Abstract
When predicting hepatic clearance using in vitro to in vivo extrapolation (IVIVE), microsomes or hepatocytes are commonly used. Here, we examine intrinsic clearance values and IVIVE results in human hepatocytes and microsomes for compounds metabolized by a variety of enzymes. The great majority of CYP3A4 substrates examined had higher intrinsic clearance values in microsomes compared with hepatocytes, whereas the values were more similar between the two incubations for substrates of other enzymes. We hypothesize that this may be due to interplay between CYP3A4 and the efflux transporter P-glycoprotein, as they have been shown to exhibit coordinated regulation. When examining the prediction accuracy for substrates of other enzymes between microsomes and hepatocytes, average fold errors as well as overall error were similar, demonstrating once again that IVIVE methods are not adequately defined and understood.
SIGNIFICANCE STATEMENT For CYP3A4 substrates, microsomes give markedly higher predictive in vitro to in vivo extrapolation than for other metabolic enzymes, which is not found for hepatocytes. We hypothesize that this could be a result of CYP3A4–P-glycoprotein interplay or coordinated regulation in hepatocytes that would not be observed in microsomes.
Introduction
Despite hepatic clearance playing an important role in the pharmacokinetics and pharmacodynamics of molecules, accurately predicting the parameter during drug discovery is still challenging. Many have found inaccuracies when implementing in vitro to in vivo extrapolation (IVIVE), where intrinsic clearance (CLint) is measured using microsomes or hepatocytes and scaled to a predicted in vivo hepatic clearance (CLH) using scaling factors and a model of hepatic disposition (Bowman and Benet, 2016; Wood et al., 2017).
It has become apparent that the mechanisms behind the current IVIVE disconnect must be discovered and considered during the IVIVE process. Many ideas surrounding the systematic underprediction have been presented, including donor variability (Floby et al., 2009), liver sample preparation and viability (Fisher et al., 2001), protein-binding discrepancies (Obach, 1999; Kochansky et al., 2008), and ignoring extrahepatic metabolism (Houston and Carlile, 1997; Chiba et al., 2009).
Although microsomes are routinely used for IVIVE, as these subcellular fractions are easy to prepare and store, it would be expected that hepatocytes would yield more accurate clearance predictions given that hepatic transporters are present in hepatocytes and not in microsomes (Lam and Benet, 2004). One review found that human hepatocytes underpredict clearance by 3- to 6-fold, whereas microsomes underpredict by 9-fold (Chiba et al., 2009). However, more recent studies have found the overall error between the systems to be more similar (Bowman and Benet, 2016; Wood et al., 2017). Furthermore, it has recently been noted that there is CLint-dependent underprediction (Hallifax et al., 2010) as well as CLH-dependent underprediction (Bowman and Benet, 2019) with increasing clearance, a finding that is more marked for data generated in hepatocytes.
Recently, Wood et al. (2017) compiled predicted clearance values from human microsomes and hepatocytes, and El-Kattan et al. (2016) compiled primary enzyme information for compounds classified in the Extended Clearance Classification System. Here, using both data sets, we examine the role different metabolic enzymes may play in the values generated in the two systems and in IVIVE accuracy.
Materials and Methods
Scaled in vitro CLint values generated in human hepatocytes and microsomes (and corrected for incubational binding) and in vivo CLint values were taken from Wood et al. (2017). In this source, in vivo CLint was back-calculated using the well stirred model accounting for protein binding. CLH was determined by subtracting renal clearance from total clearance in relevant cases where data were available.
Primary enzyme information was taken from El-Kattan et al. (2016). Of the 101 compounds with human hepatocyte values listed by Wood et al. (2017), 48 had primary metabolizing enzyme information reported by El-Kattan et al. (2016), and of the 83 compounds with human microsome values, 45 had primary metabolizing information reported. It should be noted that the enzyme assignments are qualitative, and more than one enzyme could be involved.
The bias in predictions was determined by calculating the average fold error (AFE):The accuracy of predictions was determined based on whether the predicted CLint values fell within 2-fold of the observed CLint values (Houston and Carlile, 1997):The 54 drugs investigated, organized by main metabolizing enzyme, are listed in Supplemental Table 1, with human hepatocyte and human microsome CLint values and the AFE for both hepatocyte and microsome predictions.
