Abstract
Prediction of clearance in drug discovery currently relies on human primary hepatocytes, which can vary widely in drug-metabolizing enzyme activity. Potential alternative in vitro models include the HepaRG cell (from immortalized hepatoma cells), which in culture can express drug-metabolizing enzymes to an extent comparable to that of primary hepatocytes. Utility of the HepaRG cell will depend on robust performance, relative to that of primary hepatocytes, in routine high-throughput analysis. In this study, we compared intrinsic clearance (CLint) in the recently developed cryopreserved HepaRG cell system with CLint in human cryopreserved pooled hepatocytes and with CLint in vivo for 26 cytochrome P450 substrate drugs. There was quantitative agreement between CLint in HepaRG cells and human hepatocytes, which was linear throughout the range of CLint (1–2000 ml · min−1 · kg−1) and not dependent on particular cytochrome P450 involvement. Prediction of CLint in HepaRG cells was on average within 2-fold of in vivo CLint (using the well stirred liver model), but average fold error was clearance-dependent with greater underprediction (up to at least 5-fold) for the more highly cleared drugs. Recent reporting of this phenomenon in human hepatocytes was therefore confirmed with the hepatocytes used in this study, and hence the HepaRG cell system appears to share an apparently general tendency of clearance-limited CLint in cell models. This study shows the cryopreserved HepaRG cell system to be quantitatively comparable to human hepatocytes for prediction of clearance of drug cytochrome P450 substrates and to represent a promising alternative in vitro tool.
Introduction
Clearance is of pivotal importance in drug discovery, and hence much effort to predict human clearance is focused on the early stages of drug discovery. Demand has increased as technology for screening has developed. However, despite technological advances in human liver derived in vitro systems over the last two decades and improvements in understanding prediction methodology, accurate prediction of intrinsic clearance (CLint) in vivo remains elusive.
At present, the isolated human hepatocyte represents the most versatile human hepatic in vitro system in standard use and an improvement over the established hepatic microsome standard, because of a more complete set of uptake, metabolism, and elimination routes. However, prediction of clearance from hepatocytes, as with hepatic microsomes, is imprecise with bias toward underprediction (Hallifax et al., 2010). A major source of uncertainty in prediction of clearance is the phenotypical expression of human cytochrome P450 and (by implication) transporters; in addition, the impact on these activities by liver processing and storage procedures may be a trend toward loss of activity (Hallifax and Houston, 2009). Introducing an extrapolation factor to compensate for the average bias (underprediction) in prediction of clearance from hepatocytes has been suggested (Stringer et al., 2008; Hallifax et al., 2010), but this will not improve the precision of predictions. However, the reliance on an in vitro system obtained directly from human donors, with the additional difficulty of limited supply, may be challenged by cell line systems that have been developed in recent years.
A promising potential alternative to primary hepatocytes is the HepaRG cell system, which has been derived from human immortalized hepatoma cells. HepaRG cells retain most liver functions, including cytochrome P450 drug-metabolizing enzymes (Guillouzo et al., 2007). These cells, when plated at low density with medium containing DMSO, differentiate into two cell types, resembling either adult primary hepatocytes (with bile canaliculi) or biliary epithelial cells (Parent et al., 2004) and are able to perform many specific liver functions and respond to prototypical drug metabolism inducers. Moreover, modulation of the concentration of DMSO during the differentiation process can result in expression of cytochrome P450 and transporters at levels similar to or greater than those in primary human hepatocytes in culture (Aninat et al., 2006; Lübberstedt et al., 2011). Since becoming available, HepaRG cells have been used successfully for assessment of cytochrome P450 induction (McGinnity et al., 2009; Anthérieu et al., 2010) and in toxicology studies (Dumont et al., 2010; Lübberstedt et al., 2011; McGill et al., 2011) and appear to have the potential for further understanding of the interplay between drug metabolism, cell metabolism, and liver function in preclinical toxicological studies for either known drugs and new chemical entities. Kanebratt and Andersson (2008) have demonstrated activities of several major drug-metabolizing cytochrome P450s in cultured HepaRG cells, which were of a magnitude similar to that in primary hepatocytes. This indication of the quantitative performance of the HepaRG system has been encouraging for the further development of a readily available in vitro model suitable for high-throughput use. Recently, Biopredic International (Rennes, France) has developed cryopreservation of the HepaRG cell system, which appears to offer convenience to match that of cryopreserved primary hepatocytes, such as being more amenable to automated assay.
