Advertisement
Research ArticleArticle

Evaluation of Calibration Curve–Based Approaches to Predict Clinical Inducers and Noninducers of CYP3A4 with Plated Human Hepatocytes

J. George Zhang, Thuy Ho, Alanna L. Callendrello, Robert J. Clark, Elizabeth A. Santone, Sarah Kinsman, Deqing Xiao, Lisa G. Fox, Heidi J. Einolf and David M. Stresser
Drug Metabolism and Disposition September 2014, 42 (9) 1379-1391; DOI: https://doi.org/10.1124/dmd.114.058602

Abstract

Cytochrome P450 (P450) induction is often considered a liability in drug development. Using calibration curve–based approaches, we assessed the induction parameters R3 (a term indicating the amount of P450 induction in the liver, expressed as a ratio between 0 and 1), relative induction score, Cmax/EC50, and area under the curve (AUC)/F2 (the concentration causing 2-fold increase from baseline of the dose-response curve), derived from concentration-response curves of CYP3A4 mRNA and enzyme activity data in vitro, as predictors of CYP3A4 induction potential in vivo. Plated cryopreserved human hepatocytes from three donors were treated with 20 test compounds, including several clinical inducers and noninducers of CYP3A4. After the 2-day treatment, CYP3A4 mRNA levels and testosterone 6β-hydroxylase activity were determined by real-time reverse transcription polymerase chain reaction and liquid chromatography–tandem mass spectrometry analysis, respectively. Our results demonstrated a strong and predictive relationship between the extent of midazolam AUC change in humans and the various parameters calculated from both CYP3A4 mRNA and enzyme activity. The relationships exhibited with non-midazolam in vivo probes, in aggregate, were unsatisfactory. In general, the models yielded better fits when unbound rather than total plasma Cmax was used to calculate the induction parameters, as evidenced by higher R2 and lower root mean square error (RMSE) and geometric mean fold error. With midazolam, the R3 cut-off value of 0.9, as suggested by US Food and Drug Administration guidance, effectively categorized strong inducers but was less effective in classifying midrange or weak inducers. This study supports the use of calibration curves generated from in vitro mRNA induction response curves to predict CYP3A4 induction potential in human. With the caveat that most compounds evaluated here were not strong inhibitors of enzyme activity, testosterone 6β-hydroxylase activity was also demonstrated to be a strong predictor of CYP3A4 induction potential in this assay model.

Introduction

The potential for new drug candidates to exhibit drug-drug interactions (DDI) is a significant concern during the drug development process. Because metabolism by cytochrome P450 (P450) enzymes is often a major elimination pathway, small-molecule drug candidates are evaluated for P450 inhibition or induction at various stages of development. CYP3A4 is 15–30% of hepatic P450 content (Shimada et al., 1994; Ohtsuki et al., 2012) and is estimated to account for about half of oxidations in drugs undergoing P450-mediated clearance (Wienkers and Heath, 2005). Therefore, this enzyme is critical to evaluate as a mediator of DDI.

Prediction of human DDI on the basis of in vitro data is another important goal during the early stages of drug development. Outcomes of these predictions may ultimately determine whether clinical DDI studies are conducted. Various models and frameworks have been proposed for induction prediction and these have been recently reviewed (Almond et al., 2009; Fahmi and Ripp 2010; Einolf et al., 2014). These include calibration curve–based approaches, mathematical or mechanistic static models, and physiologically-based pharmacokinetic models. Calibration curve–based models can be developed by comparing the observed clinical change in area under the curve (AUC) of a probe substrate drug (such as midazolam for CYP3A4) for a set of known inducers/noninducers of the enzyme of interest, with various in vitro induction potency parameters, such as relative induction score (RIS) (Ripp et al., 2006), AUC/F2 (Kanebratt and Andersson 2008), or Cmax/EC50 (Fahmi and Ripp 2010) obtained from specific lots (donors) of cryopreserved hepatocytes. These models, as well as others (Kato et al., 2005; Shou et al., 2008, Fahmi et al., 2009), can be used to evaluate induction potential and risk of a clinical DDI. Recent guidance from the US Food and Drug Administration (FDA, 2012) and European Medicines Agency (EMA, 2013) suggest options for evaluating induction potential, ranging from simple, conservative models (“R3” and a predefined fold-induction threshold) to more complex models as mentioned earlier (e.g., mechanistic static models, physiologically-based pharmacokinetic models, RIS). Both guidance documents advocate use of mRNA data obtained using human hepatocytes. In addition, the documents generally recommend using donor lots that have been previously characterized with a sufficient number of clinical inducers and noninducers (EMA basic method being the exception).

Several in vitro test systems have been used for assessment and prediction of CYP3A4 induction potential in humans by new drug candidates. These systems include primary cultures of cryopreserved human hepatocytes (Shou et al., 2008; McGinnity et al., 2009; Fahmi et al., 2010), human hepatocyte-like cell lines such as Fa2N-4 (Ripp et al., 2006; McGinnity et al., 2009), HepaRG (Kanebratt and Andersson 2008; McGinnity et al., 2009), and more recently a stably expressed human PXR cell line derived from HepG2 (Fahmi et al., 2012). To date, human hepatocyte cultures have been considered the gold standard for in vitro induction assessment and are currently “preferred” by regulatory agencies.

The present work describes a direct comparison of several induction parameters (RIS, R3, Cmax/EC50, and AUC/F2) generated with a set of clinical inducers and noninducers and using human hepatocytes as in vitro system, to predict the in vivo CYP3A4 induction potential from a calibration curve. In this approach, calibration curves were constructed by plotting various parameters versus the change in clinical probe AUC observed in vivo. The evaluations were conducted using endpoints of mRNA and testosterone 6β-hydroxylase activity generated using either total or unbound plasma Cmax in the models.

Materials and Methods

Materials and Reagents.

A set of 20 compounds comprising primarily clinical inducers and noninducers were evaluated in vitro at concentrations selected on the basis of previous publications (Ripp et al., 2006; Kanebratt and Andersson 2008; McGinnity et al., 2009; Fahmi et al., 2010) (Table 1) and experience within this laboratory. Dimethyl sulfoxide (DMSO), testosterone, acetonitrile, ethanol, and all test drugs were purchased from Sigma-Aldrich (St. Louis, MO) and were of the highest grade available. Inducible cryopreserved human hepatocytes, Corning hepatocyte culture medium, 6β-hydroxytestosterone, 6β-hydroxytestosterone-[D7], Corning high viability cryohepatocyte recovery kits, and collagen type I-coated 96-well plates were obtained from Corning Life Science (Tewksbury, MA). Gentamicin was obtained from Lonza (Walkersville, MD). Fungizone, l-glutamine, and Dulbecco’s phosphate buffered saline were Gibco brand (Life Technologies, Grand Island, NY). The RNeasy 96 kit and DNase I were from Qiagen (Valencia, CA). Reverse transcription kit, two-step TaqMan PCR Master Reaction Mix, and primers/probes were obtained from Applied Biosystems (Foster City, CA).

