Abstract
Assessment of time-dependent inhibition (TDI), especially CYP3A4, is an important parameter for preclinical and clinical development. The use of human liver microsomes (HLM) is the most common in vitro matrix to assess TDI, but this often leads to an overprediction of an actual effect observed clinically. Recently, the use of human hepatocytes has been hypothesized as a more relevant and possibly predictive matrix for the assessment of CYP3A4 TDI. Our work evaluates and optimizes three different human hepatocyte assays for the assessment of CYP3A4 TDI using pooled cryopreserved human hepatocytes. Using two of the optimized methods, the time-dependent inhibition kinetic parameters (KI and kinact) for four known CYP3A4 TDI (diltiazem, erythromycin, verapamil, and troleandomycin) were determined. When comparing TDI in HLM, the KI values from hepatocytes were in general 4- to 13-fold higher than that in HLM, whereas the kinact values in human hepatocytes were similar or slightly higher or lower depending on the inhibitor. The inactivation potency (kinact/KI) for four tested CYP3A4 inactivators in human hepatocytes was generally lower than that in HLM due to either lower affinity (KI) or lower inactivation rate (kinact) or both. When drug interactions were simulated with Simcyp using either HLM or human hepatocyte data, the predictions using the kinetic parameters from human hepatocytes resulted in a much better simulated change in pharmacokinetics compared with observed clinical data.
Introduction
Time-dependent inhibition (TDI), also known as mechanism-based inhibition, is associated with irreversible or quasi-irreversible inactivation of an enzyme. Drug-drug interactions precipitated from TDI are considered more serious because of the persistence of the inhibitory effects even after the inhibitor has been eliminated. There are several examples of clinically relevant drug interactions involving TDI covering several cytochrome P450 isoforms, with most interactions involving CYP3A4 (Venkatakrishnan and Obach 2007; Grimm et al., 2009). Thus, identification of potential CYP3A4 time-dependent inhibitors using appropriate in vitro systems and the accurate assessment and prediction of the relative risk of a clinically significant TDI-based drug interaction is an important aspect of drug discovery and development.
Traditionally, the in vitro evaluation of TDI has been conducted using human liver microsomes (HLM) and/or cDNA-expressed enzyme systems (Grimm et al., 2009). Assay designs vary and include several methods such as IC50 shift, progress curves, and kinetic determinations (reviewed in Grimm et al., 2009). HLM and recombinant systems provide a convenient matrix with which to assess in vitro TDI; however, they are likely not accurate representations of the in vivo situation. The use of primary human hepatocytes has been suggested to be a better representation in determining the kinetics of time-dependent inhibition (Zhao et al., 2005; McGinnity et al., 2006), especially when considering the prediction of in vivo drug-drug interactions (DDI) (Xu et al., 2009; Kirby et al., 2011). However, an in-depth assessment of experimental design and assay robustness, as well as their impact on in vivo DDI predictions, is still limited in these published studies.
Numerous groups have published on the prediction of the magnitude of the TDI effect clinically (Mayhew et al., 2000; Wang et al., 2004; Galetin et al., 2006; Obach et al., 2007), including several groups using advanced physiologically based pharmacokinetics (PBPK) modeling software packages such as Simcyp or Gastroplus (Rostami-Hodjegan and Tucker, 2004; Yang et al., 2006; Lukacova et al., 2010; Zhao et al., 2011). These simulations usually incorporate TDI kinetic parameters that were obtained from HLM experiments, and these predictions using HLM data often yield gross overpredictions of the magnitude of effect when compared with in vivo clinical data (Xu et al., 2009; Wang, 2010). The reasons for this overprediction are not clearly understood, but they may reflect inaccuracies in system parameters such as kdeg and a more complex interplay between metabolic activation and inactivation, protein degradation and synthesis, and/or hepatocellular concentrations because of active transport uptake or efflux. As mentioned previously, hepatocytes have been suggested as a more appropriate and reliable predictor of clinical TDI. In fact, the limited data published to date have shown improvements in the predictions when using hepatocyte TDI data for known time-dependent inhibitors; however, the heterogeneity in hepatocyte donors creates difficulties in any firm conclusions (Xu et al., 2009).