The aims of this study were 2-fold. First, we present an evaluation of the cryopreserved HepaRG cell system for prediction of hepatic clearance using standard high-throughput methodology. A set of 26 drug substrates was selected on the basis of availability of clinical pharmacokinetic data over a wide range of clearances together with a range of cytochrome P450-mediated pathways, lipophilicity, ionic charge, and protein binding (Table 1). CLint was determined by substrate depletion and compared with values obtained in a parallel study using pooled cryopreserved human hepatocytes. CLint was also compared with in vivo (clinical) CLint, and prediction accuracy was assessed. Second, the human hepatocyte data collected were used as a consistent dataset to explore the validity of previous hepatocyte studies of similar size (Foster et al., 2011) to investigate trends in prediction occurring and utility for in vivo prediction.
Materials and Methods
Materials.
Pooled cryopreserved human hepatocytes (lot HuP61: five white males and five white females) and cryopreserved hepatocyte recovery medium were purchased from CellzDirect (Durham, NC). HepaRG cells, cultured in medium containing DMSO (2%) to maximize expression of drug-metabolizing activities, were cryopreserved and supplied in vapor of liquid nitrogen by Biopredic International. Hanks' balanced salt solution without phenol red (10×) and HEPES buffer solution 1 were purchased from Invitrogen (Carlsbad, CA); Williams' medium E (WME), and all other chemicals not specified above were purchased from Sigma-Aldrich (St. Louis, MO).
HepaRG Cell Incubations.
HepaRG cells stored in liquid nitrogen were thawed according to the instructions from the supplier and resuspended in WME (serum- and phenol-free) at 37°C in a fresh tube. In viability counting, all cells appeared spheroid in shape, and no morphological difference (hepatocyte versus biliary cell) was detected; therefore, calculation of viability and CLint was based on the total number of cells. The substrates (Table 1) were prepared as solutions in aqueous DMSO (10 mM), diluted first to 2 mM in acetonitrile-water (50:50) and then diluted to 2 μM in WME. Each incubation volume (125 μl) was prepared by mixing an equal amount of cell suspension and compound dilution in a 96-well plate to obtain the final conditions of 0.05% of total organic solvent, 1 μM compound, and 0.25 × 106 cells/ml. The concentration of cells in the incubation was selected to match that determined for hepatocytes (below). Incubation plates were maintained at 37°C and shaken on a Thermoshake (INHECO, Munich, Germany). Reactions were stopped with an equal volume of acetonitrile at 0, 10, 20, 40, and 80 min, and samples were filtered by centrifugation in a Solvinert MultiScreen sandwich plate. The whole procedure was performed by a Multiprobe II EX HT robot with WinPrep software used for sample preparation (Perkin-Elmer Life and Analytical Sciences, Waltham, MA). Each incubation was performed in duplicate (replicate data were accepted within 10% of each other).
Hepatocyte Incubations.
Hepatocytes stored in liquid nitrogen were rapidly thawed in a shaking water bath at 37°C and recovered after centrifugation (75g; room temperature) in cryopreserved hepatocyte recovery medium. The cells were resuspended in WME (serum- and phenol-free) at 37°C in a fresh tube. Viability was confirmed to be greater than 70%, and the cell concentration was adjusted to obtain 0.5 × 106 cells/ml. The same procedure as for HepaRG cells was then followed using the same 26 substrates plus a further 15 substrates (Table 1) selected from a database of predictions of CLint (Hallifax et al., 2010); these additional substrates were used to potentially improve comparison with data reported previously from this system. Before these experiments, a trial incubation with prototypical cytochrome P450 probes (nifedipine, diclofenac, and testosterone) and verapamil was conducted to evaluate depletion rate linearity with cell concentration from 0.1 to 2 × 106 cells/ml; lower intrinsic clearance was obtained at 1 and 2 × 106 cells/ml and so the concentration of 0.25 × 106 cells/ml was selected for incubations. Each incubation was performed in duplicate (replicate data were accepted within 10% of each other).