View this table:
TABLE 1

EC50, Emax, F2, and percentage of positive control (% PC) rifampicin response obtained with CYP3A4 mRNA expression induction data from three lots of human hepatocytes

Human Hepatocyte Culture and Treatment.

Inducible cryopreserved human hepatocytes (lots 295, 312, and 318) were rapidly thawed and plated at a density of 0.6 × 106 viable cells/ml (100 μl/well) in collagen type-I-coated 96-well plates using high viability cryohepatocyte recovery kits. After approximately 4 hours, the plating medium was replaced with 100 μl of hepatocyte culture medium supplemented with 2 mM l-glutamine, 50 μg/ml gentamicin, and 0.75 μg/ml fungizone, and the cultures were maintained overnight. Hepatocyte cultures were treated for 2 days with 0.1% DMSO (negative control) and test drugs at eight concentrations each with the exception of primaquine, methotrexate, and digoxin, for which three or four concentrations were tested. All incubations were performed in triplicate. The test concentration range is shown in Table 2.

View this table:
TABLE 2

EC50, Emax, F2, and percentage of positive control (% PC) rifampicin response obtained with CYP3A4-mediated testosterone 6β-hydroxylation induction data from three lots of human hepatocytes

In Situ CYP3A4 Activity Measurement.

The testosterone 6β-hydroxylase activity assay was performed essentially as described by Zhang et al. (2010). Briefly, after treatment, hepatocyte cultures were washed with culture medium and incubated with 100 μl of culture medium containing CYP3A4 probe substrate testosterone at a concentration of 200 μM for 30 minutes. The reactions were stopped by combining an aliquot from each well with acetonitrile containing internal standard 6β-hydroxytestosterone-[D7]. The amount of metabolite formed was determined by liquid chromatography–tandem mass spectrometry using an API-4000 mass spectrometer. The culture plates were stored at –80°C until total RNA isolation.

Determination of Test Drug Concentrations in Incubation Medium.

At the end of the second day of treatment, the incubation medium from lot 295 was collected and combined with acetonitrile containing internal standard labetalol. The relative concentrations of test drug remaining in the incubation medium were quantitated by liquid chromatography–tandem mass spectrometry using an API-4000 mass spectrometer.

Total RNA Isolation and Real-Time Reverse-Transcription Polymerase Chain Reaction Analysis.

Total RNA was isolated from cells using the RNeasy 96 kit according to instructions provided by the manufacturer. The mRNA expression for CYP3A4 and the housekeeping gene β-actin was determined by Taqman real-time, reverse-transcription polymerase chain reaction (PCR) methods using the two-step assay protocol. First, a reverse transcription assay was performed using a GeneAmp PCR System 9700 (Applied Biosystems) with equal volumes of total RNA and the reverse-transcription master mixture. For the PCR assay, a PCR master mixture of reagents was prepared and a 20-μl aliquot of the master mixture was transferred to a 96-well optical reaction plate, followed by the addition of 5 μl of acquired cDNA to the appropriate wells. The PCR amplification was performed and the transcription was determined using an 7300 Real-Time PCR System.

Data Calculation.

The test drug concentrations (micromoles) remaining in the incubation medium after the 2-day treatment and the catalytic activity for CYP3A4 in hepatocytes were calculated using standard curves. The fold induction for activity data were calculated as follows: (enzyme activity of test drug)/(mean of enzyme activity of negative control). The fold induction for mRNA data was determined using the calculation of 2–ΔΔCT (Livak and Schmittgen, 2001). Percentages of positive control response for both activity and mRNA were calculated as follows: (mean of observed maximal fold – 1)/(mean of observed maximal fold by rifampicin – 1).

Curve Fitting.

To estimate EC50, the maximum response (Emax), and F2 values, concentration-response fold induction data were fitted to a sigmoid dose-response one-site-fit model (4 Parameter Logistic Model, Model 205) with XLfit (ID Business Solutions, Emeryville, CA) according to eq. 1:Embedded Image(1)where Emin is the baseline of the curve, Emax is the maximum effect, EC50 is the concentration achieving 50% of Emax, d is the slope of the curve, and C is the drug concentration. An additional parameter, F2, which is the drug concentration that causes a 2-fold increase of Emin, was also calculated. The curve fitting was conducted using the following acceptance criteria and conditions: 1) Data points were excluded from curve fitting when toxicity, insolubility, and inhibition (enzyme activity only) were apparent or when the coefficient of variation of replicate values was >40% after removing the offending outlier in the original set of triplicate samples); 2) curve fitting data were not used when R2 of the fit was <0.85; 3) when no apparent plateau was observed after the above-mentioned conditions were taken into consideration, the Emax was constrained to the observed maximal fold to prevent extrapolation of the curve fit beyond measured data, and EC50 was then obtained from the fitted curves; 4) EC50 and Emax were determined only when fold induction was at least 1.4-fold and the response was concentration-dependent.

The RIS and R3 values were calculated according to eqs. 2 and 3, following to the European Medicines Agency (EMA, 2013) and Food and Drug Administration (FDA, 2012) guidance documents, respectively.Embedded Image(2)Embedded Image(3)where [I] is total or unbound systemic plasma Cmax, and d is a scaling factor assumed to be 1 (FDA, 2012). Although R3 as defined in FDA guidance does not permit use of unbound Cmax as the value of [I], as part of our investigation, we elected to examine the effect of both total and unbound Cmax on model outcomes.

The Cmax (total and unbound)/EC50 and AUC/F2 were also calculated, where AUC is the in vivo exposure of the interacting drug, represented by the area under the plasma concentration curve over the time course (Table 3).

View this table:
TABLE 3

Clinical pharmacokinetic and drug-drug interaction data

Preparation of Calibration Curves.