These studies describe the use of human hepatocyte TDI data for the prediction of DDI, with a focus on the experimental design and assay accessibility using pooled cryopreserved human hepatocytes in a high throughput manner. Three different experimental methods were evaluated and compared. The TDI kinetic parameters for four known CYP3A4 inactivators were generated using two of the evaluated hepatocyte methods and a conventional HLM approach. These data were then used for the prediction of DDI using Simcyp, a population PBPK-based DDI simulator (Jamei et al., 2009), and compared with known clinical DDI results. The accuracy of the predictions using the in vitro data from the different approaches was assessed. Overall, the predictions using human hepatocyte data were far better than the data generated in HLM. In addition, our studies show the importance of validating commercially available prediction tools with in vitro data from well designed experiments to enhance the accuracy of in vivo DDI risk assessment.
Materials and Methods
Chemical and Biological Reagents.
Midazolam maleate salt, 1′-hydroxymidazolam, diltiazem, erythromycin, verapamil, troleandomycin (TAO), RPMI 1640 medium, Percoll, trypan blue, NADPH, dimethyl sulfoxide, formic acid, and all high-performance liquid chromatography (HPLC)-grade solvents were from Sigma-Aldrich (St. Louis, MO). Cryopreserved human hepatocytes (lot no. X008001, 10-donor pool, mixed-gender, 5 million cells/vial), were from Celsis (Baltimore, MD). Fetal bovine serum and l-glutamine were from Invitrogen (Carlsbad, CA). Human liver microsomes, 50-donor pool, were from BD Gentest (Woburn, MA).
Preparation of Cryopreserved Human Hepatocytes.
Cryopreserved human hepatocytes were quickly thawed in a 37°C water bath with gentle shaking. Once thawed, the hepatocytes were rinsed using a Percoll solution (∼30% Percoll in RPMI 1640 medium with 5% fetal bovine serum and 2 mM l-glutamine). The cell suspension was centrifuged at 100 x g for 10 min. Supernatant was removed and discarded, and the hepatocyte pellet was resuspended in 2 ml of fresh incubation media (RPMI 1640 medium with 5% fetal bovine serum and 2 mM l-glutamine). The cell viability was determined using the trypan blue method. In all experiments, the cell viability was approximately 90%. Before the initiation of all experiments, hepatocyte suspensions were placed in a 37°C incubator maintained at 5% CO2 and 95% humidity for 30 min.
Optimization of the Human Hepatocyte TDI Method.
The TDI experiments were performed using three different protocols for the purpose of method development and comparison (Table 1). All incubations were conducted in a 96-well plate in duplicate with an initial equilibration at 37°C for 30 min, as stated above. Erythromycin (3, 10, and 30 μM), verapamil (3, 10, and 30 μM), and troleandomycin (0.1, 1, and 10 μM) were used in the method development and method comparisons. The CYP3A4 activity was determined based on 1′-hydroxymidazolam formation under the different assay conditions. Acetonitrile (ACN) was used in the preparation of the stock solutions for the inactivators and midazolam, and total solvent concentration was less than 1% (v/v) in the final incubations.
Assay conditions for human hepatocyte TDI methods
The wash method was adapted from Zhao et al. (2005) with minor modifications. In brief, inhibitors (25 μl) were added to each well containing 225 μl of prewarmed hepatocyte suspensions, resulting in a final volume of 250 μl and a cell density of 0.3 × 106 viable cells/ml. The hepatocytes were preincubated with and without inhibitor at 37°C with 500 rpm of orbital shaking for 0 to 60 min. At selected time points (0, 10, 30, and 60 min), 200 μl of the incubation was transferred into a 1.5-ml microcentrifuge tube and centrifuged for 5 min at 50g at room temperature. The supernatant (150 μl) was removed and discarded, and the cell pellets were resuspended with 150 μl of fresh incubation media for washing. This step was repeated two additional times. The hepatocyte pellets were then resuspended in 170 μl of fresh incubation media containing 38.7 μM midazolam. The cell suspension containing midazolam at a final concentration of 30 μM and cell density of 0.245 × 106 cells/ml were incubated for an additional 10 min at 37°C, and the reactions were terminated by the addition of 440 μl of ACN. Quenched reactions were centrifuged at 4000 rpm, and the supernatants were collected for further analysis using liquid chromatography/tandem mass spectrometry (LC-MS/MS).