The design of the study (as a single experiment in duplicate) was due to the constraint of supply of the novel cryopreserved HepaRG cells. The comparison with hepatocytes and with in vivo data were robust in terms of the range of drugs and CLint examined and directly compared.
Liquid Chromatography-Mass Spectrometry-Time-of-Flight Analysis.
All samples were analyzed by an ultraperformance liquid chromatography LCT Premier time-of-flight mass spectrometer (Waters, Milford, MA). Samples (10 μl) were separated by a Acquity ultraperformance liquid chromatography BEH C18 column (1.7 μm, 2.1 × 50 mm; Waters) at 40°C and eluted within 1.8 min by a binary gradient from 0% at 0.3 min to 100% at 1.3 min of mobile phase B [acetonitrile with 0.1% (v/v) formic acid] against mobile phase A [water with 0.1% (v/v) formic acid]. The mass spectrometer received approximately one-fifth of the total 0.6 ml/min flow and was set in electrospray ionization-positive V scan mode, with capillary voltage 3 kV, sample cone 50 V, and temperature of desolvation and of source 350 and 120°C, respectively; flow rates of desolvation gas and cone gas were 850 and 50 l/h, respectively.
Determination of CLint from In Vitro Incubations.
Substrate concentration-time (depletion) data were fitted to the monoexponential decay model, where adequately linear (below), with a 1/y weighting (eq. 1): where Ct and C0 are the substrate concentration in the incubation medium at time t and initially, respectively, and k is the rate constant.
On the basis of the assumption that the concentration used (1 μM) was sufficiently below the Km of the test compound, the in vitro CLint was calculated by dividing the elimination constant (k) by the cell suspension concentration. Goodness of fit was assessed by visual inspection and the Akaike information criterion compared with the alternative biexponential fit. GraphPad Prism 4.03 software (GraphPad Software Inc., San Diego, CA) was used for analysis. If the Akaike value indicated a biexponential depletion or if two or more following time points were far from the linear fit, the in vitro CLint was calculated by dividing the initial compound amount with area under the curve, incubation volume, and hepatocyte concentration. The area under the curve was calculated by the linear trapezoidal method using WinNonlin (Pharsight, Mountain View, CA).
In vitro CLint values were scaled to the in vivo whole-body equivalent using values for average hepatocellularity, liver weight, and body weight (eq. 2): where SF is the hepatocellularity of 120 × 106 cells/g liver (Brown et al., 2007), HLW is human liver weight of 21.4 g liver/kg b.wt. (Brown et al., 2007), and fu, inc is the fraction unbound in the hepatocyte incubation. SF and HLW values are in close agreement with reported means of databases of published studies. The interindividual variability of these parameters may be substantial (95% confidence interval for hepatocellularity of 74–131 × 106 cells/g), but it has been demonstrated that this variability does not significantly affect the average prediction of CLint (Barter et al., 2007).
Fraction unbound in the incubation was calculated according to Kilford et al. (2008) (eq. 3): where log P/D is log P for neutral and positively charged compounds and log D for acidic compounds.
Calculation of CLint Observed In Vivo.
The intrinsic clearance observed in vivo (CLint, u, Obs) was calculated using the well stirred liver physiological model (eq. 4). This choice of liver model was based on adequacy for the objectives of this study when considering the marginal difference that we found using the most common alternative, the parallel tube model, in agreement with expectation from previous observations (Ito and Houston, 2004). where CLh is hepatic clearance, fub is fraction unbound in blood, and Qh is hepatic blood flow of 20.7 ml · min−1 · kg−1 b.wt. (Davies and Morris, 1993). Fraction unbound in blood (fub) was derived from the literature (Stringer et al., 2008; Hallifax et al., 2010).
