.2) and cross-checked to values published in the public databases DrugBank (25) and PubChem (http:// pubchem.ncbi.nlm.nih.gov). Molecular properties (logD, charge, PSA) of RO compounds were calculated using Roche proprietary in silico tools.

Statistical Analysis
All statistical analyses were performed in Graph-Pad Prism (version 7.04).

Method Optimization
Four reference compounds of different chemical character were studied first. The physicochemical properties of these drugs are summarized in Table I. Enoxacin and clarithromycin are poorly metabolized, hydrophilic drugs while saquinavir and nelfinavir are extensively metabolized, lipophilic drugs (26). All four have been previously shown to have a clear difference in their K i,app values for liver microsomes and suspended hepatocytes in the rat (Table S1). This made these compounds suitable for investigation in human liver microsomes (HLM) and hepatocytes (HH). The fraction unbound of the four compounds to HLM (f u,mic ) was determined using equilibrium dialysis and compared to f u,mic in rat liver microsomes (Fig. 1a, Table S3). Kp uu in suspended HH was determined by combining Kp and f u,cell (Table I). Kp uu in rat hepatocytes was derived from Kp values from the literature (Table S3). Enoxacin and clarithromycin had Kp uu values above 1, indicating that they accumulated in HH. In contrast, saquinavir and nelfinavir had Kp uu values below 1 (Fig. 1b).
Next, we used the f u,mic and Kp uu values in human and rat to scale apparent K i (or IC 50 ) values determined in microsomes and hepatocytes, respectively. Before scaling, these apparent values revealed a significant discrepancy between hepatocytes and microsomes, with the differences ranging up to 22-fold (Fig. 1c). To obtain corrected K i or IC 50 , we multiplied apparent K i or IC 50 in microsomes with f u,mic and apparent K i or IC 50 in hepatocytes with Kp uu (= Kp × f u,cell ). After applying these corrections, K i,app or IC 50,app values could largely be reconciled (i.e.,~2-fold differences). Nelfinavir-the most lipophilic compound in the   Tables S1 and S2. Enox, E: enoxacin; Clar, C: clarithromycin; Saq, S: saquinavir; Nel, N: nelfinavir series-appeared to be an outlier since after correction for f u,mic and Kp uu , the difference in K i or IC 50 increased rather than decreased.
Kp uu as Scaling Factor of IC 50 for an RO Discovery Series Inhibiting CYP2C9 We further applied our approach to an internal compound set (RO compounds). The nine compounds in this discovery series were outliers in the internal screening processes in which IC 50 values of HH and HLM were compared. The compounds of this series are known to inhibit CYP2C9. These nine compounds are structurally related, with a common core structure containing a sulfonamide and a secondary amide, and aromatic substituents including (iso-)thiazoles, benzothiophenes, pyridines, or furanes (Fig. 2a). They are all acidic at pH 7.4, with logD values indicating hydrophilic properties (Table II).
First, we determined the IC 50,app in HH and HLM using diclofenac as substrate. The IC 50 values of RO1, RO3, and RO4 were identical for HH and HLM, while the other compounds differed up to 12-fold (RO9) (Fig. 2b), with, a root-mean-squared error (RMSE) of 9.4 for IC 50,hep,app compared to IC 50,mic,app . As in the previous section, we next determined Kp uu and f u,mic taking care to use the same batches of hepatocytes and microsomes as for the inhibition experiments, respectively (Table II). All parameters were matched to the IC 50 assay.
We then used the Kp uu and f u,mic values as scaling factors to determine IC 50,corr . Correction using Kp uu or f u,mic alone did not improve the correlations (Fig. 2c and d). After combining Kp uu and f u,mic to give IC 50,hep,corr and IC 50mic,corr , respectively, the linear correlation between the IC 50 values improved, giving and R 2 of 0.92 and a 10-fold reduction in RMSE from 9.4 to 0.9 (Fig. 2e).
