Calculating Intrinsic Clearance.

To perform intrinsic clearance studies with hepatocytes, either formation of metabolites or depletion of the parent test drug from the incubation is followed. Since these studies are often performed earlier in drug research when metabolic routes are not well-characterized and metabolite standards are not available, substrate depletion is the most commonly used methodology (Obach and Reed-Hagen, 2002). Following in vitro incubations with hepatocytes, a rate of depletion () of the test article from the incubation is generated, which is then used to predict the in vitro Cl_int value of the compound (ml/min/kg) using in vitro scaling factors, and ultimately to predict hepatic clearance (the in vitro Cl_h value).

For the purpose of comparing published in vitro models and their performance in predicting clearance, the reported data from six studies were evaluated (Blanchard et al., 2005; Hallifax et al., 2010; Di et al., 2012, 2013; Smith et al., 2012; Chan et al., 2013). It is important to note when comparing different reports of hepatic clearance predictions using in vitro results that the data analyses were performed with a variety of scaling factors and approaches. Thus, to enable a more effective side-by-side comparison of the previously reported data, recalculations of the in vitro Cl_int and Cl_h values were conducted to ensure that the same scaling factors and assumptions were made across data sets. There are differing opinions on the modeling techniques to use for clearance predictions; however, the pros and cons of different approaches are outside the purpose of this paper (Fagerholm, 2007; Hallifax and Houston, 2009). It is also acknowledged that there is inherent variability in these in vitro estimates, from both laboratory and biologic sources. However, this analysis uses point estimates for both the in vitro and in vivo data in order to make these comparisons. Furthermore, a static modeling approach was taken that does not take into account the dynamic in vivo and in vitro environment in which metabolism is actually occurring. However, this approach is routinely used by DMPK scientists, and thus was used to focus on the techniques and outcomes in order to compare the various in vitro models for slow metabolic clearance.

To evaluate the prediction accuracy of the reported in vitro hepatocyte models, two comparisons were made: 1) scaled in vitro Cl_int versus back-calculated in vivo Cl_int from the literature total clearance in vivo (Cl_total) data (using the well-stirred model); and 2) scaled Cl_h generated from in vitro data (in vitro Cl_h) versus literature values for Cl_total (in vivo Cl_total). To predict Cl_int from in vitro data, scaling factors of 120 million hepatocytes/g liver (Hosea et al., 2009) and 25.7 g liver/kg (Davies and Morris, 1993) were used(1)To calculate in vitro Cl_h from in vitro data, eq. 2 (the well-stirred model) was used (Rowland et al., 1973), with or without the incorporation of the unbound fraction of drug in the blood (f_uB = f_up/R_B, where f_up = fraction unbound in plasma and R_B = red blood cell partitioning) and liver blood flow (Q_h = 20.7 ml/min/kg) (Davies and Morris, 1993)(2)In addition, to calculate in vivo Cl_int from observed Cl, a back-calculation using the rearranged well-stirred model (eq. 3) was used, where Cl = in vivo Cl_total [or in the case of analysis of data from Chan et al. (2013), Cl = in vitro Cl_h to back-calculate in vitro Cl_int]

(3)

Thus, data reported from the various in vitro systems reviewed herein were recalculated to assure that model evaluation was performed using a common platform (Table 1). The only deviation from this was for the literature review concerning the evaluation of the suspended hepatocyte model for clearance estimations (Hallifax et al., 2010). These data also included a correction of binding in the hepatocyte matrix as an added scaler to eq. 1, which was not teased out from the data reported herein.

Suspended Hepatocytes.

While the use of hepatocytes has improved the underprediction bias of clearance noted with microsomes, there still generally remains a bias toward underprediction (Brown et al., 2007), which worsens as clearance increases (Hallifax et al., 2010). There has been speculation that this may relate to possible loss of the drug uptake transporters in cryopreserved hepatocytes (Lundquist et al., 2014). For slowly metabolized compounds, the rapid loss of enzymatic function alone limits the use of this system since it is difficult to reliably detect small losses of test article from the incubations; however, other factors such as transporter function should also be kept in mind when using suspended hepatocytes to predict total clearance.

Hallifax et al. (2010) published a review evaluating the prediction accuracy of hepatocyte incubations to assess clearance for 89 compounds from a variety of literature sources. For this evaluation, data from many researchers were included, and thus it is acknowledged that the incubation conditions were variable, which may impact some conclusions from these prediction comparisons. Furthermore, the free fraction in the hepatocyte matrix was included in the calculations, which was not included in analyses of other hepatocyte modeling techniques reported herein. In a re-evaluation of a subset of these compounds (n = 17), where in vitro Cl_int is compared with in vivo Cl_int based on the calculations described here (eqs. 1 and 3), the conclusion of Hallifax et al. (2010) was confirmed, which indicates that there is a clearance-dependent underprediction bias exacerbated as in vivo Cl increases (Fig. 1A; Table 1). In a re-examination of compounds that exhibit low intrinsic clearance (Cl_int ≤ 15 ml/min/kg), this bias was less pronounced (Fig. 1B). Thus, using Cl_int to compare in vitro data to in vivo data, suspended hepatocyte incubations performed well for low-clearance compounds when metabolic loss was reliably detected (i.e., the major assumption using this in vitro system), but generally trends toward underprediction for higher-clearance compounds.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