Calculation of Precision and Accuracy.
The accuracy of a set of predictions of CLint for different drugs, given as the ratio of in vitro CLint (predicted) and in vivo CLint (observed) was assessed using the average fold error (AFE) (geometric mean error) determined by eq. 5 (Obach, 1997): Fold underprediction = 1/AFE.
The corresponding precision of the prediction sets was assessed using the root mean square error (RMSE) determined using eq. 6 (Sheiner and Beal, 1981):
Relation of Predicted In Vitro CLint and In Vivo CLint.
To provide a correlation of in vitro CLint with in vivo CLint, where lack of proportionality was observed, the following log linear function was fitted by least-squares regression using Microsoft Excel:
Results
Comparison of CLint, u between HepaRG Cells and Hepatocytes.
The intrinsic clearances of 41 drugs in cryopreserved primary human hepatocytes and 26 drugs in both cryopreserved HepaRG cells and hepatocytes, determined by substrate depletion, is given in Table 2. CLint, u ranged between 0.5 and 2100 ml · min−1 · kg−1 for HepaRG cells and 2.6 and 1900 ml · min−1 · kg−1 for hepatocytes. Although there were differences between the systems on an individual drug basis (notably tolbutamide; see Discussion), a direct correlation of CLint, u between the systems was apparent over the range of CLint, u investigated (Fig. 1). This correlation showed that the average ratio between the systems was close to unity, indicating a quantitative similarity in drug depletion between the HepaRG cells and the hepatocytes over the entire range of CLint examined.
Prediction of In Vivo CLint.
Intrinsic clearances for HepaRG cells and hepatocytes together with CLint, u calculated in vivo are shown in Table 2. There was a correlation between CLint, u predicted in vitro and CLint observed in vivo (CLint, u, Obs) with either HepaRG cells or hepatocytes (Fig. 2); r2 was 0.53 in both HepaRG cells (Fig. 2A) and in hepatocytes (Fig. 2B). The slopes of the correlations were 0.55 for HepaRG cells and 0.56 for hepatocytes, indicating a disproportionate relationship in both cases.
The accuracy of prediction of CLint was assessed by AFE between in vitro and in vivo. On average, HepaRG cells and hepatocytes both underpredicted in vivo CLint but to a similar extent (1.9- and 1.8-fold, respectively), confirming the direct correlation between the systems (Table 3). Precision among predictions was similar between the systems, with RMSE of 5531 for HepaRG cells and 4546 for hepatocytes (Table 3).
When drugs were subdivided into groups according to level of CLint (by order of magnitude range), there was a clear trend toward decreasing prediction ratio with increasing CLint (Table 3): compounds with CLint, u, Obs lower than 10 ml · min−1 · kg−1 were generally overpredicted in both systems by a factor of at least 3 (45.9- and 3.0-fold in HepaRG cells and hepatocytes, respectively); compounds with CLint, u, Obs between 10 and 100 ml · min−1 · kg−1 were up to 2-fold underpredicted (1.4- and 2.1-fold in HepaRG cells and hepatocytes, respectively); and compounds with CLint, u, Obs between 100 and 1000 ml · min−1 · kg−1 were more underpredicted (3.1- and 6.7-fold in HepaRG cells and hepatocytes, respectively) as were compounds with CLint, u, Obs greater than 1000 ml · min−1 · kg−1 (5.9- and 5.0-fold in HepaRG cells and hepatocytes, respectively). The trend in bias indicated by individual fold errors is illustrated in Fig. 3.