The add method was conducted as such. The inhibitor (10 μl) was added to each well containing 90 μl of prewarmed hepatocyte suspension resulting in a final volume of 100 μl and cell density of 0.3 × 106 viable cells/ml. The preincubation was conducted in the same manner as for the wash method. At each time point (0, 10, 30, and 60 min), 10 μl of incubation media containing 330 μM midazolam was added to each well, for a final midazolam concentration of 30 μM, and cell density of 0.245 × 106 cells/ml. Cell suspensions were incubated for an additional 10 min at 37°C, and reactions were terminated by the addition of 220 μl of ACN. Post-termination reactions were processed and analyzed in a similar manner to the wash method.
The dilution method was initiated by the addition of 5 μl of inhibitor to each well containing 45 μl of prewarmed hepatocyte suspension, resulting in a final volume of 50 μl and cell density of 0.6 × 106 viable cells/ml. The preincubation was conducted in the same way for the wash and add methods. At selected time points (0, 10, 30, and 60 min), 200 μl of incubation media containing 37.5 μM midazolam was added to each well. This addition resulted in a 5-fold dilution from the original volume, a final midazolam concentration of 30 μM, and a cell density of 0.125 × 106 cells/ml. Upon this addition, hepatocyte suspensions were incubated for an additional 10 min at 37°C with gentle orbital shaking. The reactions were terminated and processed, and samples were analyzed in a similar manner to the other methods.
Determination of CYP3A4 TDI Kinetics in Human Hepatocytes.
After method optimization, the time-dependent inhibition kinetic parameters (KI and kinact) for four marketed drugs that are known as time-dependent inhibitors of CYP3A4—diltiazem (1–100 μM), erythromycin (3–300 μM), verapamil (1–30 μM), and TAO (0.1–10 μM)—were determined using the add and dilution methods. The detailed experimental conditions are listed in Table 1.
Determination of CYP3A4 TDI Kinetics in Human Liver Microsomes.
The TDI of verapamil, erythromycin, TAO, and diltiazem was determined in HLM as previously described with minor modifications (Ito et al., 2003). In brief, a preincubation mixture containing 1 mg/ml HLM, 100 mM sodium phosphate (pH 7.4), 1 mM EDTA, 3 mM magnesium chloride, and varying concentrations of the inhibitors were incubated at 37°C for 3 min. The inactivation phase (preincubation) was initiated by addition of NADPH at a final concentration of 1 mM. The preincubation was allowed to proceed for 0, 2, 6, 12, and 20 min. At each time point of preincubation, an aliquot was taken and added to the reaction mixture containing 10 μM midazolam, which resulted in a 10-fold dilution. This reaction was allowed to proceed for 5 min and then quenched by the addition of an equal volume of 0.1% formic acid in ACN. The mixtures were centrifuged at 15,000g for 10 min, and supernatants were collected for analysis using LC-MS/MS.
HPLC-MS/MS Analysis.
The analysis of 1′-hydroxymidazolam was performed on a Shimadzu LC-10ADVP HPLC (Shimadzu, Columbia, MD) coupled to a Sciex API 4000 triple quadrupole mass spectrometer with an electrospray ionization source (AB SCIEX, Foster City, CA). A BDS Hyperisil C18 50 × 2.1 mm column (Thermo Scientific, West Palm Beach, FL) was used for separation. The mobile phase consisted of 5 mM ammonium acetate in 0.1% formic acid in water (mobile phase A) and 0.1% formic acid in methanol/acetonitrile 50:50 (mobile phase B). The initial condition was set at 5% mobile phase B for 1 min, increased to 95% mobile phase B over 1 min, stayed at 95% mobile phase B for 2 min, and was finally brought back to the initial conditions in 0.1 min. The retention times for 1′-hydroxymidazolam and 7-hydroxycoumarin (internal standard) were 1.65 and 1.32 min, respectively. The total run time was 5 min and the flow rate was maintained at 0.45 ml/min. The tune parameters of the mass spectrometry detector and the scan function parameters, including cone voltages and collision energies, were optimized for detection of 1′-hydroxymidazolam and 7-hydroxycoumarin.
Data Calculation.
The enzyme activity of control incubations without inhibitors was set to 100% at each preincubation time point. The formation of 1′-hydroxymidazolam in incubations with different inhibitor concentrations at each time point was normalized to this control incubation. For each test inhibitor concentration, the remaining enzyme activity after preincubation was calculated as the percentage of control at the corresponding concentration [I] without preincubation. The natural log (ln) of the percentage remaining activity was plotted against the preincubation time. The slope (− kobs, observed rate) of each line was then calculated for the period of 0 min to the last linear time point of the preincubation phase. A negative value of kobs was considered as zero. The kinact (maximal inactivation rate) and KI (apparent inactivation constant) were obtained from the nonlinear fitting of eq. 1 to − kobs determined in HLM and hepatocytes using WinNonlin (version 5.2.1; Pharsight, Mountain View, CA).