In vivo Cl_int versus in vitro Cl_int generated in hepatocytes in suspension (standard and relay) and plated monoculture. The solid lines represent the line of unity; the dashed lines represent 2- and 3-fold differences from the line of unity. (A) Full data set. (B) Data set focused on low in vivo Cl_int ≤ 15 ml/min/kg. ▪, suspended hepatocytes (Hallifax et al., 2010); n = 17/n = 6 for total and low Cl_int, respectively. ▲, relay method with suspended hepatocytes (Di et al., 2012, 2013); n = 11/n = 7 for total and low Cl_int, respectively. ○, plated hepatocytes (Blanchard et al., 2005; Smith et al., 2012); n = 3/n = 1 for total and low Cl_int, respectively.

Using the well-stirred model (eq. 2) with this data set to predict Cl_h from in vitro Cl_int, the conclusion of Hallifax et al. (2010) was again confirmed, with the incorporation of the free fraction yielding on average an underprediction of in vivo Cl (Fig. 2A; Table 1). Focusing on those compounds with reported low in vivo Cl (n = 13; ≤5 ml/min/kg; ∼25% of Q_h), underprediction of clearance was again observed when the free fraction was included in the well-stirred model with prediction accuracies of 38% and 54% within 2- and 3-fold, respectively (Fig. 2, B and C). Strikingly, when the free fraction was not included in the well-stirred model, a strong overprediction bias was observed (Fig. 2, A and B, open symbols). It should be noted that the reliability of the in vitro data (e.g., half-life extrapolation rules) used for these predictions of low-clearance compounds was not reported. Data from an in vitro substrate depletion assay becomes less reliable as extrapolation of the t_1/2 for substrate loss extends beyond the length of the incubation; therefore, overextrapolation from these studies is a possibility. For the aforementioned examinations, in vitro assay conditions were used, and subsequent data evaluation and analysis rules were unable to be assessed. It is clear that in vitro clearance must be sufficiently robust to accurately measure suspended hepatocytes following maximal 4-hour incubation. If this is not the case, then incubation conditions such as the number of cells or the length of incubation must be optimized to accurately estimate in vitro metabolic clearance.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

In vivo Cl_total versus in vitro Cl_h generated in hepatocytes in suspension (standard and relay) and plated monoculture. The solid lines represent the line of unity; the dashed lines represent 2- and 3-fold differences from the line of unity. (A) Full data set. (B) Data set focused on low Cl ≤ 5 ml/min/kg. (C) Data set focused on very low Cl ≤ 1 ml/min/kg for ease of visualization. ▪, suspended hepatocytes with f_uB correction, □, suspended hepatocytes without f_uB correction in the well-stirred model (eq. 2) (Hallifax et al., 2010); n = 17/n = 13 for total and low clearance, respectively. ▲, suspended hepatocyte relay model with f_uB correction, ∆, relay model without f_uB correction using the well-stirred model (eq. 2) (Di et al., 2012, 2013); n = 11/n = 7 for total and low clearance, respectively. •, plated hepatocytes with f_uB correction, ○, plated hepatocytes without f_uB correction in the well-stirred model (eq. 2) (Blanchard et al., 2005, Smith et al., 2012); n = 3/n = 2 for total and low clearance, respectively.

Relay Method.

Since short-term incubations with hepatocytes have shown good, albeit mixed, success in the prediction of the clearance of test articles when there is sufficient metabolism, one approach reported by Di et al. (2012, 2013) to resolve the incubation time limitation issue for suspended hepatocytes is to perform a series of up to five 4-hour incubations, or a relay. In this system, incubations of pooled hepatocytes at 0.5 × 10⁶ cells/ml are performed in 0.5 ml incubation volumes in 24-well plates. The incubation mixture at the end of a 4-hour incubation period is centrifuged to pellet the hepatocytes with the supernatant subsequently drawn off and frozen until the next incubation period. This supernatant is thawed the next day and 300 µl introduced into a freshly prepared cryopreserved hepatocyte suspension (0.5 × 10⁶ cells/ml) to continue the incubation period (Di et al., 2012, 2013). This technique allows the evaluation of low-clearance compounds over a period up to 20 hours (5 days of 4-hour incubation/day) (Di et al., 2013), and since every 4 hours a newly thawed hepatocyte suspension is used, the cells are metabolically competent over the total time period.

Evaluating the Di et al. (2012, 2013) data (Fig. 1A; Table 1), and using in vitro Cl_int values to compare with in vivo Cl_int (from total Cl using eq. 3), the predictions appear somewhat closer to the line of unity than when using a single incubation of suspended hepatocytes, albeit with fewer compounds evaluated. Moreover, there is a similar trend in underprediction bias with high-clearance compounds. This underprediction bias was not as pronounced for compounds with Cl_int values of ≤15 ml/min/kg. Using the well-stirred model to predict Cl_h, a pattern similar to that observed with suspended hepatocytes was observed (Fig. 2A). For the low-clearance compounds, the predictions that included f_uB in the well-stirred model (eq. 2) exhibited 56% and 78% prediction accuracies within 2- and 3-fold, respectively. While underprediction bias is observed with the relay method for high-clearance compounds, similar to that seen with suspended hepatocytes, this bias is reduced or eliminated for low-clearance compounds. A similarly performed relay system using cryopreserved rat or dog hepatocytes with compounds, which exhibited an array of clearance values from low to high, gave similar performance outcomes. This is an important finding, since the successful use of in vitro preclinical models to predict in vivo outcomes helps to assure the relevance of the human clearance predictions using human in vitro models (Di et al., 2013).