Segregation of drugs according to the major cytochrome P450 or phase II clearance pathway showed that for the all major routes except CYP2D6, underprediction was within 2-fold (Table 4); for those drugs for which CYP2D6 is a major pathway, underprediction was only marginally greater, at 3- and 2-fold for HepaRG and hepatocytes, respectively (Table 4), indicating minimal difference between the systems and between the in vitro and in vivo situations in terms of cytochrome P450 bias.
Discussion
The novel hepatic in vitro system, the HepaRG cell, has in recent years emerged as a viable new tool for drug discovery, with some success to date as a model for liver toxicity and induction of drug metabolism (Aninat et al., 2006; Guillouzo et al., 2007; Anthérieu et al., 2010; Dumont et al., 2010; McGill et al., 2011). HepaRG cells can be routinely produced with expression levels of cytochrome P450 similar to those found in primary hepatocytes and so have potential for use as a model for prediction of clearance—a vital component of drug discovery and one that currently relies on hepatocytes from human liver donors (an unsatisfactory situation because of limits in supply, quality, and variability as discussed widely in the literature). However, despite the potential of the HepaRG system for prediction of clearance, it would remain impractical as a widely available drug discovery tool without the successful demonstration of cryopreservation.
Cryopreservation of primary hepatocytes, despite the well documented drawbacks, has enabled a degree of standardization of a human primary hepatocyte system through pooling and other processes, but the impact of this will probably remain limited by the high level of uncertainty inherent in this system. One particular developer of the HepaRG system, Biopredic International, has recently established successful cryopreservation, enabling commercial supply. The purpose of the study reported here was to provide a quantitative evaluation of the performance of this new cryopreserved HepaRG cell system in a high-throughput situation. With reference to previously reported quantitative evaluations of hepatocytes, a range of 26 drugs of diverse properties (clearance, charge, and lipophilicity) was selected with which to compare intrinsic clearance in the HepaRG cell both with hepatocytes (cryopreserved pool) and in vivo.
CLint in HepaRG cells was on average equal to CLint in the hepatocytes and directly proportional over the entire range of CLint (1–2000 ml · min−1 · kg−1), indicating that the HepaRG cell system is quantitatively very similar, in terms of prediction of hepatic clearance, to cryopreserved hepatocytes; there was no clearance-dependent bias between the systems as has been reported between human liver microsomes and hepatocytes, in a recent database analysis of predictions (Hallifax et al., 2010). Although the ratio of CLint between the systems varied depending on drug, segregation of predictions according to primary metabolic pathway showed no evidence that this variation was associated with any specific cytochrome P450. One previous report that characterized HepaRG on the basis of prototypical probe substrate found a lower level of metabolism of 7-ethoxycoumarin and dextromethorphan demethylase (Kanebratt and Andersson, 2008), whereas in another report, CYP1A and CYP2D6 activities were considered similar to those in hepatocyte preparations (Turpeinen et al., 2009; Lübberstedt et al., 2011). In the present study, although this pathway analysis was partly obscured by the incidence of multiple cytochrome P450 pathways, it may be encouraging that there were no major deviations in CLint between the systems with regard to any prior concerns about potentially unrepresentative expression of cytochrome P450 in the HepaRG system (with the exception of tolbutamide: a low CLint in hepatocytes relative to microsomes indicated a more specific cause, such as uptake transport, because good CYP2C9 expression was indicated by warfarin and diclofenac; Brown et al., 2007).
When CLint in the HepaRG system (scaled to whole body) was compared with in vivo CLint (calculated using the well stirred liver model), the average ratio indicated an underprediction bias (by AFE) of just less than 2-fold. This level of bias may be considered minor with respect to the historical average prediction bias from hepatocytes of approximately 5-fold (Hallifax et al., 2010). Average prediction of CLint from the hepatocytes in this study was also within 2-fold, indicating above-average (historical) metabolic rates within the sample of hepatocytes used.
