Discussion

Accurate prediction of in vivo hepatic clearance is an integral part of successful and efficient drug development. Although IVIVE methodologies are commonly used to predict in vivo hepatic clearance, it has been demonstrated for decades by many investigators that there are significant limitations in predictability of clearance. A tendency for significant underprediction for primarily metabolized drugs has been reported (Bowman and Benet, 2016; Wood et al., 2017; Benet and Sodhi, 2020) and for drugs where clearance is rate limited by hepatic uptake (Zou et al., 2013; Kim et al., 2019). Moreover, it has been reported that the predictability of in vivo hepatic clearance is poor for the drugs that are highly bound to protein in blood (Francis et al., 2021). As we (Bowman and Benet, 2018) and Poulin et al. (2016) have reviewed, when conducting in vitro studies, many groups have reported that adding plasma or serum to hepatocyte incubations and adding albumin to microsome incubations can cause decreases in free Michaelis constant (K_m) values and improved IVIVE results (Ludden et al., 1997; Carlile et al., 1999; Baba et al., 2002; Shibata et al., 2002; Tang et al., 2002; Blanchard et al., 2004, 2005, 2006; Rowland et al., 2007, 2008a,b; Chao et al., 2009; Wattanachai et al., 2011, 2012, 2015; Gill et al., 2012; Mao et al., 2012; Walsky et al., 2012; Palacharla et al., 2017; Fujino et al., 2018; Bowman et al., 2019; Bteich et al., 2019; Kim et al., 2019; Nishimuta et al., 2019) (values presented in Supplemental Tables 1 and 2). Here, we review these 24 studies together to test the validity of this conclusion. For each of these investigations, we emphasize the importance of measuring incubational binding for all CL_int determinations, to accurately predict clearance based on the in vitro conditions in each laboratory as well as to account for nonspecific binding or high protein binding characteristics inherent for certain compounds. We also acknowledge the potential challenges of accurately measuring in vitro intrinsic clearance values for highly bound drugs in the presence of protein (particularly those that exhibit low in vitro turnover) and for poorly soluble or highly lipophilic drugs in the absence of protein due to the low amounts of available drug in the incubation.

First, as seen for all drugs in Table 1, the addition of serum only yields minimal increases in twofold predictability for hepatocyte blood clearance measurements (32.4% twofold predictability without serum addition and 41.1% with serum addition) and hepatocyte plasma clearance measurements (17.9% without serum and 27.5% with serum). Slightly higher percentage increases are observed for microsomal measurements, but still not good predictability as seen in Table 2 (20.8% without albumin versus 37.5% with albumin addition for blood measurements, and 20.8% without versus 33.3% with albumin addition for plasma measurements). From the 73 individual blood clearance predictions with human hepatocytes for 52 different drugs and the 24 individual blood clearance predictions with human microsomes for 15 different drugs, seven drugs were evaluated in both hepatocytes (14 predictions; 28.6% predicted within twofold without serum and this increased to 42.9% with serum addition) and microsomes (12 predictions; 25.0% predicted within twofold without albumin addition and this increased to 33.3% with albumin addition) as seen in Supplemental Tables 1 and 2. The percentage of predictions within twofold for this subset were very similar to that reported in Tables 1 and 2 for the entire data set. Comparison of low, moderate, and high extraction ratio for all drugs in Tables 1 and 2 yielded no obvious consistent differences based on ER for the four different measurements (hepatocyte and microsome, blood and plasma values). We believe that it is relevant that no dependence on ER was found, as low ER compounds would provide clearance measurements that are the same for all the hepatic models of organ elimination.

In Fig. 2 and Tables 3 and 4, we tested the hypothesis that predictability of in vivo hepatic clearance is poor for drugs that are highly bound to protein in blood as proposed by Soars et al. (2007) by using the cutoff of f_u, p < 0.1. For hepatocyte incubations, IVIVE success of CL_H, b within twofold for highly bound drugs increases from 14.7% without serum to 40.0% with serum, whereas for poorly bound drugs, IVIVE success decreases from 48.6% without serum to 42.1% with serum (Table 3). Similar results were observed for plasma measurements. In contrast, for microsomal incubations, IVIVE success for blood measurements within twofold for highly bound drugs does not change with albumin (28.6%), whereas poorly bound drugs IVIVE success increases from 10% without albumin to 50% with albumin (Table 4). Again, similar results are seen with plasma measurements. However, the contrasting trends for microsomes versus hepatocytes were surprising as it was expected that serum addition for highly bound compounds would improve IVIVE predictions in both matrices.