A set of nine clinical inducers and clinical noninducers with known midazolam AUC changes after single dosing and clinical pharmacokinetic data were used for the preparation of calibration curves. The compounds included three strong inducers: rifampicin (with four clinical study data points), phenytoin, and carbamazepine; four moderate/weak inducers: troglitazone, terbinafine, pleconaril, and pioglitazone; and two clinical noninducers: nifedipine and clotrimazole (Table 3, as indicated with asterisks). Flumazenil was excluded from the calibration curves because very weak induction was observed in only one of the three donors (mRNA only) and its extremely low Cmax would have yielded a data point far removed from the range of the other points on the calibration curve. The remaining interacting drugs examined here were not used to generate the calibration curves. This is because we used only those compounds for which associated clinical data were obtained with the well established CYP3A4 probe midazolam as the victim drug (see Table 3). In vitro data generated for those compounds with non-midazolam clinical data were then evaluated against the curve generated with the aforementioned nine compounds. The calibration curves were constructed with the observed midazolam AUC change against calculated induction parameters (RIS, R3, Cmax/EC50, and AUC/F2) using eq. 4:Embedded Image(4)where A is the baseline of the curve constrained to 0%, B is the maximum AUC change constrained to ≤100%, C is the values of induction parameters x (RIS, R3, Cmax/EC50, and AUC/F2) achieving 50% of AUC, change and d is the slope of the curve. The cut-off values for a positive inducer were defined as the induction parameter values leading to a 20% decrease in predicted midazolam AUC change (FDA, 2012). The analysis included determination of the 95% confidence interval for the cut-off values (95% probability that the predicted cut-off values will occur) and correlation coefficient (R2) for goodness of fit. Statistical parameters were determined using the Statistics Designer function with in XLfit software.

Comparison of Model Predictability.

To compare the prediction accuracy of each model, the root mean square error (RMSE) was calculated as described in eq. 5, with greater accuracy being shown by the lower RMSE. The fold changes of predicted DDI and the observed DDI were calculated as AUCinduced/AUCcontrol. The bias of the prediction models was determined by the geometric mean fold error (GMFE) in eq. (6), which weighs over- and underpredictions equally. The lowest GMFE value would represent the lowest prediction bias.

Embedded Image(5)Embedded Image(6)

Results

Concentration-Dependent Induction Response of CYP3A4 mRNA and Activity in Human Hepatocytes.

Hepatocytes from lots 295 (aged 41, female and Caucasian), 312 (aged 56, male and Caucasian), and 318 (aged 58, male and African American) were treated for 2 days with a medium change and compound replenishment after 24 hours. Both CYP3A4 mRNA levels and catalytic activities were measured. The parameters EC50, Emax, and F2 were determined from the concentration-response curves, and percentage of positive control response was also determined. Overall, EC50 values obtained from both CYP3A4 mRNA and enzyme activity were similar (e.g., within 3-fold) within and between donors, with some notable exceptions (Tables 1 and 2). For example, EC50 for rifampicin was 0.12 and 0.18 for mRNA and enzyme activity, respectively, for lot 295, but was 1.4 and 1.1 μM for the same endpoints, respectively, for lot 312. Emax values obtained from mRNA data in general were greater than those from the activity results. The weak clinical inducers sulfinpyrazone and probenecid produced a potent induction response of CYP3A4 mRNA and/or activity and the response did not reach plateau at the highest concentration within any of the lots (data not shown). Flumazenil caused no induction for both mRNA and activity in lots 295 and 318; however, a slight induction response for CYP3A4 mRNA was observed in lot 312 at the high end of the concentration-response curve. As expected, most compounds previously shown to be inducers in vivo and in vitro caused a greater than 2-fold induction over vehicle control and exhibited concentration-dependence, the criteria to demonstrate a positive induction result as described in the EMA guidance (EMA, 2013). A few moderate and/or weak clinical inducers, such as pleconaril and pioglitazone, failed to reach cut-off values for either mRNA or enzymatic activity for some of the three lots of hepatocytes. Quinidine at the concentration range of 0.11–250 μM caused an induction of CYP3A4 mRNA in two of the three lots but not for enzyme activity for all lots. No induction of CYP3A4 mRNA and activity was observed for primaquine, methotrexate, and digoxin at the concentrations tested in any of the three lots.

Comparison of Calibration Curves for RIS, R3, Cmax/EC50, and AUC/F2.

To examine the relationship between induction data generated in vitro and data observed in clinical studies (Table 3), the RIS, R3, and Cmax/EC50 values were calculated on basis of both total and unbound Cmax (Supplemental Tables 1–3; Table 5). The calibration curves were then prepared with the percentage observed AUC changes of midazolam as a function of parameters RIS, R3, Cmax/EC50, and AUC/F2 as shown in Figs. 1–4 (see Supplemental Figs. 1–8 for lots 312 and 318). The proposed cut-off values corresponding to a 20% of predicted midazolam AUC change in vivo, 95% confidence interval for the cut-off values, and correlation coefficient R2 for the calibration curves are summarized in Table 4. Overall, excellent correlation between the induction parameters and observed midazolam AUC changes was obtained with the choice of model with reasonable 95% confidence intervals and R2 values (0.84–0.995 for mRNA and 0.78–0.99 for activity). Cut-off values were within 3-fold for both mRNA and activity across all three lots for RIS, R3, and Cmax/EC50. Relative to other parameters, the cut-off values for AUC/F2 appeared to vary more across all three lots of hepatocytes.

Fig. 1.
Fig. 1.

Calibration curve of midazolam AUC change (%) as a function of RIS, obtained from induction data for CYP3A4 mRNA [(A) on the basis of Cmax-t; (B) on the basis of Cmax-u], and testosterone 6β-hydroxylase activity [(C) on the basis of Cmax-t; (D) on the basis of Cmax-u] in lot 295 human hepatocytes. The RIS values were calculated on the basis of both total and unbound Cmax. The interacting drugs used for the curve fitting are shown in Table 3. See Supplemental Data for lots 312 and 318.

Fig. 2.
Fig. 2.

Calibration curve of midazolam AUC change (%) as a function of R3, obtained from induction data for CYP3A4 mRNA [(A) on the basis of Cmax-t; (B) on the basis of Cmax-u] and testosterone 6β-hydroxylase activity [(C) on the basis of Cmax-t; (D) on the basis of Cmax-u] in lot 295 human hepatocytes. The R3 values were calculated on the basis of both total and unbound Cmax. The interacting drugs used for the curve fitting are shown in Table 3. See Supplemental Data for lots 312 and 318.

Fig. 3.
Fig. 3.

Calibration curve of midazolam AUC change (%) as a function of Cmax/EC50, obtained from induction data for CYP3A4 mRNA [(A) on the basis of Cmax-t; (B) on the basis of Cmax-u] and testosterone 6β-hydroxylase activity [(C) on the basis of Cmax-t; (D) on the basis of Cmax-u] in lot 295 human hepatocytes. The Cmax/EC50 values were calculated on the basis of both total and unbound Cmax. The interacting drugs used for the curve fitting are shown in Table 3. See Supplemental Data for lots 312 and 318.

Fig. 4.
Fig. 4.

Calibration curve of observed midazolam AUC change (%) as a function of AUC/F2, obtained from induction data for CYP3A4 mRNA (A) and testosterone 6β-hydroxylase activity (B) in lot 295 human hepatocytes. The interacting drugs used for the curve fitting are shown in Table 3. See Supplemental Data for lots 312 and 318.