It has recently been shown that the bias in prediction of CLint from hepatocytes is clearance-dependent, with greater bias observed in vitro with increasing level of CLint (Hallifax et al., 2010; Foster et al., 2011). Segregation of in vitro/in vivo ratio of CLint in the present study showed that bias in the HepaRG system was also clearance-dependent, confirming that use of average bias can be misleading and that substantial underprediction of CLint can occur for highly cleared drugs. This result was supported by a similar clearance dependence in bias from the hepatocytes in the present study, which appears to substantiate this effect as a more general in vitro cellular phenomenon [supported by the additional substrates used with hepatocytes in the present study (n = 41) and the similar correlation slopes in this and previous studies, as above]. Only marginal differences in predictions of CLint arise between use of the well stirred and parallel tube liver models and clearance dependence in prediction bias has been shown to be independent of liver model (Hallifax et al., 2010). At present, the cause of this rate limitation can only be speculated on. Huang et al. (2010) demonstrated that underprediction of a set of in-house compounds by hepatocytes was due to biliary efflux in vivo, a phenomenon that is not accounted for in in vitro systems; however, in the present study, this is unlikely to apply because substrates of transporters (particularly P-glycoprotein) were avoided. More appropriate to the drugs used in this study, Foster et al. (2011) speculated that clearance dependence in prediction of CLint may be due to a capacity limitation, such as cofactor exhaustion or, alternatively, permeation rate limitation in vitro.
The sets of predictions of CLint from the two in vitro systems in this study were characterized with a similar level of imprecision, which reinforces a broad quantitative similarity in the performance of the novel HepaRG system with the alternative cryopreserved hepatocyte system, although this imprecision was greater than that reported for a larger database of published predictions (Hallifax et al., 2010). Overall, the similarity in performance with hepatocytes that this study has demonstrated is reassuring with regard to the quantitative application of HepaRG cells, providing that precautions for the clearance-dependent bias, can be applied. The similarity with hepatocytes indicates that a simple correction could be applied, as has been suggested for hepatocytes (Foster et al., 2011), to avoid critical misidentification of very highly cleared compounds at the drug candidate screening stage; in the case of the HepaRG system, this might simply require adjustment of the intercept of the suggested empirical correction (to reflect lower average bias). Because of the nature of its production, the HepaRG system appears to represent a more consistent in vitro model with regard to reproducible expression of cytochrome P450s and possibly other metabolic enzymes which, considering the variability in cytochrome P450 activity that can arise in terms of phenotypic expression with current in vitro systems, may be a key benefit of this novel system.
In conclusion, this study has shown the potential of the cryopreserved HepaRG cell system as a quantitative in vitro model for prediction of intrinsic clearance, which may provide a standardized alternative to the currently used liver donor-derived hepatocyte. In addition, the clearance-dependent in vivo prediction of human hepatocytes is confirmed.
Authorship Contributions
Participated in research design: Zanelli, Hallifax, and Houston.
Conducted experiments: Caradonna and Turlizzi.
Performed data analysis: Zanelli.
Wrote or contributed to the writing of the manuscript: Zanelli, Hallifax, and Houston.
Acknowledgments
We acknowledge Dr. Chiara Ghiron at Siena Biotech for useful discussion on the manuscript and Dr. Christophe Chesné and Bénédicte Shevchenko at Biopredic International for the supply of cryopreserved differentiated HepaRG cells within the partnership of the Liintop Project.
Footnotes
This work was supported by funding from the European Union Sixth Framework Programme for Optimisation of Liver and Intestine In Vitro Models for Pharmacokinetics [Liintop/STREP 037499]; and by a consortium of pharmaceutical companies (GlaxoSmithKline, Lilly, Pfizer, and Servier) within the Centre for Applied Pharmacokinetic Research at the University of Manchester.
Article, publication date, and citation information can be found at http://dmd.aspetjournals.org.
-
ABBREVIATIONS:
- CLint
- intrinsic clearance
- DMSO
- dimethyl sulfoxide
- WME
- Williams' medium E
- AFE
- average fold error
- RMSE
- root mean square error.
- Received August 16, 2011.
- Accepted October 13, 2011.
- Copyright © 2012 by The American Society for Pharmacology and Experimental Therapeutics