Because the compounds were analogous to each other, we also analyzed the possible influence of substituents on Kp uu . For instance, in five of the nine compounds, the R2 substituent was kept constant while the R1 substituent varied. However, no systematic influence of the change in substituent   could be observed. We conclude that the compound series was too small to allow a proper substituent analysis; this underscores the importance of experimental analysis of small series of compounds.

Kp uu in IVIVE and DDI Models of the Azole Antifungal Family
Finally, we investigated if Kp uu could improve predictions by DDI models that are based on in vitro K i values. Compounds of the azole family are commonly used as inhibitors of CYP3A4 in preclinical and clinical DDI studies (2,27). A concentrative cellular uptake of ketoconazole, itraconazole, and posaconazole has been widely described, but uptake mechanisms are not fully elucidated (28)(29)(30)(31). This concentrative uptake requires correction of K i (which is typically determined in HLM) in order to match in vivo K i values in DDI models. Therefore, compound-specific correction factors (referred to as Bhepatic enrichment factor^or Bhepatic uptake scalar^) have been introduced to improve the IVIVE in several models, which are summarized in Fig. 3a. We therefore investigated if Kp uu values measured in suspended HH agreed with these correction factors (Table III and Fig. 3b). The Kp uu values were high, ranging from 7 to 38, in line with the previously described concentrative uptake. Interestingly, the Kp uu values were within 2-fold of the empirically determined correction factors in the DDI models, indicating that Kp uu is a major contributor to the Bhepatic enrichment factor^for the azole series.

DISCUSSION
In this work, we evaluated the potential of Kp uu as scaling factor for CYP enzyme inhibition studies, using three experimental setups.
For the first set of compounds, we investigated if the determination of intracellular unbound concentrations could be used to reconcile differences in potency in CYP inhibition observed between microsomes and hepatocytes. CYP enzymes can be considered as targets or off-targets facing the cell interior. We have previously shown that F ic or Kp uu can explain differences in potency between isolated intracellular targets in biochemical high-throughput screening assays and the corresponding cellular assays (19). Drug response for intracellular targets in a variety of subcellular compartments including the cytosol, endosomes, and the nucleus has been successfully predicted, using our methodology for determination of F ic , F cyto , or F endo (19,32). Intracellular unbound drug concentrations are determined by combining the cellular accumulation ratio (Kp) with the fraction of unbound compound in the cell (f u,cell ). f u,cell is determined using the homogenization method which relies on the assumption that the degree of unspecific binding of drugs is not altered upon homogenization of cells (7,18).
In the first series of experiments, we used reference compounds with previously reported differences in K i of CYPs in rat liver microsomes and rat hepatocytes, as well as Kp, f u,cell , and f u,mic values in the rat (20).