Contrary to using standard hepatocyte suspension incubations to predict clearance, this suspended hepatocyte relay model allows the extension of incubation times up to 20 hours for low-clearance compounds. Evaluation of this limited data set suggests that use of the relay system for low-clearance compounds exhibits more accurate predictions of in vivo Cl_int compared with traditional hepatocyte suspensions (78% and 54% predicted within 3-fold for relay versus traditional suspended, respectively) (Table 1). However, a limitation in using the relay method in a discovery setting is the use of many vials of expensive hepatocytes. In addition, it is relatively labor intensive (4 to 5 consecutive days in laboratory). To offset the cost incurred though the use of up to five vials of human hepatocytes, a large number of compounds would need to be evaluated simultaneously.

Plated Hepatocytes.

An alternative approach for extending the incubation time when using hepatocytes is to plate the hepatocytes into a monolayer or sandwich culture on a collagen-coated plate (12-, 24-, 48-, or 96-well design), typically at a cell density in the range of 0.3–1.0 × 10⁶ cells/ml (Lau et al., 2002; Griffin and Houston, 2005; Smith et al., 2012). Plating hepatocytes enables them to recover from the isolation procedure and develop more liverlike function over a longer period of time. In addition, there have been several reports of utilization of hepatocyte monolayer culture systems in sandwich configuration in an effort to develop and then maintain liver-specific function for extended periods (e.g., days to weeks), such as albumin secretion and enzyme activity (Dunn et al., 1989; Koebe et al., 1994). However, data in this model for the purposes of estimating metabolic drug clearance over an extended incubation period for low-clearance drugs are currently lacking. One of the first reports proposing the use of a hepatocyte monolayer culture system to enable more accurate assessment of clearance for low-turnover drugs was that by Griffin and Houston (2005). In this series of studies, the Cl_int values of seven compounds were compared in freshly isolated rat hepatocyte suspensions and monolayer cultures (Griffin and Houston, 2005). While the in vitro Cl_int of only two (tolbutamide and 7-ethoxycoumarin) of the seven compounds tested correlated well (within 2-fold) with the in vivo clearance data, the overall rank order of Cl_int between compounds was the same in both models. Interestingly, the prediction of in vitro Cl_int for high-turnover compounds was lower for monolayer cultures compared with that observed in hepatocyte suspension, presumably due to rate-limited uptake of compound into the static (i.e., no shaking) monolayer hepatocytes and the reduced surface area for drug diffusion, thereby impacting the apparent enzyme kinetics (increased K_m and/or decreased V_max). Comparatively, when drug was incubated with hepatocytes in suspension with shaking (900 rpm), more rapid dispersion may take place, resulting in higher observed intrinsic clearance. The possibility of differential transporter expression in suspended compared with plated hepatocytes contributing to differences in cellular uptake of drug is also a consideration, where lack of efflux transport out of suspended hepatocytes may lead to higher rates of metabolism. This is speculative, and certainly does not apply to drugs that are not substrates for transport. Intriguingly, monolayer cultures gave a higher estimation of in vitro Cl_int for the low-turnover drug S-warfarin compared with suspensions (following equal incubation time of 60 minutes). Thus, the conclusion was made that hepatocyte monolayer cultures may be more suitable for predicting the Cl_int of low-clearance compounds (below 0.1 µl/min/10⁶ cells), although this was based on a single observation using rat hepatocytes (Griffin and Houston, 2005), thus there is a need for further testing of this approach. Blanchard et al. (2005) also directly compared the metabolic clearance (fitting drug depletion to Michaelis-Menten constants) in suspensions and plated cultures of human hepatocytes prepared from three donors using naloxone, midazolam, and bosentan. As reported by Griffin and Houston (2005) using rat hepatocytes, the Cl_int of the high-clearance compound, naloxone, was higher in human hepatocyte suspensions compared with primary cultures for two of the three donors. The suggested explanation for this observation was the continuous mixing of the suspensions, resulting in an increased rate of drug uptake into the suspended cells relative to the primary cultures. This explanation is comparable to the aforementioned work of Griffin and Houston (2005), with assay conditions for suspension incubations (shaking at speeds of up to 900 rpm) being quite different from monolayer cultures (no shaking). It is also the authors’ experience that shaking versus nonshaking has a large impact on turnover rates (unpublished observations). In contrast, to the results with the high-clearance compound, the clearance of the two other (medium and low-clearance) compounds was comparable in both systems, which supports the notion that plated systems may be suitable for generating sufficient turnover for low-clearance drugs. Confounding these results was the fact that within the three donors tested there was variable activity, and only three test compounds were evaluated.