Simcyp DDI Simulation.
The Simcyp population-based ADME simulator (version 10.0; Simcyp Limited, Sheffield, UK) was used to perform time-based simulations of 30 clinical drug-drug interaction studies. The Sim-Healthy Volunteer population within Simcyp was used with special trial designs, including number of subjects, age range, gender ratio, and dose regimen replicated based on information described in the respective publications (supplemental data). Thirty clinical studies involving interactions between the three inhibitors and nine substrates from the available compound library within Simcyp were simulated.
The input parameters including physiological; in vitro absorption, distribution, metabolism, and excretion; and pharmacokinetic parameters for the inhibitors and substrates were the default values in the Simcyp library unless otherwise stated. The key parameters affecting DDI predictions used by the simulator were described previously (Yang et al., 2006; Obach et al., 2007; Grimm et al., 2009; Jamei et al., 2009). The contribution of CYP3A4 (fm,CYP3A4) to the clearance of each substrate was calculated based on the enzyme kinetics (Vmax and Km) within Simcyp. The median values of fm,CYP3A4 for each substrate used in Simcyp simulations were 0.89 (alprazolam), 0.999 (cyclosporine), 0.95 (midazolam), 1.00 (nifedipine), 0.85 (sildenafil), 0.86 (simvastatin), 0.97 (triazolam), 0.09 (metoprolol), and 0.002 (theophylline). The contribution of the gut wall to the interaction was incorporated in the model as such: the fraction of gut metabolism (Fg) was predicted based on the nominal flow using the gut model (Qgut) within Simcyp (Yang et al., 2007). The median values of Fg defined in the model for each substrate were 0.98 (alprazolam), 0.65 (cyclosporine), 0.58 (midazolam), 0.69 (nifedipine), 0.55 (sildenafil), 0.13 (simvastatin), 0.66 (triazolam), 0.99 (metoprolol), and 1 (theophylline). The concentration-time profiles for each inhibitor (I) were simulated using a first-order absorption and one compartment distribution model with default pharmacokinetic parameters. The CYP3A4 turnover rate constant (Kdeg) of 0.0077 h − 1 (equivalent to t1/2 = 90 h) in liver and of 0.03 h − 1 (equivalent to t1/2 = 23 h) in gut were the default values and were used in the predictions. The TDI kinetic parameters (KI and kinact) determined in hepatocytes and HLM from this study were the only user-input parameters used. The free fraction in microsomes (fu,mic) and hepatocytes (fu,inc) were either used as supplied in the software (fu,mic) or from literature (fu,inc) (Xu et al., 2009). The values (fu,mic and fu,inc) were 0.90 and 0.89 for diltiazem, 0.972 and 0.936 for erythromycin, and 0.77 and 0.936 for verapamil, respectively.
The time-based DDI simulations of 30 clinical trials (supplemental data) were performed using the Simcyp TDI-based DDI model (Yang et al., 2006). The magnitude of the interaction was determined as the fold increase in the area under the curve (AUC; expressed as the AUC ratio) of a substrate in the presence and absence of an inhibitor. The simulated median AUC ratios were compared with the corresponding AUC ratios observed from clinical studies.
The comparison of predictability using the TDI kinetic parameters generated from different methods was conducted. The geometric mean-fold error (GMFE) that weights over- and underprediction equally was used to determine the prediction error. The root mean square error (RMSE) was calculated to measure the precision of the prediction, and the mean residual sum (MRS) was used to assess the bias between the subsequent methods.



Results
Comparison of Three Different TDI Methods.
The effects of the dilution of the inhibitor from the preincubation phase to the reaction phase were compared for each of the three methods. The wash method was intended to remove the inhibitor at the end of preincubation through the washing of the cells three times before the incubation phase. This entire process took approximately 20 min, and the final inhibitor concentration in the reaction medium was estimated to be approximately 18-fold lower than that in the preincubation. For the add (1.25-fold) and dilution methods (5-fold), there were no extra steps between the preincubation and reaction phases (Table 1).