View this table:
TABLE 4

Summary of cut-off values, confidence interval, and correlation coefficients (R2) of calibration curves for RIS, R3, Cmax/EC50, and AUC/F2

RIS, R3, Cmax/EC50 were calculated using both Cmax-t and Cmax-u. Cut-off values for RIS, R3, Cmax/EC50 and AUC/F2 were defined as that corresponding to 20% predicted AUC change.

Assessment of R3 Cut-Off Value in Prediction of CYP3A4 Inducers.

The R3 values for each interacting drug were calculated on the basis of both total and unbound Cmax and are presented in Table 5. These values were compared with 0.9, a cut-off value for a likely inducer in vivo as proposed in the FDA draft guidance (FDA, 2012). As shown in Table 5, R3 values calculated with both total systemic plasma concentration (Cmax-t) and unbound systemic plasma concentration (Cmax-u) classified the strong clinical inducers well but were less accurate in categorizing midrange or weak inducers. For example, using Cmax-t, the calculated R3 values for some of clinical noninducers such as nifedipine, rosiglitazone, omeprazole, and quinidine were <0.9, resulting in false positive assignments. In contrast, R3 values calculated on the basis of Cmax-u were >0.9 for some moderate and weak inducers, such as troglitazone, terbinafine, pleconaril and pioglitazone, leading to false negative assignments. This was true of all three donors regardless of using mRNA or activity as the endpoint.

View this table:
TABLE 5

R3 values calculated on the basis of total and unbound Cmax for induction response of CYP3A4 mRNA and enzyme activity in three lots of human hepatocytes

Predicted AUC Changes Using the Calibration Curves.

Using the constructed calibration curves, the AUC changes for 16 interacting drugs were predicted from all three lots and compared with the observed AUC changes (Figs. 5–7; Supplemental Tables 4–7). As expected, the predicted AUC changes were close to fitted values for the nine drugs interacting with midazolam as the victim drug, although a slight overprediction was observed for troglitazone and terbinafine using RIS and R3, calculated on the basis of Cmax-t (both activity and mRNA) in some of these lots. The correlation plots between the observed and predicted midazolam AUC changes for three hepatocyte lots were prepared for all induction parameters (Figs. 5–7). As anticipated, a strong correlation (R2 = 0.85–0.97) for both mRNA and activity was obtained with the observed AUC changes, regardless of the parameters used. No obvious difference in the robustness of the prediction in midazolam AUC changes was observed using the mRNA versus the activity data. The prediction accuracy and bias for each model were analyzed by RMSE and GMFE using a set of data from both three lots and a single lot. Table 6 shows similar GMFE and RMSE values calculated from this set of three lots for the different prediction methods using mRNA or activity as the measured endpoint. RMSE and GMFE analysis from a single lot provided similar results (data not shown). We found that the correlation was largely improved for all parameters (RIS, R3, and Cmax/EC50) for both mRNA and activity (R2 > 0.94) when using Cmax-u. Consistent with these observations, lower RMSE and GMFE values were obtained when using Cmax-u instead of Cmax-t in the prediction methods (Table 6). No apparent correlation was observed between the observed midazolam or nonmidazolam AUC changes and percentage of positive control response for both activity and mRNA for all three lots (R2 = 0.13–0.41) (Fig. 8).

Fig. 5.
Fig. 5.

Correlation analysis of observed midazolam and nonmidazolam AUC change (%) and predicted AUC changed (%) from RIS [(A) on the basis of Cmax-t; (B) on the basis of Cmax-u], R3 [(C) on the basis of Cmax-t; (D) on the basis of Cmax-u], and Cmax/EC50 [(E) on the basis of Cmax-t; (F) on the basis of Cmax-u] derived from CYP3A4 mRNA for three lots of human hepatocytes. Solid black line represents unity. Dashed lines represent the boundary denoting predicted changes within ±20% of observed. The data points outside of these lines were considered either under- or overpredicted.

Fig. 6.
Fig. 6.

Correlation analysis of observed midazolam and nonmidazolam AUC change (%) and predicted AUC changed (%) from RIS [(A) on the basis of Cmax-t; (B) on the basis of Cmax-u], R3 [(C) on the basis of Cmax-t; (D) on the basis of Cmax-u], and Cmax/EC50 [(E) on the basis of Cmax-t; (F) on the basis of Cmax-u] derived from CYP3A4 enzyme activity for three lots of human hepatocytes. Solid black line represents unity. Dashed lines represent the boundary denoting predicted changes within ±20% of observed. The data points outside of these lines were considered either under- or overpredicted.

Fig. 7.
Fig. 7.

Correlation analysis of observed midazolam and nonmidazolam AUC change (%) and predicted AUC changed (%) from AUC/F2 derived from CYP3A4 mRNA (A) and enzyme activity (B) for three lots of human hepatocytes. Solid black line represents unity. Dashed lines represent the boundary denoting predicted changes within ±20% of observed. The data points outside of these lines were considered either under- or overpredicted.

View this table:
TABLE 6

Accuracy and bias in the prediction of clinical CYP3A induction of midazolam and nonmidazolam CYP3A substrates using the calibration-curve based approaches

The predicted fold changes in AUC from all three hepatocyte lots were used in the calculations.

Fig. 8.
Fig. 8.

Correlation analysis of observed midazolam and nonmidazolam AUC change (%) and percentage of positive control response, derived from CYP3A4 mRNA (A) and enzyme activity (B) for three lots of human hepatocytes.

The prediction for the interacting drugs with nonmidazolam victim drugs was also conducted with these calibration curves. Weak correlations between the observed and predicted nonmidazolam AUC changes for three hepatocyte lots for all induction parameters were found (Figs. 5–7) (R2 < 0.40). However, parameters predicted clinical noninducers reasonably well except for quinidine, where a significant overprediction (32–93% midazolam AUC change) was found using RIS, R3, and Cmax/EC50, generated from mRNA data on the basis of Cmax-u for lots 295 and 312. However, no induction was predicted with AUC/F2 from both mRNA and activity data for quinidine across all three lots. For moderate/weak inducers, the AUC change for nonmidazolam drugs was predicted with varied accuracy. In general, the prediction accuracy was lower and bias was greater in the prediction of the AUC change of CYP3A substrates that were not midazolam. This is evident in the lower RMSE and GMFE values for midazolam trials, as shown in Table 6. In a few cases, either over- or underprediction was also observed for the in vivo AUC changes of nifedipine by phenobarbital, alprazolam by carbamazepine, and simvastatin by troglitazone and pioglitazone, depending on parameters and hepatocyte lots (Supplemental Tables 4–7). Significant overprediction was consistently found for the AUC changes of R-warfarin by sulfinpyrazone (22% observed AUC change versus 59–94% predicted midazolam AUC change) and of carbamazepine by probenecid (20% observed AUC change versus 70–94% predicted midazolam AUC change), with all parameters for both activity and mRNA across all lots except for AUC/F2 for lot 318 (Supplemental Tables 4–7).