Most recently, Francis et al. (2021) also reported increased predictability in the presence of protein for highly protein bound drugs. There are significant differences between the Francis et al. (2021) analysis and the material presented here that readers should consider. Francis et al. (2021) evaluated both human and rat hepatocyte data, but no microsome measurements for 22 drugs, of which nine were from the published study of Kim et al. (2019) for acidic Extended Clearance Classification System (ECCS) (Varma et al., 2015) 1B substrates where clearance is rate limited by hepatic OATP uptake. The present analysis of 52 drugs for 73 experimental human hepatocyte investigations, includes all 22 drugs analyzed by Francis et al. (2021) from 36 experimental human hepatocyte measurements. Thus, it is possible that better predictability occurs for highly bound ECCS class 1B drugs (OATP substrates), but our results suggest this is not carried over to all highly bound drugs.

IVIVE success with and without protein addition was investigated with respect to BDDCS class (Wu and Benet, 2005). Both BDDCS class 1 and 2 drugs are primarily metabolized due to their high permeability characteristics; however, BDDCS class 1 drugs have high solubility, whereas BDDCS class 2 drugs have poor solubility. It has been proposed that the high solubility characteristics of BDDCS class 1 drugs allow for rapid membrane passage of drug at high soluble concentrations, deeming any potential involvement of xenobiotic transporters clinically insignificant. However, due to the low solubility characteristics of BDDCS class 2 drugs, the potential for transporter involvement cannot be ruled out as passive diffusion may not occur at high enough concentrations to overwhelm any potential transporter-mediated involvement in drug disposition. Thus, the primarily metabolized BDDCS class 2 drugs may or may not also be transporter substrates. BDDCS class 3 and 4 drugs are expected to be substrates for transporter mediated processes but differ in terms of solubility.

In the present study, the addition of serum into the hepatocyte incubations did not improve in vivo CL_H, b prediction for highly soluble BDDCS class 1 and 3 (39.5% without serum; 43.2% with serum) (Table 5). Poorly soluble BDDCS class 2 and 4 drugs exhibited lower predictability without serum (21.4%), which increased to a comparable 37.9% with serum addition (Table 5). We had speculated that the presence of protein in in vitro incubations may be enhancing assay conditions for low-solubility compounds, by preventing nonspecific binding and potentially boosting drug concentrations available for metabolism. We also analyzed the effect of albumin addition in the microsomal incubations on IVIVE predictability, where the potential for protein-facilitated uptake is not a possibility due to the lack of cell membranes and transporters in such incubations. No data for BDDCS class 3 and 4 drugs are available. In Table 6 for the microsome incubations, improvement of IVIVE success is noted with serum addition, with a marked increase from 28.6% to 57.1% for BDDCS class 1 blood measurements but no change from 14.3% for plasma measurements. However, only seven BDDCS class 1 drug measurements were available in microsomes (versus 17 for BDDCS class 2) so the class 1 change in blood measurements may be suspect. Class 2 compounds displayed increases in IVIVE predictability with albumin addition (from 17.6% to 29.4% for blood values and 23.5% to 41.2% for plasma values), although overall predictability within twofold is still quite low with protein addition.