Our experimental human f u,cell values differed more than 100-fold for the two lipophilic drugs compared to those in the rat study that was calculated based on lipophilicity and cellular volume (33) (saquinavir 0.14 vs. 0.0003 and nelfinavir 0.035 vs. 0.0002, rat vs. human, respectively, Table S3). There was a better agreement for the two more hydrophilic compounds (enoxacin 0.99 vs. 1.00 and clarithromycin 0.86 vs. 0.10 in rat vs. human, respectively). The homogenization/membrane dialysis method has been compared with the temperature and lipophilicity methods for estimating the f u,cell of a limited series of structural unrelated compounds by Riede et al. (15). In their study, the homogenization method gave generally lower values than the a b  temperature and lipophilicity methods. This suggests an overestimation of intracellular binding by the latter two. However, a large variability (up to 13-fold differences) between laboratories has been observed (7,15,17,34,35). f u,cell is dependent on accurate dilution factors for scaling f u,hom to f u,cell (Eq. 4 and 5). In the comparative study by Riede et al. (15), very low cell concentrations (130,000 cells/mL) were used, in comparison to 10 million cells/mL in our studies or 50 million cells/mL by Riccardi et al. (34,35). Such experimental differences may not allow a reliable comparison between laboratories, but in general, all experimental setups agree on significantly lower f u,cell for highly lipophilic drugs as compared to calculations based on logD. This indicates the importance of using a cellular matrix with relevant lipid and protein binding sites of drugs for the determination of f u,cell (8). Moreover, no significant inter-species variability in f u,cell has been observed in a recent comparative study (34). Therefore, for the four reference compounds, the methodological differences were more significant than the expected interspecies variability in f u,cell , and we used our experimental f u,cell values for determination of Kp uu in both rat and human. On the other hand, inter-species variability is expected in Kp values, which can be influenced by active uptake and elimination processes. Therefore, the specific Kp in rat or human hepatocytes was used for determination of Kp uu (Table S3). For the four reference compounds, Kp for the human and rat hepatocytes followed the same rank order and were in the same order of magnitude, with slightly higher values in rat hepatocytes for three out of the four compounds. It should be noted that clarithromycin is a weak base that can be subject to lysosomal trapping. Using pH partitioning theory (18), the calculated cytosolic Kp uu (or F cyto ) in human hepatocytes would result in 1.3, instead of the experimental value of 2.7, which does not account for subcellular localization of the drug. Two of the four reference compounds, enoxacin and clarithromycin, displayed a K i ratio (K i,mic /K i,hep ) or IC 50 ratio and Kp uu above one, indicating a concentrative uptake of the compounds in hepatocytes. By contrast, the K i ratio or IC 50 ratio and Kp uu were less than one for saquinavir and nelfinavir, suggesting that the lower potency in rat hepatocytes was driven by poor access to the intracellular compartment (and to the CYP enzyme) as opposed to the microsomes where the CYP enzyme is freely exposed in the medium. For three out of the four compounds, K i or IC 50 values in liver microsomes and hepatocytes were in good agreement after applying relevant Kp uu and f u,mic as scaling factors (Fig. 1c) (34)(35)(36). For nelfinavir, scaling K i or IC 50 values with Kp uu and f u,mic resulted in a larger discrepancy rather than an improvement. Interestingly, nelfinavir was identified as an outlier also in our previous Kp uu studies in MDCK cells, suggesting the involvement of unknown active processes that are not captured by the Kp uu methodology (7). Nelfinavir is very lipophilic, poorly soluble, and a substrate of active efflux (P-gp)-these properties suggest that the compound is prone to nonspecific binding and that there are significant confounding effects occurring in the hepatocyte experiment. Indeed, poor mass balance (~50%) was observed for nelfinavir and the similarly lipophilic saquinavir in the hepatocyte experiments. As the mass balance was recovered in the presence of the CYP inhibitor ABT, this confirms that the loss of compound is due to hepatic metabolism (Fig. S4). The hepatocyte metabolism of nelfinavir and saquinavir has been reported previously in a study that identified these two compounds as outliers. This was based on their intrinsic clearance in microsomes being much higher than in hepatocytes, compared to other compounds with similar clearance mechanisms (37).
It should be noted, however, that in our studies, the hepatocyte metabolism of nelfinavir and saquinavir was not a limiting factor for establishing the concentration equilibrium, since Kp values were unaffected by the presence or absence of ABT (Fig. S4). Furthermore, the use of albumin or an increase in temperature of wash buffers from 4 to 37°C (38,39) did not change the results significantly (Fig. S5), which precludes the likelihood that non-specific binding was affecting the results. Thus, the discrepancy in the K i,corr values of nelfinavir could not be explained by CYP-mediated metabolism or non-specific binding and is likely driven by other factors. Despite the discrepancy for nelfinavir, we overall obtained a significant harmonization of the K i ratios between microsomes and hepatocytes, suggesting the validity of this approach and indicating the value of further elaboration.