Concentration of Test Compounds in the Medium.

The results of such testing in the present study are shown in Supplemental Table 8 for lot 295. Within this set of compounds, concentrations ranged from close to nominal to well below nominal.

Discussion

In this study, model compounds were evaluated for CYP3A4 induction in human hepatocytes and calibration curves constructed to predict responses in vivo. As expected, we observed notable interdonor differences in EC50 and Emax values (e.g., rifampicin EC50 values), which supports regulatory agency guidance recommending calibration of hepatocyte donors for response with a set of inducers and noninducers. Using resulting calibration curves, inducers were predicted with variable accuracy, whereas noninducers were generally well predicted.

Isolated false positive and false negative outcomes were observed. For example, phenobarbital was predicted as a noninducer with the victim drug nifedipine when the calibration curves of total Cmax/EC50 from the enzyme activity and/or mRNA data were used (lots 295 and 312). As phenobarbital is a clinical inducer, these results suggest that evaluating multiple parameter endpoints would be conservative. Quinidine was also incorrectly classified as an in vivo inducer when mRNA was used as the predictor in the RIS, R3, and Cmax/EC50 calibration-curve models obtained on the basis of Cmax-u with donors 295 and 312. This outcome was attributable to the concentration-dependent induction response of CYP3A4 mRNA, but not activity. Quinidine has been classified as a moderate CYP3A4 inhibitor in vivo (Isoherranen et al., 2012), suggesting that any induction in vivo could be masked. These data highlight the value of acquiring enzyme activity results to help one consider additional, more complex models. For example, the “net-effect” model (Fahmi et al., 2009) incorporates parameters of competitive and time-dependent inhibition that may provide a more informed prediction of clinical DDI.

Our data also suggest that midazolam calibration curves may overpredict AUC changes of nonmidazolam victim drugs for weak/moderate inducers, as evidenced by higher midazolam AUC change compared with those obtained with other victim drugs. This is illustrated with predictions of alprazolam AUC change by carbamazepine and simvastatin AUC change by pioglitazone and troglitazone (Table 3; Supplemental Tables 4–7). Similarly, significant overprediction of in vivo response using the midazolam curve was found for sulfinpyrazone and probenecid compared with the observed responses with victim drugs, R-warfarin and carbamazepine. This finding was consistent for all three hepatocytes lots with all modeled induction parameters with exception of AUC/F2 for lot 318, regardless of activity or mRNA endpoint. Fahmi et al. (2012) also reported an overprediction for sulfinpyrazone using midazolam-RIS calibration curves in DPX2 cells. As midazolam exhibits a very high fraction of metabolized CYP3A (fm,CYP3A), it is likely more susceptible to CYP3A4 induction than victim drugs (e.g., R-warfarin) cleared by additional pathways (Ripp et al., 2006; Xu et al., 2011). These findings underscore the use of midazolam as a preferential and sensitive clinical probe for DDI investigations. Notably, two clinically weak inducers, pioglitazone and pleconaril failed to always reach the 2-fold minimum induction response in vitro that would classify a compound as an inducer according to EMA guidance. However, this criterion was met for one or more of the other donors underscoring the value of using three donors in the standard test.

All calculated induction parameters incorporated in vivo total or unbound plasma concentrations of the interacting drugs. Regulatory guidance from the EMA (2013) and FDA (2012) recommend that Cmax-u be used for RIS calculation and Cmax-t for the R3 calculation, respectively. Our results showed that the AUC changes were reasonably well predicted when using either Cmax-t or Cmax-u to calculate parameters and was the case for both mRNA and enzyme activity. However, use of Cmax-u resulted in a better correlation between observed and predicted midazolam AUC change (Figs 5, 6, 7) with an improved accuracy and precision of predicting the DDI, as RMSE and GMSE were lower (Table 6). These observations are consistent with a previous report (Ripp et al., 2006). Conversely, Fahmi et al. (2012) demonstrated that use of total systemic drug concentration in an RIS evaluation resulted in significant improvement in DDI correlations in DPX2 cells, possibly attributable to inclusion of 10% serum in the incubation medium that likely affected the free fraction in the medium.

The FDA draft guidance (FDA, 2012) indicates that an investigational drug is likely to be a P450 inducer when the calculated R3 value is below 0.9. We showed that R3 cut-off values predicting a 20% midazolam AUC change (e.g., a DDI) were much lower than 0.9 (ranging from 0.44 to 0.65 for CYP3A4 mRNA, as well as enzyme activity, across the three lots). Accordingly, we observed a relatively high rate of false positives (e.g., up to 50% exhibited R3 < 0.9). When R3 values were calculated using Cmax-u, we found several false negative outcomes (Table 5). These data indicate that the 0.9 cut-off value along with the prescribed use of Cmax-t proposed by the FDA is conservative. In our evaluation we set the scaling factor d equal to 1, as this is the “assumed” value according to the guidance. Modifying the d value (or the R3 cut-off value) may improve the accuracy of the classifications.

Both regulatory agency guidance documents recommend mRNA as the endpoint for testing induction potential. Fahmi et al. (2010) showed that the measurement of CYP3A4 mRNA was more sensitive in detecting induction in hepatocytes compared with enzyme activity, while both endpoints were found effective at classifying clinical induction response. Our results support a similar conclusion. In general, we selected compounds in our test set biased away from potent inhibitors of CYP3A4 enzyme, to avoid the potentially confounding effects of enzyme inhibition. Clotrimazole, which was shown to exhibit a Ki value for liver microsomal CYP3A4 of 0.25 nM (Gibbs et al., 1999), is the notable exception. In this case, metabolic depletion and/or the wash steps conducted prior to testosterone addition likely precluded significant inhibition. Enzyme activity alone would probably have limited value as a predictor when compounds found to strongly or irreversibly inhibit enzyme activity within hepatocytes (e.g., ritonavir) are examined, as this may not always show a corresponding result in vivo (Kirby et al., 2011). Both midazolam (in vivo probe) and testosterone (in vitro probe) are substrates of CYP3A5 (Williams et al., 2002). This weakly inducible enzyme (Fahmi et al., 2010) exhibits polymorphic expression [e.g., expressed in 10–30% of Caucasians and 50–70% of African Americans (Daly, 2006)]. Whether clinical subject and hepatocyte-donor CYP3A5-genotype status would help explain some variability in the models is not known.