Results
When comparing CLint values generated in hepatocytes versus microsomes, it became apparent that the values generated in microsomes for CYP3A4 substrates are often higher than those generated in hepatocytes (Fig. 1A). Of the 14 CYP3A4 substrates that had values generated in both systems, 13 had higher CLint values in microsomes, where diltiazem was the only example in which the value was higher in hepatocytes (35 vs. 27 ml/min per kilogram in hepatocytes vs. microsomes). In comparison, the values between the two systems were more similar for CYP2C (n = 7) and CYP2D6 (n = 7) substrates and fell on both sides of the line of unity (Fig. 1B). For uridine diphosphate glucuronosyltransferase (UGT) substrates (n = 8) (Fig. 1C), lower-clearance compounds fell on both sides of the line of unity, whereas hepatocytes yielded higher values as CLint increased.
These CLint values generated for the same drugs in both hepatocytes and microsomes were then compared with in vivo CLint to see the effect on IVIVE accuracy (Fig. 2; Table 1). There were 39 overlapping compounds, and CYP3A4 substrates had the highest AFE of 5.88 in hepatocytes (compared with 2.03 in microsomes). Although only three drugs were substrates of CYP1A2, it is interesting to note that the AFE was the lowest in both systems.
Given the potential of CLint-dependent underprediction with hepatocytes in particular, the highest CLint compounds were examined across enzymes. With the drugs examined here, substrates of CYP3A4 and CYP2D6 had similar observed highest CLint values (Table 2). Despite having similar observed values, the difference in predicted values between microsomes and hepatocytes was not as marked for CYP2D6 substrates as for CYP3A4. For instance, with midazolam (CYP3A4) with an observed CLint of 390 ml/min per kilogram, the CLint measured in microsomes was 7.50-fold higher than the value measured in hepatocytes. With carvedilol (CYP2D6) with a similar observed CLint value of 427 ml/min per kilogram, the CLint measured in microsomes was only 1.30-fold higher than the value measured in hepatocytes. In Fig. 3, the difference between the values generated in the two systems appeared to more notably increase with observed CLint only with CYP3A4 substrates.
Finally, all compounds with primary enzyme information were examined (n = 48, hepatocytes; n = 45, microsomes). When examining the number of compounds with accurate, under-, and overpredictions (Table 3), almost all of the errors were due to underprediction, agreeing with the systematic underprediction noted throughout the field. Substrates of CYP3A4 had the most accurate predictions in microsomes. When examining the human hepatocyte AFEs (Table 4), aldehyde oxidase was an obvious outlier, with only one compound as an example (zaleplon), which had a 22.1-fold error. Excluding aldehyde oxidase and with the additional compounds added, CYP3A4 still had the highest AFE of 7.87. Upon further inspection, there was an outlier in this category as well (nitrendipine has a 668-fold error); however, after removing this drug, the AFE remained the highest at 5.96. When examining the AFE for human microsomes, CYP3A4 had the second-lowest AFE of 2.08. The highest AFE was for UGT with 7.54; however, here again there was an outlier (fenoprofen had a 159-fold error), and after removing it, the AFE dropped to 4.88, a value more comparable to that of the other enzymes.
Discussion
Although predicting hepatic clearance with IVIVE using hepatocytes and microsomes is commonly done, there is still systematic underprediction. When comparing CLint values measured in microsomes and hepatocytes, it became apparent that CYP3A4 substrates frequently had higher CLint values in microsomes. Stringer et al. (2008) also saw similar results for five CYP3A4 substrates and found that the CLint values for the same drugs were 10- to 50-fold higher in microsomes than hepatocytes. Foster et al. (2011) measured the clearance of compounds in hepatocytes and microsomes from the same donor livers and found that the CLint values for the highest clearance substrates—in their case, substrates of CYP3A4—were higher when measured in microsomes versus hepatocytes, but the values were comparable between the systems for a low-clearance CYP3A4 substrate. The authors hypothesized there could be cofactor rate limitation or permeation limitation for high clearance compounds in hepatocytes. For the compounds examined here, substrates of CYP3A4 and CYP2D6 had similar observed high CLint values, but the difference in predicted values between the systems was not as marked for CYP2D6 substrates as for CYP3A4, making the cofactor limitation hypothesis less likely.
Although CYP3A4 substrates yielded higher CLint values in microsomes, inherently microsomes are not mechanistically better predictors. The overall percentage of inaccurate predictions (77% for hepatocytes and 69% for microsomes) and the AFE (5.19 for hepatocytes and 3.47 for microsomes) were still high, emphasizing that present IVIVE methods are not adequately understood.