In a comprehensive assessment of metabolic profiles and temporal stability of phase I and II activities in human hepatocytes, Smith et al. (2012) evaluated the time-course of activity under various conditions in donor-matched livers that were incubated in both suspension formats, as well as following plating as adherent monolayers. In these studies, cryopreserved human hepatocytes from four donors were seeded in 24-well collagen-coated plates (0.4 × 10⁶ cells/well). Following a 4-hour attachment period, specific activities of CYP1A2 (phenacetin) and CYP3A (testosterone) were assessed following 4, 6, 8, 12, 24, 48, and 72 hours of preincubation. Activities for these two enzymes were sustained for roughly 12 hours after plating and, on average, activities dropped to 38% ± 6% and 50% ± 31% of the 4-hour postattachment activity by 24 hours for CYP1A2 and CYP3A4, respectively (Smith et al., 2012). Similar to their assessment of intrinsic activity for enzymes in suspended hepatocytes previously discussed (inhibitory E_max model), the mean IT₅₀ values from plated cells from the four donors were 21.3 ± 2.1 and 28.8 ± 20.4 hours for CYP1A2 and CYP3A, respectively (Smith et al., 2012). These results indicate increased enzyme stability relative to suspensions, where the IT₅₀ values for these two enzymes were 2.69 and 1.62 hours, respectively. While this assessment of the time-course of enzyme activity in cultured human hepatocytes was limited in scope (i.e., only CYP1A2 and CYP3A activity was studied), the authors speculate that monolayer cultures sustain liverlike physiology, which prolongs metabolic enzyme activity relative to suspended hepatocytes. Diazepam and tolbutamide were incubated in both suspended and plated hepatocytes from the same donor to support this conclusion. While the 2-hour incubation in suspended hepatocytes resulted in insufficient turnover for both substrates, in vitro Cl_int values of 0.917 ± 0.377 and 0.364 ± 0.253 µl/min/10⁶ cells were able to be obtained for diazepam and tolbutamide, respectively, following incubation in plated hepatocytes for up to 24 hours (Smith et al., 2012). However, the basal activities in hepatocytes in monolayer culture relative to suspension are still in question, and some believe that activities generally decrease substantially once cells are plated in monoculture, despite prolonged activity. Thus, the ability to capture clearance for diazepam and tolbutamide is likely a function of incubation time and not enhanced basal activity in a monolayer culture.

Similar to the analysis of data with hepatocytes in suspension, eqs. 1 and 2 were used to analyze the Blanchard et al. (2005) and Smith et al. (2012) data with respect to Cl_int of these three low-clearance compounds, which were common to the data set in Table 1 [midazolam; mean of four observations (Blanchard et al., 2005); diazepam and tolbutamide; mean of three observations (Smith et al., 2012)]. While this data set is very limited, with respect to Cl_int prediction there is clear underprediction bias for two of the three compounds, the exception being tolbutamide (Fig. 1A). When examining calculated in vitro Cl_h without f_uB correction, an overprediction of 3- to 6-fold is observed (Figs. 2, A and B; Table 1). However, when calculated Cl_h is corrected for f_uB, predictions are within 3-fold for midazolam (2.9-fold) and tolbutamide (1.4-fold), while a 6-fold underprediction for diazepam is observed (Fig. 2, A–C; Table 1), suggesting that correction for f_uB may be appropriate for low-turnover compounds.

Overall, this limited data set suggests that plated hepatocytes may be an in vitro system option for drug molecules requiring longer incubation times to achieve sufficient turnover. However, it is worth keeping in mind the possible loss in enzyme activity over the first 24 hours following plating (Smith et al., 2012). In addition, Richert et al. (2006) reported that gene expression for a number of phase I and II metabolic enzymes and transporters is profoundly decreased in cultured hepatocytes following plating, presumably due to oxidative stress responses. Current recommended practices are thus to begin incubations following visual verification of successful cell attachment, which tends to be in the 2–4 hour window following seeding, in an effort to capture enzyme activity near its peak, and minimize underpredictions of metabolic clearance. Also, due to the reported loss in enzyme activity over time, incubations using plated monoculture hepatocytes for clearance estimations should be limited to 24 hours or less, which may still be an insufficient incubation time for drug molecules that are metabolized extremely slowly.

Hepatocyte Coculture: HepatoPac (Hepregen).

To address the relatively short timeframe (e.g., 4–24 hours) of metabolic enzyme activity in standard hepatocyte suspension or monolayer cultures, technologies such as the hepatocyte-fibroblast coculture system have been developed (Khetani and Bhatia, 2008). In this innovative in vitro approach, a micropatterned coculture of human hepatocytes is created by seeding hepatocytes on collagen-patterned matrices that enable selective cell adhesion, a concept that began development in the early 1990s (Bhatia et al., 1994). Following a 2 to 3 hour period, unattached cells are washed off, leaving ∼10,000 adhered hepatocytes on 37 individual 500 µm collagen-coated islands (24-well configuration). This plate configuration is flexible, and can be scaled to 96-well plates. The next day, supportive mouse 3T3-J2 fibroblasts are seeded to create the cocultures. With this design, hepatocytes have been shown to maintain good morphology, viability, and liver-specific cytochrome P450 and phase II enzyme gene expression for up to 6 weeks (Khetani and Bhatia, 2008). In addition, liver-specific functions such as albumin secretion and urea synthesis were maintained for several weeks, compared with pure hepatocytes where these activities were only maintained for a few days. The longevity of drug-metabolizing enzyme activity and general cellular health has enabled numerous ADME applications, including the ability to conduct long-term incubations for low-turnover drug molecules (Chan et al., 2013) and to study the interplay between drug metabolism and transport (Ramsden et al., 2014).