Calculations of F2, RIS, R3, and Cmax/EC50, require preparation of a dose-response curve, ideally with sigmoidal shape and well defined maxima and minima. While minima were reasonably well defined, we noted that approximately 70% of compounds did not reach clear maxima, likely owing to compound incomplete solubility, cytotoxicity, enzyme inhibition, or a combination thereof. In about 15% of the curves, a plateau was not reached because the concentration range was likely insufficient. For these cases, we deployed a strategy of constraining Emax to the observed maximal fold induction level that exhibited no evidence of insolubility or cytotoxicity; the EC50 parameter was then obtained from the curve-fitting model. An alternative approach to not reaching well defined maxima is to use the slope of the curve or AUC/F2 as predictors (Kanebratt and Andersson, 2008; Shou et al., 2008). Our data support the value of obtaining the AUC/F2 parameter.

In an in vitro induction assay, nominal and final (e.g., at the end of the treatment period) intracellular concentrations may differ and could impact model predictivity. Differences may be attributable to cellular uptake, metabolic depletion, compound degradation, binding to cellular components or the plate, or a combination of these. As recommended by the EMA guidance, we investigated drug concentrations in the medium on the last day of incubation. For 11 out of 17 compounds, concentrations were within approximately 2-fold of nominal at the concentrations closest to the reported Cmax-u. However, six compounds exhibited concentrations <20% of nominal (Supplemental Table 8), suggesting that intracellular unbound concentrations were substantially less than those used to derive EC50 and Emax. When we used the time-weighted average concentrations to derive these parameters, in general, EC50 values were lower and Emax values were unchanged. Somewhat surprisingly, this exercise showed no improvement on RMSE and GMFE for any parameter (data not shown).

In conclusion, in vivo CYP3A4 induction responses were well predicted by the plated-hepatocyte model, using parameters RIS, R3, Cmax/EC50, and AUC/F2 in calibration curve–based models. Our data provide no strong basis for selecting a preferential model for predicting an induction response, although AUC/F2 was somewhat less accurate and exhibited higher prediction bias. Enzyme activity and mRNA were equally effective as endpoints. If only one endpoint can be generated, mRNA is preferred, because of the potential confounding effects of enzyme inhibition. However, we found examples (e.g., quinidine) where integrating results of both mRNA and enzyme activity could provide a higher level of confidence in the evaluation compared with either endpoint alone. In a general evaluation scheme, considering the resources needed to construct calibration curves as well as the potential need for range-finding, we would suggest using a three-donor screening test to first classify a potential inducer from a basic method (such as described in the EMA guidance), followed by the more comprehensive RIS testing in calibrated hepatocytes for those compounds exhibiting induction.

Acknowledgments

The authors thank Dr. Charles Crespi from Corning Life Sciences for reviewing the manuscript and for providing useful suggestions.

Authorship Contributions

Participated in research design: Zhang, Stresser.

Conducted experiments: Zhang, Ho, Callendrello, Clark, Santone, Xiao.

Performed data analysis: Zhang, Stresser, Einolf, Kinsman.

Wrote or contributed to the writing of the manuscript: Zhang, Stresser, Einolf, Fox.

Footnotes

    • Received April 15, 2014.
    • Accepted June 12, 2014.

  • Parts of the work were presented at the following meetings: Zhang J (2012, 2013) International Society for the Study of Xenobiotics (ISSX) 18th North American Regional Meeting, Oct. 14–18, 2012, Dallas, TX, and 10th international ISSX meeting, Sept. 29–Oct. 3, 2013, Toronto, ON, Canada.

  • dx.doi.org/10.1124/dmd.114.058602.

  • ↵Embedded ImageThis article has supplemental material available at dmd.aspetjournals.org.

Abbreviations

AUC
area under curve
Cmax-t
total systemic plasma concentration
Cmax-u
unbound systemic plasma concentration
DDI
drug-drug interaction
EC50
the concentration achieving 50% of the maximum response
EMA
European Medicines Agency
Emax
the maximum response
FDA
Food and Drug Administration
F2
the concentration causing 2-fold increase from baseline of the dose-response curve
GMFE
geometric mean fold error
P450
cytochrome P450
PC
positive control
PCR
polymerase chain reaction
R2
correlation coefficient
R3
a term indicating the amount of P450 induction in the liver, expressed as a ratio between 0 and 1
RMSE
root mean square error
RIS
relative induction score