Three compounds—verapamil, erythromycin, and TAO—were used to test the CYP3A4 inactivation in a 1-h preincubation with human hepatocytes. The percentage loss of CYP3A4 activity after this preincubation was calculated by comparing the 1′-hydroxymidazolam formation to that in control samples without preincubation and without inhibitor (time 0). The percentage loss of CYP3A4 activity in hepatocytes was comparable among the three methods for each of the inhibitors tested (Table 2), and the loss of CYP3A4 activity by each inhibitor was concentration dependent. However, it was found that even without preincubation, the CYP3A4 enzyme activity in the controls containing inhibitor lost a significant amount compared with samples without inhibitor. This loss of enzyme activity in the controls (without preincubation) was different among the three methods and was inhibitor dependent (Fig. 1). Without preincubation, there was an approximately 30% loss of CYP3A activity in the presence of verapamil (30 μM) using the add and dilution methods and more than 50% loss using the wash method (Fig. 1A). In addition, there was almost no loss of CYP3A activity in the presence of erythromycin (up to 30 μM) independent of the method used (Fig. 1B). In contrast, the wash method with TAO led to more than 70% loss of CYP3A activity at 10 μM without preincubation, whereas more than 80% enzyme activity remained when using either the add or wash methods (Fig. 1C).
Inactivation of CYP3A4 in human hepatocytes using three different TDI methods
Remaining CYP3A4 activity in controls (without preincubation) with inactivator verapamil (A), erythromycin (B), and troleandomycin (C and D) from different methods.
The significant loss of CYP3A activity without preincubation with TAO was investigated by comparing two modifications of the wash methods: 1) washing the preincubation mixture immediately after adding the inhibitor (no preincubation) and then adding midazolam after the wash, or 2) washing the preincubation mixture without adding the inhibitor and then simultaneously adding the inhibitor and midazolam. As shown in Fig. 1D, washing in the presence of TAO resulted in significant loss of CYP3A activity at 10 μM, whereas almost no loss of CYP3A activity was observed using the second wash method. This suggests that the loss of CYP3A activity was not caused during the reaction/incubation phase with midazolam but rather during the washing period.
CYP3A4 Time-Dependent Inactivation Kinetics.
Because of the significant inactivation during the wash in the wash phase, this method was not used for any further experiments. Thus, using the optimized add and dilution methods, the full TDI kinetic profiles for diltiazem, erythromycin, verapamil, and TAO were determined. The time- and concentration-dependent inhibition of CYP3A4 activity was assessed based on 1′-hydroxymidazolam formation using data analyses described under Materials and Methods, and the fitted curves of inactivation rate (kobs) versus inhibitor concentration are shown in Fig. 2. The KI and kinact generated from each assay and the calculated kinact/KI ratios are presented in Table 3.
CYP3A4 inactivation in human hepatocytes determined using the add and dilution methods, respectively, by diltiazem (A1 and A2), erythromycin (B1 and B2), TAO (C1 and C2), and verapamil (D1 and D2). The inactivation rate constant (kobs) was plotted against concentrations of each inactivator, and the maximal inactivation rate (kinact) and the inactivator concentration (KI) that reaches half of the kinact were determined by nonlinear regression analysis.
TDI kinetic parameters generated in human hepatocytes and human liver microsomes
The TDI kinetic parameters generated using pooled human hepatocyte suspensions were reproducible between two replicate assays using either the add or dilution method (Table 3). The KI and kinact values obtained from the dilution method tended to be slightly higher than that from the add method for diltiazem, verapamil, and troleandomycin, but not for erythromycin. However, the ratios of kinact to KI (inactivation efficiency) obtained using the add and dilution methods were very similar for all four inhibitors (Table 3).
When comparing the TDI kinetic parameters in HLM generated in-house, the KI values generated in human hepatocytes for diltiazem (6.29 and 8.91 μM from add and dilution method, respectively), erythromycin (60 and 67.9 μM), and troleandomycin (3 and 3.44 μM) were approximately 5-, 12-, and 8-fold higher than that in HLM, respectively. The kinact values for diltiazem (0.019 and 0.023 min − 1 from add and dilution method, respectively) and erythromycin (0.081 and 0.079 min − 1) in hepatocytes were similar to that determined in HLM, whereas the kinact values for verapamil (0.020 and 0.027 min − 1) in hepatocytes were approximately 3-fold lower than the kinact in HLM, and for troleandomycin (0.24 and 0.36 min − 1) was roughly 4-fold higher than that in HLM. However, the overall inactivation efficiency (kinact/KI ratio) obtained from hepatocytes was consistently lower than that determined in HLM for all four inactivators (Table 3).