The first report to evaluate the hepatocyte-fibroblast (HepatoPac) coculture system for measuring metabolism of low-turnover drugs was published by Chan et al. (2013). In this evaluation, 17 compounds with in vivo nonrenal clearance values ranging from low (≤5 ml/min per kg) to intermediate (>5 and <15 ml/min/kg) were incubated in the HepatoPac coculture system in a 96-well format, with 5000 hepatocytes per well. Incubations were carried out continuously for 7 days without media change, and three separate individual hepatocyte donors were evaluated to account for individual variability in enzyme activity. An evaluation of the Chan et al. (2013) data for HepatoPac was performed using the common modeling approaches outlined previously (Figs. 3 and 4; Table 1). This method yielded an excellent correlation between the in vitro Cl_int compared with the in vivo Cl_int (Fig. 3A). All five compounds (alprazolam, warfarin, tolbutamide, theophylline, and prednisolone), in which the modeled in vivo Cl_int was low (<15 ml/min/kg), predicted within 3-fold of the actual value (Fig. 3B). In scaling to in vitro Cl_h using f_uB binding correction in the well-stirred model (eq. 2), over the full range of compounds, within 2- and 3-fold prediction accuracies were 71% and 82%, respectively (Fig. 4A). Focusing on the 14 compounds that exhibit low in vivo Cl (≤5 ml/min/kg), the accuracies were 71% and 86%, respectively (Fig. 4B). It is noted that this favorable outcome was observed despite the fact that only 5000 hepatocytes were present in each well (96-well format) compared with the relay system in which 250,000 hepatocytes were used per well (24-well format). Even accounting for well size differences, there are significantly fewer hepatocytes present in the HepatoPac system. It should also be noted that the recalculated outcome compares even more favorably than that reported by Chan et al. (2013), in which comparisons to in vivo nonrenal Cl were made and only f_up was incorporated into the well-stirred model, whereas herein comparisons were made to total in vivo Cl using f_uB in the well-stirred model. The lack of impact of using in vivo hepatic Cl for comparison instead of total Cl likely reflects the Chan et al. (2013) observation that few of these compounds had a significant renal component in their clearance.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

In vivo Cl_int versus in vitro Cl_int generated in the coculture HepatoPac system (Chan et al., 2013). The solid lines represent the line of unity; the dashed lines represent 2- and 3-fold differences from the line of unity. (A) Full data set. (B) Data set focused on low Cl_int ≤ 15 ml/min/kg.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

In vivo Cl_total versus in vitro Cl_h generated in the coculture HepatoPac system. The solid lines represent the line of unity; dashed lines represent 2- and 3-fold differences from the line of unity. (A) Full data set. (B) Data set focused on low Cl ≤ 5 ml/min/kg. ▪, in vitro data generated with f_uB correction in the well-stirred model, □, in vitro data generated without f_uB correction in the well-stirred model (eq. 2).

Again, as in the suspended hepatocyte evaluation where no incorporation of protein binding into the well-stirred model resulted in significant overprediction, this observation also held true for the HepatoPac system. When f_uB was not factored into the calculations, predictions were considerably less accurate [only one (theophylline) of six compounds predicted within 3-fold for the low-clearance compounds], with a definite trend toward overprediction (Fig. 4B). Chan et al. (2013) noted that underprediction was observed when incorporating protein binding for the intermediate to high-clearance compounds; however, this is not the finding when using total Cl and f_uB. In any case, there were only six compounds in this category, and thus this observation merits further investigation.

Evaluation of this data set suggests that use of the HepatoPac system for the low-clearance compounds is an excellent choice for accurate predictions of in vivo Cl_total (100%, 71%, and 86% predicted within 3-fold for HepatoPac, suspension, and relay methods, respectively) (Table 1). It appears that incorporation of f_uB into the well-stirred model (eq. 2), significantly improved prediction accuracy across metabolic clearance categories. Limitations of using this method in a discovery setting include the higher expense and increased labor intensiveness relative to cryopreserved suspended hepatocytes. However, for later-staged chemical series where metabolism is consistently hard to measure, and for differentiation of late-staged candidates within the same series, a coculture model may currently be the best option.

Donor Selection Considerations.