References

    1. Ahdieh H
    1. Adams M,
    2. Pieniaszek HJ Jr., and
    3. Gammaitoni AR
    , andAhdieh H (2005) Oxymorphone extended release does not affect CYP2C9 or CYP3A4 metabolic pathways. J Clin Pharmacol 45:337–345.
    OpenUrlCrossRefPubMed
    1. Ahonen J,
    2. Olkkola KT, and
    3. Neuvonen PJ
    (1995) Effect of itraconazole and terbinafine on the pharmacokinetics and pharmacodynamics of midazolam in healthy volunteers. Br J Clin Pharmacol 40:270–272.
    OpenUrlPubMed
  3. ↵
    1. Almond LM,
    2. Yang J,
    3. Jamei M,
    4. Tucker GT, and
    5. Rostami-Hodjegan A
    (2009) Towards a quantitative framework for the prediction of DDIs arising from cytochrome P450 induction. Curr Drug Metab 10:420–432.
    OpenUrlCrossRefPubMed
    1. Backman JT,
    2. Olkkola KT, and
    3. Neuvonen PJ
    (1996a) Rifampin drastically reduces plasma concentrations and effects of oral midazolam. Clin Pharmacol Ther 59:7–13.
    OpenUrlCrossRefPubMed
    1. Backman JT,
    2. Olkkola KT,
    3. Ojala M,
    4. Laaksovirta H, and
    5. Neuvonen PJ
    (1996b) Concentrations and effects of oral midazolam are greatly reduced in patients treated with carbamazepine or phenytoin. Epilepsia 37:253–257.
    OpenUrlCrossRefPubMed
    1. Backman JT,
    2. Kivistö KT,
    3. Olkkola KT, and
    4. Neuvonen PJ
    (1998) The area under the plasma concentration-time curve for oral midazolam is 400-fold larger during treatment with itraconazole than with rifampicin. Eur J Clin Pharmacol 54:53–58.
    OpenUrlCrossRefPubMed
  7. ↵
    1. Daly AK
    (2006) Significance of the minor cytochrome P450 3A isoforms. Clin Pharmacokinet 45:13–31.
    OpenUrlCrossRefPubMed
    1. Eap CB,
    2. Buclin T,
    3. Cucchia G,
    4. Zullino D,
    5. Hustert E,
    6. Bleiber G,
    7. Golay KP,
    8. Aubert AC,
    9. Baumann P,
    10. Telenti A,
    11. et al.
    (2004) Oral administration of a low dose of midazolam (75 microg) as an in vivo probe for CYP3A activity. Eur J Clin Pharmacol 60:237–246.
    OpenUrlPubMed
  9. ↵
    1. Einolf HJ,
    2. Chen L,
    3. Fahmi OA,
    4. Gibson CR,
    5. Obach RS,
    6. Shebley M,
    7. Silva J,
    8. Sinz MW,
    9. Unadkat JD,
    10. Zhang L,
    11. et al.
    (2014) Evaluation of various static and dynamic modeling methods to predict clinical CYP3A induction using in vitro CYP3A4 mRNA induction data. Clin Pharmacol Ther 95:179–188.
    OpenUrlCrossRef
  10. ↵
    EMA (2013) Guideline on the investigation of drug interactions (final). http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2012/07/WC500129606.pdf.
  11. ↵
    1. Fahmi OA,
    2. Hurst S,
    3. Plowchalk D,
    4. Cook J,
    5. Guo F,
    6. Youdim K,
    7. Dickins M,
    8. Phipps A,
    9. Darekar A,
    10. Hyland R,
    11. et al.
    (2009) Comparison of different algorithms for predicting clinical drug-drug interactions, based on the use of CYP3A4 in vitro data: predictions of compounds as precipitants of interaction. Drug Metab Dispos 37:1658–1666.
    OpenUrlAbstract/FREE Full Text
  12. ↵
    1. Fahmi OA,
    2. Kish M,
    3. Boldt S, and
    4. Obach RS
    (2010) Cytochrome P450 3A4 mRNA is a more reliable marker than CYP3A4 activity for detecting pregnane X receptor-activated induction of drug-metabolizing enzymes. Drug Metab Dispos 38:1605–1611.
    OpenUrlAbstract/FREE Full Text
  13. ↵
    1. Fahmi OA and
    2. Ripp SL
    (2010) Evaluation of models for predicting drug-drug interactions due to induction. Expert Opin Drug Metab Toxicol 6:1399–1416.
    OpenUrlCrossRefPubMed
  14. ↵
    1. Fahmi OA,
    2. Raucy JL,
    3. Ponce E,
    4. Hassanali S, and
    5. Lasker JM
    (2012) Utility of DPX2 cells for predicting CYP3A induction-mediated drug-drug interactions and associated structure-activity relationships. Drug Metab Dispos 40:2204–2211.
    OpenUrlAbstract/FREE Full Text
  15. ↵
    FDA (2012) Guidance for industry—Drug interaction studies—study design, data analysis, implications for dosing, and labeling recommendations (draft guidance). http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm292362.pdf.
    1. Furukori H,
    2. Otani K,
    3. Yasui N,
    4. Kondo T,
    5. Kaneko S,
    6. Shimoyama R,
    7. Ohkubo T,
    8. Nagasaki T, and
    9. Sugawara K
    (1998) Effect of carbamazepine on the single oral dose pharmacokinetics of alprazolam. Neuropsychopharmacology 18:364–369.
    OpenUrlCrossRefPubMed
  17. ↵
    1. Gibbs MA,
    2. Kunze KL,
    3. Howald WN, and
    4. Thummel KE
    (1999) Effect of inhibitor depletion on inhibitory potency: tight binding inhibition of CYP3A by clotrimazole. Drug Metab Dispos 27:596–599.
    OpenUrlAbstract/FREE Full Text
    1. Harris RZ,
    2. Inglis AM,
    3. Miller AK,
    4. Thompson KA,
    5. Finnerty D,
    6. Patterson S,
    7. Jorkasky DK, and
    8. Freed MI
    (1999) Rosiglitazone has no clinically significant effect on nifedipine pharmacokinetics. J Clin Pharmacol 39:1189–1194.
    OpenUrlPubMed
    1. Inglis AM,
    2. Miller AK,
    3. Culkin KT,
    4. Finnerty D,
    5. Patterson SD,
    6. Jorkasky DK, and
    7. Freed MI
    (2001) Lack of effect of rosiglitazone on the pharmacokinetics of oral contraceptives in healthy female volunteers. J Clin Pharmacol 41:683–690.
    OpenUrlCrossRefPubMed
  20. ↵
    1. Isoherranen N,
    2. Lutz JD,
    3. Chung SP,
    4. Hachad H,
    5. Levy RH, and
    6. Ragueneau-Majlessi I
    (2012) Importance of multi-p450 inhibition in drug-drug interactions: evaluation of incidence, inhibition magnitude, and prediction from in vitro data. Chem Res Toxicol 25:2285–2300.
    OpenUrlCrossRefPubMed
  21. ↵
    1. Andersson TB
    1. Kanebratt KP
    andAndersson TB (2008) HepaRG cells as an in vitro model for evaluation of cytochrome P450 induction in humans. Drug Metab Dispos 36:137–145.
    OpenUrlAbstract/FREE Full Text
  22. ↵
    1. Kato M,
    2. Chiba K,
    3. Horikawa M, and
    4. Sugiyama Y
    (2005) The quantitative prediction of in vivo enzyme-induction caused by drug exposure from in vitro information on human hepatocytes. Drug Metab Pharmacokinet 20:236–243.
    OpenUrlCrossRefPubMed
    1. Kim KA,
    2. Oh SO,
    3. Park PW, and
    4. Park JY
    (2005) Effect of probenecid on the pharmacokinetics of carbamazepine in healthy subjects. Eur J Clin Pharmacol 61:275–280.
    OpenUrlCrossRefPubMed
  24. ↵
    1. Kirby BJ,
    2. Collier AC,
    3. Kharasch ED,
    4. Whittington D,
    5. Thummel KE, and
    6. Unadkat JD