In Vivo DDI Prediction Using Simcyp.
To assess the accuracy of in vivo DDI predicted using the in vitro TDI kinetic data generated from human hepatocytes and HLM, 30 clinical drug-drug interaction studies involving diltiazem, erythromycin, and verapamil were simulated using Simcyp. Most victim drugs used in the interaction studies were known sensitive CYP3A4 substrates, with fmCYP3A4 > 85% (Simcyp-calculated median value based on Km and Vmax), except for metoprolol and theophylline, which were considered as negative controls for CYP3A4 TDI. The DDI predicted using in vitro TDI data generated from human hepatocytes and HLM are listed and compared with the clinically observed interactions (see supplemental data).
The DDI predicted using in vitro TDI data generated in HLM generally overestimated the severity of the observed DDI (25 of 30 interaction pairs). All in vivo interactions (11 of 11) involving erythromycin were overpredicted (i.e., the interaction predicted was more severe than the actual situation [supplemental data; Fig. 3A]). Nearly 50% (13 of 25) of the overpredicted DDI were beyond a 2-fold range of the actual value. In contrast, most (28 of 30) of the predicted DDI using hepatocyte TDI data were within 2-fold of the observed clinical DDI (Fig. 3, B and C). The DDI predicted using hepatocyte TDI data from two methods (add method and dilution method) were very similar and consistent (R2 = 0.947).
Correlations of observed versus predicted DDI (AUC ratio) from TDI kinetic data generated using the HLM method (A), the hepatocyte add method (B), and the hepatocyte dilution method (C).
The accuracy of the DDI predictions using TDI kinetic parameters from different in vitro assays were analyzed and are summarized in Table 4. The predictions using HLM TDI data yielded greater error than those using hepatocyte TDI data for erythromycin, diltiazem, and verapamil, with the greatest error for erythromycin-based DDI prediction (GMFE 2.75). A relatively high degree of variability was observed between predicted DDI using HLM data and the observed values, especially for erythromycin (RMSE 10.8) and diltiazem (RMSE 4.15). In contrast, the variability (RMSE 1.01–1.22) between predicted DDI using hepatocyte data and observed DDI was much smaller. The predictions using hepatocyte TDI data had no significant bias for verapamil and erythromycin and some level of underprediction for diltiazem, but there was a consistent overprediction for all three inhibitors observed (MRS 1.6–7.7) when HLM data were used.
Summary of DDI prediction accuracy using TDI kinetic parameters (Ki, kinact) generated from HLM method and HH methods
Discussion
Hepatocytes are considered to be the closest recapitulation in an in vitro system of liver function in vivo retaining the full spectrum of phase I/II enzymes, transporters, and cofactors at physiological levels. Cryopreserved human hepatocyte suspensions in 24-, 48-, or 96-well plates are routinely used to determine in vitro metabolic stability for the prediction of in vivo metabolic clearance (Jouin et al., 2006). To date, only limited studies have been conducted using cryopreserved hepatocytes to evaluate in vitro CYP3A4 TDI and its impact on the prediction of in vivo DDI (Xu et al., 2009). Zhao et al. (2005) were the first group to publish on the use of cryopreserved hepatocytes to assess CYP3A4 TDI for six known inactivators using what would be described in this work as the wash method; however, TDI kinetic data were not determined, thus the prediction of in vivo interactions using hepatocyte TDI data were not conducted. A recent study investigated the prediction of TDI-based DDI using TDI kinetic data generated in hepatocytes with a modified wash method (Xu et al., 2009). However, the KI and kinact were generated using hepatocytes from two single donors, and the feasibility of different experimental designs was not assessed. The work presented here demonstrates the feasibility of using pooled hepatocytes for the in vitro assessment of TDI by comparing three different experimental designs and assessing the impact of the TDI kinetic values derived from the different hepatocyte methods on the prediction of in vivo DDI.