As previously mentioned, variability in drug-metabolizing enzyme expression and activity between hepatocyte preparations is a concern, especially when considering that metabolic clearance rate data from hepatocytes are commonly used for prediction of human clearance as an important component of pharmacokinetic predictions. Thus, it is important to use donor hepatocytes that are thoroughly characterized from an activity perspective, in particular for polymorphically expressed cytochrome P450 (e.g., CYP2C9, CYP2C19, and CYP2D6), and for non-P450 enzymes such as glucuronosyltransferases and aldehyde oxidase (Hutzler et al., 2014). One approach to addressing this is to use pooled hepatocytes, typically composed of 10–20 donors. This approach is also commonly employed for hepatocyte suspensions and various subcellular fractions (e.g., liver microsomes or cytosol) to acquire data from what is perceived to be the average human. In plated hepatocyte systems, using pooled hepatocytes that are also plateable has been a challenge because not all donors provide plateable hepatocytes, and plating efficiencies may be variable between donors. Thus, a recommendation when using plated hepatocytes systems (standard plating or coculture systems) is to test multiple donors, as was conducted in the aforementioned HepatoPac clearance work where three separate donors were plated and clearance data from each were averaged for predictive purposes (Chan et al., 2013). A more novel approach is the use of pooled cryoplateable hepatocytes, which are now available from multiple vendors. The availability of pooled plateable hepatocytes theoretically means that average enzyme activities can be acquired similar to hepatocytes suspensions. The plating efficiency of each donor in the pool, as well as how the cells in this format communicate and how active they are, is a question that needs further validation and research to ensure the system truly represents an equal representation of each donor. Nonetheless, this plateable pooled approach is beginning to be used more often.

In Vitro Systems for Metabolite Generation.

Following the release of the Food and Drug Administration guidance for safety testing of drug metabolites (USFDA, 2008), and later the European Medicines Agency guidance on nonclinical safety studies (EMA, 2009), a number of publications evaluating the predictive power of human in vitro systems to predict human circulatory and excretory metabolites emerged (Anderson et al., 2009; Dalvie et al., 2009; Leclercq et al., 2009), exemplifying the importance of a strategy for this assessment. The evaluation of human circulating metabolites enables one to address the metabolite in safety testing thresholds by assessing whether major human metabolites (defined as those representing ≥10% total circulating drug material) have been appropriately exposed and tested for safety in preclinical toxicology studies (Nedderman et al., 2011; Gao et al., 2013). In addition, the Food and Drug Administration (USFDA, 2012) and European Medicines Agency (EMA, 2013) drug interaction guidances reiterate that an understanding of human metabolism, including both metabolic clearance pathways and circulating human metabolites, should be understood prior to large-scale clinical trials. For drugs with moderate to higher clearance, the assessment of human metabolic pathways is typically conducted early in drug development using in vitro systems such as microsomes and/or suspended hepatocytes. Optimally, an understanding of the primary human metabolic pathways should occur early in the drug development process, with in vitro cross-species metabolite data customarily being generated prior to preclinical toxicity studies and clinical testing, and in vivo metabolite scouting data (e.g., human plasma and urine) being generated as part of the single and multiple ascending dose (early phase 1 studies) human clinical trials. To validate this approach for a given drug candidate, in vivo metabolism is typically assessed in the relevant preclinical species to ensure that the selected in vitro system (whether microsomes or hepatocytes) appropriately represents the observed in vivo pathways. Additionally, phenotyping of the enzymes responsible for the primary metabolic clearance routes is evaluated using in vitro systems (such as recombinant enzymes with relative activity scaling factors, selective inhibitors of drug metabolism enzymes, and/or liver bank cross-liver correlation analyses) to determine the potential for victim-based drug-drug interactions (DDI) and the potential for pharmacokinetic variability in drug clearance due to polymorphic expression of drug-metabolizing enzymes.

As previously noted, the successful use of early ADME screens to reduce metabolic liability and enable once-a-day dosing has resulted in an increased percentage of lower turnover rates for compounds in pharmaceutical drug portfolios. This change has resulted not only in added challenges in predicting drug clearance, but also in understanding the human-relevant metabolites for these lower-clearance drugs. For low-clearance drugs, the traditional in vitro systems (e.g., HLMs or hepatocytes) are inadequate for generating a predictive metabolite profile for a drug that is slowly, but extensively, metabolized. The inability to measure drug metabolites in vitro can thereby force companies to proceed into clinical testing prior to being able to effectively evaluate the likely human metabolic pathways, including whether the species selected (typically, one rodent and one non-rodent) for in vivo toxicity testing were in fact representative of the human metabolism. This represents a potential safety risk for human subjects, highlighting the requirement for novel metabolically competent in vitro models where drugs can be incubated for sufficient periods of time to enable turnover more representative of the in vivo systems.

A thorough evaluation of the ability of traditional in vitro systems including HLMs, S9 fractions, and suspended hepatocytes to predict major human metabolites (defined as metabolites representing ≥10% of total circulating drug-related material or ≥10% of excreted dose) for 48 drugs possessing quantitative assessments of in vivo human metabolism was conducted by Dalvie et al. (2009). It should be noted that the reported compounds represent a broad range of clearances, and without access to reported clearance values it is not possible to comment specifically on the success rates of the various methods for low-clearance compounds. Based on this assessment, HLM, S9 fraction, and hepatocytes successfully produced major in vivo human metabolites (both circulating and excretory) for 33%, 44%, and 54% of the compounds assessed, respectively. The predictive value increased when considering only primary metabolites (metabolites only one biotransformation step removed from parent drug) but decreased when evaluating only secondary metabolites (metabolites greater than one biotransformation step removed from parent drug) with HLM, S9, and hepatocytes successfully yielding the primary metabolites 66%–69% of the time, but secondary metabolites only 12%–56% of the time. The lower predictive value for metabolites requiring multiple, sequential metabolic processes very likely reflects the limited incubation times, and thereby metabolite residence times, for these systems. It is also possible that the high concentrations of substrate often used in incubations (e.g., 10–20 µM or higher) for metabolite identification purposes may compete out low-level primary metabolite for enzyme, assuming the same enzyme is also involved in the secondary metabolism. Thus, while these traditional in vitro systems were fairly successful at generating relevant human metabolites, there is certainly much room for improvement.