    (2011) Complex drug interactions of HIV protease inhibitors 1: inactivation, induction, and inhibition of cytochrome P450 3A by ritonavir or nelfinavir. Drug Metab Dispos 39:1070–1078.
    OpenUrlAbstract/FREE Full Text
    1. Leizorovicz A,
    2. Piolat C,
    3. Boissel JP,
    4. Sanchini B, and
    5. Ferry S
    (1984) Comparison of two long-acting forms of quinidine. Br J Clin Pharmacol 17:729–734.
    OpenUrlCrossRefPubMed
  26. ↵
    1. Livak KJ and
    2. Schmittgen TD
    (2001) Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Δ Δ C(T)) Method. Methods 25:402–408.
    OpenUrlCrossRefPubMed
    1. Ma JD,
    2. Nafziger AN,
    3. Rhodes G,
    4. Liu S, and
    5. Bertino JS Jr.
    (2006) Duration of pleconaril effect on cytochrome P450 3A activity in healthy adults using the oral biomarker midazolam. Drug Metab Dispos 34:783–785.
    OpenUrlAbstract/FREE Full Text
  28. ↵
    1. McGinnity DF,
    2. Zhang G,
    3. Kenny JR,
    4. Hamilton GA,
    5. Otmani S,
    6. Stams KR,
    7. Haney S,
    8. Brassil P,
    9. Stresser DM, and
    10. Riley RJ
    (2009) Evaluation of multiple in vitro systems for assessment of CYP3A4 induction in drug discovery: human hepatocytes, pregnane X receptor reporter gene, and Fa2N-4 and HepaRG cells. Drug Metab Dispos 37:1259–1268.
    OpenUrlAbstract/FREE Full Text
    1. Mihaly GW,
    2. Ching MS,
    3. Klejn MB,
    4. Paull J, and
    5. Smallwood RA
    (1987) Differences in the binding of quinine and quinidine to plasma proteins. Br J Clin Pharmacol 24:769–774.
    OpenUrlCrossRefPubMed
  30. ↵
    1. Ohtsuki S,
    2. Schaefer O,
    3. Kawakami H,
    4. Inoue T,
    5. Liehner S,
    6. Saito A,
    7. Ishiguro N,
    8. Kishimoto W,
    9. Ludwig-Schwellinger E,
    10. Ebner T,
    11. et al.
    (2012) Simultaneous absolute protein quantification of transporters, cytochromes P450, and UDP-glucuronosyltransferases as a novel approach for the characterization of individual human liver: comparison with mRNA levels and activities. Drug Metab Dispos 40:83–92.
    OpenUrlAbstract/FREE Full Text
    1. O’Reilly RA
    (1982) Stereoselective interaction of sulfinpyrazone with racemic warfarin and its separated enantiomorphs in man. Circulation 65:202–207.
    OpenUrlAbstract/FREE Full Text
    1. Prueksaritanont T,
    2. Vega JM,
    3. Zhao J,
    4. Gagliano K,
    5. Kuznetsova O,
    6. Musser B,
    7. Amin RD,
    8. Liu L,
    9. Roadcap BA,
    10. Dilzer S,
    11. et al.
    (2001) Interactions between simvastatin and troglitazone or pioglitazone in healthy subjects. J Clin Pharmacol 41:573–581.
    OpenUrlCrossRefPubMed
  33. ↵
    1. Ripp SL,
    2. Mills JB,
    3. Fahmi OA,
    4. Trevena KA,
    5. Liras JL,
    6. Maurer TS, and
    7. de Morais SM
    (2006) Use of immortalized human hepatocytes to predict the magnitude of clinical drug-drug interactions caused by CYP3A4 induction. Drug Metab Dispos 34:1742–1748.
    OpenUrlAbstract/FREE Full Text
    1. Schellens JH,
    2. van der Wart JH,
    3. Brugman M, and
    4. Breimer DD
    (1989) Influence of enzyme induction and inhibition on the oxidation of nifedipine, sparteine, mephenytoin and antipyrine in humans as assessed by a “cocktail” study design. J Pharmacol Exp Ther 249:638–645.
    OpenUrlAbstract/FREE Full Text
    1. Selen A,
    2. Amidon GL, and
    3. Welling PG
    (1982) Pharmacokinetics of probenecid following oral doses to human volunteers. J Pharm Sci 71:1238–1242.
    OpenUrlCrossRefPubMed
  36. ↵
    1. Shimada T,
    2. Yamazaki H,
    3. Mimura M,
    4. Inui Y, and
    5. Guengerich FP
    (1994) Interindividual variations in human liver cytochrome P-450 enzymes involved in the oxidation of drugs, carcinogens and toxic chemicals: studies with liver microsomes of 30 Japanese and 30 Caucasians. J Pharmacol Exp Ther 270:414–423.
    OpenUrlAbstract/FREE Full Text
    1. Shord SS,
    2. Chan LN,
    3. Camp JR,
    4. Vasquez EM,
    5. Jeong HY,
    6. Molokie RE,
    7. Baum CL, and
    8. Xie H
    (2010) Effects of oral clotrimazole troches on the pharmacokinetics of oral and intravenous midazolam. Br J Clin Pharmacol 69:160–166.
    OpenUrlCrossRefPubMed
  38. ↵
    1. Shou M,
    2. Hayashi M,
    3. Pan Y,
    4. Xu Y,
    5. Morrissey K,
    6. Xu L, and
    7. Skiles GL
    (2008) Modeling, prediction, and in vitro in vivo correlation of CYP3A4 induction. Drug Metab Dispos 36:2355–2370.
    OpenUrlAbstract/FREE Full Text
    1. Soons PA,
    2. van den Berg G,
    3. Danhof M,
    4. van Brummelen P,
    5. Jansen JB,
    6. Lamers CB, and
    7. Breimer DD
    (1992) Influence of single- and multiple-dose omeprazole treatment on nifedipine pharmacokinetics and effects in healthy subjects. Eur J Clin Pharmacol 42:319–324.
    OpenUrlPubMed
  40. ↵
    1. Wienkers LC and
    2. Heath TG
    (2005) Predicting in vivo drug interactions from in vitro drug discovery data. Nat Rev Drug Discov 4:825–833.
    OpenUrlCrossRefPubMed
  41. ↵
    1. Wrighton SA
    1. Williams JA,
    2. Ring BJ,
    3. Cantrell VE,
    4. Jones DR,
    5. Eckstein J,
    6. Ruterbories K,
    7. Hamman MA, and
    8. Hall SD
    , andWrighton SA (2002) Comparative metabolic capabilities of CYP3A4, CYP3A5, and CYP3A7. Drug Metab Dispos 30:883–891.
    OpenUrlAbstract/FREE Full Text
  42. ↵
    1. Xu Y,
    2. Zhou Y,
    3. Hayashi M,
    4. Shou M, and
    5. Skiles GL
    (2011) Simulation of clinical drug-drug interactions from hepatocyte CYP3A4 induction data and its potential utility in trial designs. Drug Metab Dispos 39:1139–1148.
    OpenUrlAbstract/FREE Full Text
  43. ↵
    1. Zhang JG,
    2. Ho T,
    3. Callendrello AL,
    4. Crespi CL, and
    5. Stresser DM
    (2010) A multi-endpoint evaluation of cytochrome P450 1A2, 2B6 and 3A4 induction response in human hepatocyte cultures after treatment with β-naphthoflavone, phenobarbital and rifampicin. Drug Metab Lett 4:185–194.
    OpenUrlCrossRefPubMed
Back to top

In this issue

Download PDF
Article Alerts
Email Article
Citation Tools

Research ArticleArticle

Prediction of CYP3A4 Induction Using Calibration Curves

J. George Zhang, Thuy Ho, Alanna L. Callendrello, Robert J. Clark, Elizabeth A. Santone, Sarah Kinsman, Deqing Xiao, Lisa G. Fox, Heidi J. Einolf and David M. Stresser
Drug Metabolism and Disposition September 1, 2014, 42 (9) 1379-1391; DOI: https://doi.org/10.1124/dmd.114.058602
Twitter logo Facebook logo Mendeley logo

Jump to section

Advertisement