The continued inactivation of enzyme by inhibitor during the incubation phase is a major concern in generating accurate in vitro TDI data. A 10- to 20-fold dilution is commonly used in TDI assays using HLM to minimize further enzyme inactivation and/or competitive inhibition during the incubation phase; however, it is technically challenging to do this in hepatocyte-based assays. In this study, three methods with different approaches in dilution were compared using pooled human hepatocytes in a 96-well plate format. It was found that at a given inhibitor concentration the wash method led to greater loss of CYP3A4 activity in controls containing inhibitors without preincubation (Fig. 1). This phenomenon was more pronounced in the presence of TAO, a potent time-dependent and competitive inhibitor. Although the wash method intends to remove the inhibitor from the preincubation mixture, it appeared that it did not prevent ongoing inactivation during the washing and centrifugation of the cells (Fig. 1D). The extra steps prolong the presence of the inhibitor in the cell incubation, which resulted in even more loss of CYP3A4 activity, possibly because of continued inactivation and/or competitive inhibition. In fact, the add and dilution methods controlled for this loss of enzyme activity far better than the wash method (Fig. 1).
Theoretically, insufficient dilution of the inhibitor from the preincubation to incubation phase may lead to an overestimation of TDI. However, the significant loss of enzyme activity in controls using the wash method could lead to an inaccurate determination of the loss of enzyme activity at sequential preincubation time points, which could underestimate time-dependent inactivation. This was observed in studies with TAO, in which the kobs (slope of enzyme activity against preincubation time) reached a plateau rapidly as the inhibitor concentration increased. The remaining enzyme activity was extremely low in the controls and left a very narrow range for further enzymatic loss during the preincubation phase (i.e., a decrease in the dynamic range of the assay). In addition, the wash method is far more labor-intensive and has the potential for the introduction of more error because of more sample handling when compared with either the dilution or add methods.
Because the wash method provided suboptimal experimental conditions and parameters, only the add and dilution methods were further optimized for the evaluation of CYP3A4 TDI kinetics. The kinact and KI obtained from the dilution method tended to be slightly higher than those from the add method. However, overall the 4-fold difference in dilution factor appeared to have no significant impact on either the KI or kinact of the four time-dependent inhibitors tested. The dilution efficiency in hepatocyte incubations may not be the same as that observed in HLM incubations because the inhibitor is present within the cell. The lower dilution used in the add method did not result in an apparent higher rate of inactivation. This may be due to a multitude of factors, including the low inactivation potency and/or apparent weaker competitive inhibition in hepatocytes, the prevention of inactivation due to the presence of saturating levels of substrate (Ghanbari et al., 2006), and/or the normalization of the data at each time point to the control values. A higher dilution may be beneficial for potent inhibitors that would need to be tested at higher concentrations in hepatocytes. In addition, if the metabolite of an inactivator was a strong competitive inhibitor of the enzyme metabolizing the probe substrate, the formation of metabolite from probe substrate is likely to be less with insufficient dilution, which may result in an apparently higher rate of inactivation.
The known time-dependent inhibition of CYP3A4 in HLM by diltiazem, erythromycin, verapamil, and TAO was confirmed in these hepatocyte studies. The time-dependent inhibition of these inhibitors in hepatocytes was approximately 2- to 10-fold lower than that in HLM when comparing the ratio of kinact to KI. The reduced inactivation potency was mainly due to the higher KI values for diltiazem (4–6-fold), erythromycin (11–13-fold), and TAO (7–8-fold), and to a lesser extent a lower kinact [e.g., verapamil (3–4-fold)]. The higher KI in hepatocytes when compared with those in HLM remained, although the values are corrected for the unbound fraction in both systems using previously published parameters from Xu et al., 2009 (data not shown). The apparent differences between the TDI kinetic values measured using hepatocytes and those for HLM could be attributed to factors governing the accessibility to the enzyme and/or the concentration of the inactivators at the target enzyme in hepatocytes (including involvement of transporters, metabolic consumption through phase II enzymes, and/or nonspecific binding to cellular constituents) that are not accounted for in HLM.
The apparent discrepancy of 2- to 10-fold difference in TDI potency between hepatocytes and HLM affected the accuracy of in vivo DDI predictions. As shown in the correlation analysis in Fig. 3, DDI predictions using TDI parameters generated in hepatocytes were consistently lower than those predicted using parameters from HLM. In general, compared with hepatocyte-based predictions, the HLM-based predictions overestimated the TDI-based DDI observed in vivo, which is consistent with previously published data (Xu et al., 2009; Wang, 2010) that were obtained using similar models. In contrast, approximately 90% of the predictions using hepatocyte TDI data were within 2-fold of the observed clinical DDI. A likely explanation for the better predictability using hepatocyte TDI parameters could be that the effective inactivator concentrations that drive the potency inside of the cell could be significantly lower compared with the nominal concentration added in the HLM.