Wang et al. (2010) expanded the Dalvie et al. (2009) analysis, evaluating 27 of the original 48 drugs (again representing a range of clearances) using a micropatterned coculture system from Hepregen. Incubations were conducted in 24-well plates using approximately 30,000 cells per well, and drug was incubated for up to 7 days with no change of media. Based on this analysis, HLM, S9, suspended hepatocytes, and cocultured hepatocytes successfully yielded the major excretory metabolites (n = 39 metabolites) 49%, 56%, 64%, and 82% of the time, and circulatory metabolites (n = 40 metabolites) 43%, 48%, 53%, and 75% of the time, respectively. These data suggest that the in vitro systems showed only slightly better predictive power for excretory metabolites than for circulatory metabolites, and that the predictive value increases when moving to the cocultured hepatocyte system with prolonged incubation times. The cocultured system again yielded primary metabolites (88–94% success rate) more effectively than secondary metabolites (67%–74% success rate), and showed notable improvement over the traditional HLM, S9, and suspended hepatocyte systems with shorter incubation times (69%–75% and 25%–57% success rates for primary and secondary metabolites, respectively). These data strongly suggest that the micropatterned, cocultured hepatocytes yield more in vivo relevant human metabolites than the standard in vitro systems. Again, it is not possible to comment specifically on the success rates for low-clearance compounds, but that said, the prolonged longevity of the cocultured hepatocytes would be expected to provide an advantage when assessing metabolic pathways for such low-clearance compounds. A nice example of the value of long-term incubations (up to 7 days) for two lower-clearance drugs using micropatterned cocultures was reported in Wang et al. (2010) (Fig. 5), where human-relevant metabolites for linezolid (in vivo Cl = 1 ml/min/kg) and ziprasidone (in vivo Cl = 5 ml/min/kg) were generated in more abundance when compared with 4-hour suspension incubations. Additionally, Ballard et al. (2014) evaluated a select subset of the previously reported compounds (a total of five compounds, three of which had not yielded all of the major in vivo human metabolites in the cocultured hepatocyte system) using the suspended hepatocyte relay method. For this evaluation, 4-hour incubations were conducted in 24-well plates using approximately 250,000 cells per well, with metabolite generation assessed after a total of 2, 3, or 5 relays (corresponding to 8, 12, and 20 hours of incubation). For these five compounds, the cocultured hepatocyte system and relay method generated 10 and 13 of the 16 relevant human metabolites, for success rates of 63% and 81%, respectively. Although this reflects a limited data set, it was noted that no new metabolites were generated after 2 relays (8 hours), suggesting that fewer relays may be required for metabolite generation than for clearance predictions.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

High-performance liquid chromatography UV chromatogram of linezolid (left) and ziprasidone (right) in micropatterned human hepatocyte cocultures (0, 4, and 48 hours, and 7 days) and suspended human hepatocytes (4 hours). L, linezolid; a and b, morpholine ring-opened acid metabolites; Z, ziprasidone; 1, N-dealkylziprasidone S-oxide; 2, ziprasidone S-oxide; 3, S-methyldihydroziprasidone. Used by permission from R.S. Obach, and previously published in Drug Metabolism and Disposition (Wang et al., 2010).

As a targeted example of a low-clearance compound, Ramsden et al. (2014) presented the metabolism of faldaprevir, an inhibitor of the hepatitis C virus, using the HepatoPac cocultured hepatocyte system. Faldaprevir was initially reported to be metabolically stable in vitro (Duan et al., 2012), whereas the ¹⁴C-human ADME study revealed two abundant monohydroxylated fecal metabolites together accounting for 41% of the dose (Chen et al., 2014). To better characterize the observed in vivo metabolism, ¹⁴C-faldaprevir was evaluated in the HepatoPac system using a 24-well plate format with approximately 32,000 cells per well and sampling times up to 96 hours. Metabolites generated using HepatoPac were quantitatively compared with those reported in feces for the ¹⁴C-human ADME study. HepatoPac successfully yielded five of the seven reported faldaprevir fecal metabolites, with two monohydroxylated products (M2a and M2b) representing the largest metabolites in the HepatoPac system (accounting for 5% and 10% of the dose, respectively) and in the feces (accounting for 22% and 19% of the dose, respectively). The remaining five fecal metabolites were minor, individually accounting for <2% of the dose. Interestingly, the HepatoPac system also yielded faldaprevir glucuronide (accounting for ∼2% of dose). Direct glucuronidation was a major clearance pathway in rat, but was not observed in the ¹⁴C-human ADME study, either reflecting the lack of direct glucuronidation in humans or excretion of faldaprevir glucuronide into feces with subsequent hydrolysis. The authors suggest that based on in vitro to in vivo correlation using rat HepatoPac incubations, the human HepatoPac results suggest that direct glucuronidation may play a role in faldaprevir clearance in humans, thereby providing additional mechanistic information not directly obtained from the ¹⁴C-human ADME study.