In Fig. 3, we plot the individual unbound intrinsic clearance values from hepatocyte incubations (Fig. 3A) and microsome incubations (Fig. 3B) with respect to BDDCS class, visually highlighting the tendency of the addition of albumin into the incubations to increase unbound intrinsic clearance of both BDDCS class 1 and class 2 drugs in both hepatocyte and microsomal incubations. Although the mechanism of the protein effect in hepatocytes has frequently been cited as protein-facilitated uptake, in microsomes it has been suggested that the protein effect is due to albumin trapping the unsaturated long-chain fatty acids that can inhibit the activity of cytochrome P450s and UDP-glucuronosyltransferases (Rowland et al., 2007, 2008a,b; Wattanachai et al., 2011, 2012, 2015; Palacharla et al., 2017). The unsaturated long-fatty acids are present in the hepatocytes, but most of them are trapped to fatty acid-binding proteins, and thus it is thought that the concentration of the unsaturated long-fatty acids would not be high enough to inhibit cytochrome P450 and UDP-glucuronosyltransferase isoforms activities (Fujino et al., 2018). It would be expected that trapping of the inhibitory fatty acids via protein addition in microsomal incubations would increase in vitro unbound CL_int values of BDDCS class 1 and 2 drugs equally, as both classes are primarily metabolized. The microsomal data in Table 6 confirms that the protein effect is apparent for both BDDCS class 1 and 2; however, a greater percentage of class 2 drugs displayed increases in unbound intrinsic clearance greater than twofold (70.6% versus 42.9%), suggesting that additional factors (such as improvement of solubility) may also be contributing.

Our laboratory has recently pointed out (Benet and Sodhi, 2020) that although values of f_u, b are required to predict in vivo blood clearance using the IVIVE method, there is potential for error in calculation of f_u, b from experimental determination of f_u, p and measurements of R_BP ratios. Experimental f_u, b values are usually calculated by eq. 5.

To derive eq. 5, one must assume that free drug concentrations are equal in the blood cells and the plasma. However, transporters have been identified in the blood cell membrane, and this assumption may not hold for transporter substrates expressed in erythrocytes. Therefore, it is possible that this inaccuracy of f_u, b may impact poor IVIVE prediction (Benet and Sodhi, 2020). And thus, we also conducted the analysis of CL_H, p IVIVE to avoid using f_u, b. However, the CL_H, p analyses did not result in improved IVIVE predictions as compared with the CL_H, b analysis. We generally have greater faith in the blood level measurements. When attempting to use the plasma values, we had to assume that plasma flow was (1-hematocrit) multiplied by hepatic blood flow with a resulting plasma flow of 11.4 ml/min/kg as given in eq. 3. It is generally accepted that the CL_H, p value cannot exceed Q_H, p. However, we found that 18 of the 49 drugs for which plasma measurements were analyzed, CL_H, p values were greater than the Q_H, p value of 11.4 mL/min/kg (Supplemental Tables 1 and 2). In contrast, three of the 49 drugs reported Q_H, b values greater than 20.7 mL/min/kg. The difference in frequency of the plasma clearances exceeding plasma hepatic flow is sixfold greater than when it occurred for the blood clearances exceeding blood flow.

We also investigated the effect of charge class on the predictability within twofold of observed in vivo hepatic clearance as depicted in Table 7 for hepatocyte measurements and Table 8 for microsome measurements. In the presence of protein, there appears to be a modest increase in predictability for neutrals in hepatocytes but not for microsomes and a doubling of the percentage accuracy for acids in both hepatocytes and microsomes, but accuracy is never greater than 50%. For bases, serum addition caused little improvement in accuracy in hepatocytes. Only one base was studied in microsomes. It is interesting to note that the IVIVE success in the presence of protein between hepatocytes and microsomes was similar (ranging from 31.6%–50%) for acids, a class of drugs for which the rate-limiting hepatic uptake is well-documented throughout the literature (Kim et al., 2019). One may have anticipated that if rate-limiting steps based on the Extended Clearance Model (as we reviewed, Benet et al., 2018) or ECCS were relevant, that there would be differences between the IVIVE success of hepatocytes (where the possibility of rate-limited uptake exists) versus microsomes (where there is no such possibility), yet IVIVE success was approximately 31.6%–50% for both matrices in the presence of protein.