Several factors should be considered in the prediction of TDI-based DDI. In addition to the TDI kinetic parameters that could be highly variable depending on the in vitro system used, as well as the experimental design, the rate of enzyme degradation (kdeg) in vivo is often associated with great uncertainty, and no consensus has been reached because of the impossibility of direct measurement in humans. As a result, a relatively wide range of kdeg values has been reported for CYP3A4 (t1/2 = 23–140 h; Yang et al., 2008; Grimm et al., 2009). In the work presented here, a kdeg value for CYP3A4 of 0.0077 h − 1 (t1/2 = 90 h), which is the default in the simulation software, was used. A recent study (Wang, 2010) has revealed that a better prediction is obtained when a kdeg value of 0.03 h − 1 (t1/2 = 23 h) is used in the model with the HLM-derived TDI data default in Simcyp. This approach suggests that by altering a physiological parameter, a better prediction could be achieved. However, this approach may not be applied universally to increase the reliability of the prediction. More importantly, it neglects other physiologically important effects such as competing metabolic pathways of the inhibitor and/or active transport processes that combined may affect cellular inhibitor concentration and thereafter the TDI liability.
Other parameters, such as the fm(CYP) and Fg of the substrate and free fraction of inhibitor concentration, can also affect the prediction. The effect of these parameters were assessed in Simcyp and incorporated in the DDI prediction. A good understanding of the source of uncertainty of these parameters and its impact on the prediction is also important. In addition, the accurate prediction of DDI involving competitive and time-dependent inhibition by the parent drug and its metabolite would require further development of a mechanistic PBPK model.
This study investigated the impact of in vitro TDI kinetic data generated from different systems on the DDI prediction while the other parameters were kept consistent as defined in the Simcyp model. The pooled HLM (50-donor) and cryopreserved HH (10-donor) were used in the assay to minimize the variation of enzyme activity among individuals. Still, the effect of variability associated with different systems on the prediction should be considered carefully. Overall, because hepatocytes represent the closest physiological in vitro system to the liver, they can be considered a more relevant system, and TDI determined in hepatocytes may be used for a true risk assessment of the in vivo TDI-based DDI. Although the HLM assay is useful to identify issues and screen out liabilities, the cryopreserved hepatocyte assay may be more reliable for the prediction of the magnitude of in vivo DDI.
In conclusion, this study demonstrates the use of pooled cryopreserved hepatocytes as an alternative model for the determination of in vitro TDI kinetics. The potency of enzyme inactivation in hepatocytes is apparently weaker than that in HLM. This may represent a physiological effect that is possibly due to a lower intracellular concentration of the inhibitor in the hepatocytes. The prediction of DDI using hepatocyte TDI data suggests that pooled hepatocyte suspension incubations are a more robust and predictive system to evaluate a compound's potential to cause in vivo TDI-based DDI. Combined results from our studies suggest that the use of properly designed hepatocyte TDI experiments are valuable for the risk assessment of TDI-based drug-drug interaction and will reduce the number of inaccurate conclusions when only HLM-based predictions are used.
Authorship Contributions
Participated in research design: Chen, Monshouwer, and Fretland.
Conducted experiments: Liu.
Performed data analysis: Chen.
Wrote or contributed to the writing of the manuscript: Chen and Fretland.
Acknowledgments
We thank Wen-Chen Hsu for valuable advice and training in hepatocyte experimental systems.
Footnotes
Article, publication date, and citation information can be found at http://dmd.aspetjournals.org.
doi:10.1124/dmd.111.040634.
↵
The online version of this article (available at http://dmd.aspetjournals.org) contains supplemental material.
-
ABBREVIATIONS:
- TDI
- time-dependent inhibition (or inhibitor)
- DDI
- drug-drug interactions
- ACN
- acetonitrile
- HLM
- human liver microsomes
- HH
- human hepatocytes
- LC-MS/MS
- liquid chromatography/tandem mass spectrometry
- TAO
- troleandomycin
- AUC
- area under the curve
- PBPK
- physiologically based pharmacokinetics
- HPLC
- high-performance liquid chromatography
- FBS
- fetal bovine serum
- GMFE
- geometric mean-fold error
- RMSE
- root mean square error
- MRS
- mean residual sum
- KI
- inhibitor concentration that supports half the maximal rate of inactivation
- kinact
- maximal rate of enzyme inactivation.
- Received May 17, 2011.
- Accepted August 11, 2011.
- Copyright © 2011 by The American Society for Pharmacology and Experimental Therapeutics