An overall evaluation of all of the reported data sets clearly suggests that use of the longer duration hepatocyte systems, such as the cocultured hepatocytes or the suspended hepatocyte relay method, provides improved predictive power for human-relevant metabolites, whether excretory or circulatory, over traditional in vitro methods. As more data are generated, it seems very likely that these systems will prove particularly valuable when assessing metabolic pathways, as well as enzyme phenotyping, for low-clearance compounds such as demonstrated by the faldaprevir case study. Additionally, coculture vendors such as Hepregen and HµREL now offer 24-, 48-, and 96-well plates containing both preclinical species and human hepatocytes to enable cross-species comparisons using a single plate. Given the trend toward lower-clearance drug candidates, such alternate in vitro systems are likely to become increasingly mainstream approaches.

Future Perspective.

Despite considerable progress in recent years in developing new in vitro approaches and technologies to enable longer incubation times and measurement of intrinsic clearance with higher confidence, more research is certainly needed to further improve extrapolation to the in vivo situation. For example, most of the studies performed to evaluate different models for estimating intrinsic clearance have done so using predominantly cytochrome P450 substrates. While this family of drug-metabolizing enzymes remains the most important in drug metabolism, other metabolic enzymes such as UDP-glucuronosyltransferases and aldehyde oxidase are becoming increasingly relevant (Hutzler et al., 2013; Oda et al., 2015; Zientek and Youdim, 2015). Thus, the long-term stability and activity of these and other non-P450 enzymes in the discussed monoculture and coculture models still deserve better characterization. A direct comparison of various in vitro models using the same lot of hepatocytes also has yet to be reported. Given known distinct donor composition and thus enzyme activities between lots of hepatocytes, the most direct comparison for the performance of the reviewed hepatocyte models is one in which the different models are compared using the same lot of hepatocytes. In addition, phenotyping for enzyme involvement in drug metabolism is especially challenging for low-turnover drugs, in particular for early discovery in the absence of knowledge about predominant metabolites or access to metabolite standards. An in vitro model that enables longer duration incubations with the appropriate concentration of cytochrome P450 inhibitor that maintains concentration (e.g., minimal inhibitor depletion) and selectivity throughout long-term incubations still needs to be developed. Finally, microfluidic flow systems coupled with cocultured hepatocytes (e.g., HµREL) seem to be an area of emerging science, and may be used more routinely in drug metabolism laboratories in the near future. For example, Novik et al. (2010) demonstrated that integration of fluid flow with a hepatocyte coculture system provides the ability to differentiate compounds for clearance rates, including buspirone (high clearance), sildenafil (medium clearance), and indomethacin (low clearance). In addition, this coculture dynamic flow system showed an increased metabolite formation rate for several cytochrome P450 enzymes compared with both static systems and the monoculture flow system. Most recently, a model integrating drug metabolism with disease (e.g., inflammatory liver models) using a three-dimensional liver bioreactor culture system has been published (Sarkar et al., 2015), where hydrocortisone intrinsic clearance was predicted that correlated with human data and key metabolites were observed. A similar three-dimensional dynamic flow model has been shown to maintain human hepatocyte function for up to 7 weeks while generating more reproducible estimates of clearance for 7-ethoxycoumarin compared with hepatocytes in suspension (Choi et al., 2014). Finally, human hepatocytes seeded in a microfluidic LiverChip system were investigated for their phenotype, expression levels of 22 drug metabolism and transporter genes, and metabolic capacity in comparison with two-dimensional static monocultures (Vivares et al., 2015). The LiverChip device consists of a liver tissue–engineered array of perfused open-well bioreactors, each containing an integrated micropump for controlling flow of culture medium. The results demonstrated stable cellular function, and mRNA levels of major genes involved in drug metabolism were maintained over a 7-day culture period. Also, cytochrome P450 activity was comparable or higher (depending on the cytochrome P450 enzyme) after 4 days of culture compared with 1 day of culture in two-dimensional static conditions. Certainly, the in vitro metabolism field seems to be moving toward dynamic flow systems, some being coupled with cocultures of hepatocytes as the model of choice for definitive characterization of drug candidates. However, more optimization and validation for in vitro-in vivo extrapolation purposes is needed for low-turnover compounds.

Overall, it appears that most simplistic models for studying drug metabolism (e.g., microsomes and cryopreserved hepatocyte suspension), while serving a very important purpose in drug discovery, have effectively peaked in terms of their predictive utility. Thus, the current wave of drug metabolism research is now focused on identifying liverlike models containing hepatocytes with prolonged drug metabolism capacity that are more predictive of the in vivo situation for metabolic clearance and metabolite identification, in particular, for low-turnover drug molecules.