Returning to ECCS class 1B drugs, it is instructive to note that Bi et al. (2021) investigated eight ECCS class 1B drugs in human hepatocytes (with the addition of plasma compared with buffer) that were also reported by Kim et al. (2019) for studies in human hepatocytes (with the addition of 5% HSA compared with buffer). Both Bi et al. (2021) and Kim et al. (2019) report that adequate predictions can only be achieved by using a scaling factor, i.e., 2.44 (Kim et al., 2019) and 3.8–5.3 (Bi et al., 2021). This approximately twofold difference in scaling factor reflects that significant differences in predictability are found in different, highly respected laboratories for the same drugs and emphasizes the need for future studies comparing the effects of different amounts of added protein as well as different experimental systems. It is important to note that the unbound CL_int of Kim et al. (2019) for uptake of OTAP substrates was determined using the oil-spin method, whereas CL_int values of the other data sets cited in our analysis were mainly determined by the conventional hepatocyte stability assay (measuring the compound disappearance in hepatocyte incubation). The two methods could lead to significant differences in underestimation of hepatic clearance for hepatic uptake transporter substrates, although neither method gives IVIVE predictions within twofold with protein addition. Even more recently, the Sugiyama laboratory, guided by its understanding of the facilitated-dissociation model obtained from the 10 highly protein-bound anionic drugs analyzed by Kim et al. (2019), proposed a simplified method to predict the hepatic uptake clearance in vivo (in the presence of the physiologic 5% albumin concentration) for these 10 drugs (Miyauchi et al., 2021). In this simplified method, Miyauchi et al. (2021) introduced the concept of the Extent of Facilitated Dissociation (EFD) where the increase of intrinsic permeability clearance of unbound drug via influx into hepatocytes at percentage of albumin concentration (PS_{u,inf, 5%}) for these 10 anionic substrates may be predicted from the measured intrinsic uptake clearance in the absence of albumin (PS_u,inf,0%) using eq. 6.

They reported that an EFD value of 11 μM and a dissociation equilibrium constant for the binding of a ligand to albumin value of 46 μM could be used for all the 10 substrates rather than measuring the 5% albumin PS value. Miyauchi et al. (2021) report that the prediction of hepatic clearance obtained by the simplified method was comparable to those obtained by the previously proposed method utilizing the measurement of the protein binding and hepatic uptake at varying concentrations of albumin. Whether this facilitated dissociation model for a particular subset of drugs (highly protein-bound OATP substrates) (Kim et al., 2019; Miyauchi et al., 2021) and other studies proposing protein-facilitated uptake via the presence of an albumin-specific receptor on the hepatocyte surface, rate-limiting dissociation of ligand from a protein-ligand complex, rate-limiting diffusion of ligand through an unstirred water layer, interactions with the hepatocyte cell surface (as reviewed by Poulin et al., 2016 and Bowman and Benet, 2018), as well as our proposed transporter-induced protein binding shift (Bowman et al., 2019), we emphasize that all of these studies do not include the appropriate controls, i.e., drugs giving good IVIVE predictions without the addition of protein. For example, another very recent article by Bteich et al. (2021) shows excellent improvement in hepatic clearance predictivity for fluoxetine and perampanel using a newly proposed extrapolation approach, but like other proposed methodologies does not validate the approach by testing a drug showing good predictability without protein addition. We are not suggesting that this last study and others do not show increased predictability with the addition of protein, just that the conclusions regarding the proposed hypotheses being tested by these investigators (regardless of the methodology used) can only be validated by also including drugs that exhibit good predictability without protein addition. In all the analyses reviewed thus far, there appears to be minimal support for the hypothesis that protein addition provides a clinically meaningful increase in IVIVE predictability when all published drugs investigated in hepatocyte and microsomal incubations are considered.

When a presentation such as this proposes that a well-accepted methodology has little use, the scientific community, of course, questions the data upon which this conclusion is made. Here, we only review the published data. However, in our very recently accepted manuscript (Benet and Sodhi, 2021), we report that the original concept of the “albumin mediated hepatic uptake” is based on invalid assumptions and that, in fact, there is no theoretical basis upon which IVIVE should be expected to work, as it doesn’t. However, the scattered literature data upon which our conclusions here are based probably requires that prospective, well-controlled experimental studies in a single laboratory will be needed to rigorously determine whether adding protein in any form to in vitro determinations of clearance improves the IVIVE and whether any improvement can be tied to chemical structure, physical properties (e.g., solubility, charge), and clearance mechanism, before the scientific community abandons the concept. We would welcome such a study.