When examining the AFE for overlapping compounds as well as all compounds, CYP3A4 had the highest AFE in hepatocytes and CYP1A2 had the lowest AFE in both systems. If extrahepatic metabolism is ignored and hepatic clearance is assumed to be total clearance (or if only renal clearance is subtracted from total clearance), then only using measurements from liver microsomes or hepatocytes could lead to IVIVE underpredictions. De Kanter et al. (2004) found that multiorgan (liver, lung, kidney, small intestine, colon) precision-cut slices from rats could predict drug clearance better than when only considering the contribution of the liver. Given that CYP1A2 is the only enzyme of those examined with no evidence of intestinal metabolism whereas CYP3A4 contributes 80% to the total cytochrome P450 abundance in the intestine (Paine et al., 2006), this could potentially explain the trend noted with hepatocytes. However, a similar trend would have been expected in microsomes, and in vivo CLH values are typically taken from intravenous studies when available, making a potential contribution of intestinal metabolism smaller. While genetic polymorphisms could also be an explanation for the errors seen for several of the examined enzymes, both over- and underpredictions would be expected (Chiba et al., 2009), which was not found here.
When considering the difference in CYP3A4 CLint values between hepatocytes and microsomes (and the corresponding different AFEs between the systems), a possible explanation could be due to transporter-enzyme interplay that is present in hepatocytes but not in microsomes. Several publications have noted the common substrate specificity, tissue localization, and coinducibility of CYP3A4 and the efflux transporter P-glycoprotein (P-gp) and proposed that the enzyme and transporter could play complementary roles in the absorption, distribution, metabolism, and elimination of compounds (Wacher et al., 1995; Benet et al., 1996; Hall et al., 1999; Zhang and Benet, 2001). In the intestine, where drugs will contact P-gp prior to CYP3A4, they can be effluxed back into the lumen before diffusing into enterocytes to be metabolized, forming more metabolites than without P-gp (Benet, 2009). In the liver, where drugs will contact CYP3A4 prior to P-gp, drugs will be pumped out by P-gp, forming fewer metabolites than without P-gp (Benet, 2009). Therefore, CYP3A4 substrates evaluated in hepatocytes may have lower CLint values because they can be effluxed by P-gp, whereas when they are evaluated in microsomes with no P-gp present, they are subject to more metabolism by CYP3A4. Lam and Benet (2004) found that elacridar, a P-gp inhibitor, had no effect on digoxin metabolism in rat microsomes, whereas in rat hepatocytes, the P-gp inhibitor caused increased metabolism at low concentrations (1 μM) that did not change digoxin uptake. Cummins et al. (2002) also demonstrated that in CYP3A4-transfected Caco-2 cells, inhibiting P-gp reduced CYP3A metabolism in the apical to basolateral direction (similar to the intestine) but increased metabolism in the basolateral to apical direction (similar to the liver). Bow et al. (2008) found that P-gp is internalized after hepatocyte isolation and suggested that “drug efflux from suspended hepatocytes are not an appropriate system to study apical efflux/canalicular excretion of drugs.” However, they did not negate the finding of Lam and Benet (2004) and others that there can be a transporter effect potentially from the internalized proteins in hepatocytes that is different than in microsomes. We hypothesize and will test whether this transporter internalization could have a negative effect on metabolic activity of CYP3A4, potentially even for drugs that are not strong substrates of P-gp as a result of coordinated regulation.
In conclusion, when examining CLint values generated in microsomes and hepatocytes, values were almost always larger for CYP3A4 substrates in microsomes, perhaps due to CYP3A4–P-gp interplay present in hepatocytes but not microsomes. Although IVIVE predictions were better for these substrates in microsomes, overall percentage inaccuracies were similar between the two systems, highlighting that IVIVE methods are not adequately understood.
Authorship Contributions
Participated in research design: Bowman, Benet.
Conducted experiments: Bowman.
Performed data analysis: Bowman, Benet.
Wrote or contributed to the writing of the manuscript: Bowman, Benet.
Footnotes
- Received June 28, 2019.
- Accepted September 13, 2019.
C.M.B. was supported in part by the Pharmaceutical Research and Manufacturers of America Foundation Pre Doctoral Fellowship in Pharmaceutics and the National Science Foundation Graduate Research Fellowship Program [Grant 1144247]; L.Z.B. is a member of the UCSF Liver Center supported by the National Institutes of Health [Grant P30 DK026743].
↵This article has supplemental material available at dmd.aspetjournals.org.
Abbreviations
- AFE
- average fold error
- CLH
- total hepatic clearance
- CLint
- intrinsic clearance
- IVIVE
- in vitro to in vivo extrapolation
- P-gp
- P-glycoprotein
- UGT
- uridine diphosphate glucuronosyltransferase
- Copyright © 2019 by The American Society for Pharmacology and Experimental Therapeutics