In the next set of experiments, we therefore harmonized the experimental conditions between the CYP inhibition experiments and the Kp uu experiments. Instead of comparing Kp uu to CYP inhibition constants from the literature, we determined IC 50 in HH and HLM with methodologies as similar as possible to each other to generate a consistent dataset. As a result, the Kp uu -and f u,mic -corrected IC 50 values for the discovery series of CYP2C9 enzyme inhibitors were in excellent agreement for HH and HLM (Fig. 2e). We conclude that the best agreement between the two systems is obtained by applying simultaneous correction factors for both systems, along with taking into consideration differences in assay setup and inter-batch variability.
Given the significance of Kp uu in the two series of in vitro experiments, we next evaluated if it could also be used as a scaling factor in in vivo studies. More specifically, we investigated if Kp uu could be used to estimate the empirical hepatic uptake scalar required in PBPK models of ketoconazole, itraconazole, and posaconazole as CYP inhibitors. Many models use such empirically determined scalars (Fig. 3a). However, others avoid them because the uptake mechanism of these azole compounds are not mediated via classic hepatic uptake transporters such as the OATP or OCT transporter family (29,40). Kp uu is a measure of all combined effects in a cell and provides a mechanistic tool for estimation of uptake scalars without the need for elucidating the underlying factors contributing to the net drug disposition in the cell. We found that Kp uu values in suspended HH were in good agreement with the empirically determined hepatic uptake scalars in the models for both rat and human (Fig. 3). In the case of itraconazole, the three major metabolites are equally or more potent inhibitors of CYP3A4 than the parent compound, which is a complicating factor in PBPK modeling (28,40,41). The measurement of Kp uu of the parent compound alone was however sufficient for recovering the hepatic uptake scalars. Therefore, we believe that Kp uu can be rationally integrated into prospective PBPK modeling approaches. The use of HH instead of its rat counterpart reduces the potential effects of inter-species differences in drug transporter expression and metabolic enzyme sensitivity to inhibition. It is also the most relevant in vivo model system for making extrapolations to humans. The relevance of Kp uu to in vivo studies is further supported by a recent study in which a series of four statins was assessed for in vitro and in vivo liver enrichment (35); in vivo liver-to-plasma, Kp uu in rat and human were in good agreement with in vitro Kp uu obtained in cryopreserved suspended hepatocytes. Similar scenarios are anticipated for, e.g., rifampicin, which has an empirical hepatic uptake scalar of 16.9 (42) and for bosentan, with a hepatic uptake scalar of 5 to 6 (43).
Determination of K i,app directly in HH can be an alternative to the incorporation of hepatic uptake scalars. However, most K i screens are based on HLM. To make our approach more generally applicable, we suggest the use of K i,corr values, and account for intracellular unbound concentrations in the target cells in any tissue of interest (Fig. 4). Our approach is not limited to the liver and can be applied on any cell of interest with an intracellular target, including studies of DDIs in intestinal enterocytes.

CONCLUSION
In summary, our results indicate that Kp uu in HH is an easy-to-interpret in vitro parameter that can be used as a human-relevant scaling factor to bridge the differences of experimental systems such as liver microsomes and hepatocytes and gives a mechanistic understanding of any result discrepancies (Fig. 4). Since Kp uu is the net result of all processes that affect the intracellular unbound drug concentration in HH, no prior knowledge of uptake and elimination processes is required. Indeed, very high or low Kp uu values can indicate that active cellular processes are playing a significant role in drug disposition. Therefore, Kp uu can be used as a decision tool (e.g., if the Kp uu is lower than 0.5 or higher than 2) for more detailed investigations. Very lipophilic compounds have been found to be more challenging for assessment of Kp uu and future work should focus on this compound class. Further, Kp uu has been shown to reflect hepatic uptake scalars used in literature for the triazole antifungal family and we believe that the use of Kp uu is an approach that can be pursued to provide a mechanistic understanding of scalars used in PBPK models that predict drug exposure and DDI potential.