INTRODUCTION PATRICK POULIN, 1,2 SAMI HADDAD 2

Transcription

1 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism Albumin and Uptake of Drugs in Cells: Additional Validation Exercises of a Recently Published Equation that Quantifies the Albumin-Facilitated Uptake Mechanism(s) in Physiologically Based Pharmacokinetic and Pharmacodynamic Modeling Research PATRICK POULIN, 1,2 SAMI HADDAD 2 1 Consultant, Québec city, Québec, Canada 2 Department of Occupational and Environmental Health, School of Public Health, IRSPUM, Université de Montréal, Québec, Canada Received 1 September 2015; revised 18 September 2015; accepted 18 September 2015 Published online in Wiley Online Library (wileyonlinelibrary.com). ABSTRACT: The impact of albumin concentration on the uptake of drugs in cells might involve mechanisms going beyond the free drug concentration hypothesis. Proceeding from the assumption that both the unbound and protein-bound drug fractions can be available for uptake, several authors have argued that the uptake of highly bound drugs in cells might be driven mainly by the albumin-facilitated uptake mechanism(s). Hence, a novel approach quantifying the additional contribution of the protein-bound drug complex and ph gradient effect in diverse in vitro-to-in vivo extrapolation (IVIVE) procedures of drug uptake and clearance has been proposed and extensively validated by Poulin et al. (2015. J Pharm Sci. Epub ahead of print); this approach consisted of replacing the unbound fraction in plasma (fu p ) with an adjusted fu p value (fu p-adjusted ). After a second review of literature, the objective of the present study was to perform further validation exercises of the concept of fu p-adjusted by using additional case examples of IVIVEs that covered diverse drug properties and experimental settings with varied albumin concentrations (e.g., perfused liver, isolated and suspended hepatocytes, and cultured cells overexpressing transporters). Again, the novel IVIVE method based on fu p-adjusted was the best-performing prediction method of the uptake rate (or clearance) as a function of protein binding compared with the conventional method based on the fu p theory (absolute average fold error of 1.4 vs. 7.4). Therefore, the present study confirms the utility of fu p-adjusted compared with fu p in IVIVE procedures for drugs highly bound to albumin, and the improvement was observed particularly in the higher range of albumin concentrations. From these findings, we may conclude that uptake of these drugs in cells is primarily driven by the albumin-bound form. Consequently, it is suggested to estimate the uptake kinetic parameters with cell-based assays incubated in 100% human serum or to make a correction while the experimental data are generated either without albumin or with varied albumin concentrations, in order to predict more accurately the in vivo conditions in physiologicallybased pharmacokinetic and pharmacodynamic (PBPK/PD) modeling research. Overall, the protein-facilitated uptake mechanism(s) could be another paradigm shift in addition to a previous paradigm related to the ph gradient effect. C 2015 Wiley Periodicals, Inc. and the American Pharmacists Association J Pharm Sci Keywords: ADME; clearance; disposition; drug safety; hepatocytes; IVIVEs; isolated perfused liver; PKPD; pharmacokinetics; pharmacodynamics INTRODUCTION An organ (e.g., liver) may remove a diverse group of drugs from the circulation with remarkable efficiency, despite extensive binding of these solutes to the extracellular proteins (e.g., albumin). Poulin et al. 1 have recently argued that uptake of highly bound drugs in cells might be driven by the albumin-facilitated uptake mechanisms. Several published experimental studies confirmed that for several highly protein-bound drugs, the real free drug concentration taken by the cells in the presence of extracellular binding proteins could be substantially higher than the calculated free drug concentration from the unbound fraction in plasma (fu p ) measured in vitro Hence, the hepatic removal of drugs in the isolated perfused rat liver (IPRL) (and hepatocyte suspensions) was faster than expected when serum (plasma) is used as the perfusate compared with serumfree perfusate This is also true for the removal of drugs Correspondence to: Patrick Poulin ( patrick-poulin@videotron.ca) Journal of Pharmaceutical Sciences C 2015 Wiley Periodicals, Inc. and the American Pharmacists Association in the isolated perfused rat kidneys. 12,13 In addition, recent drug drug interaction (DDI) studies at the transporter level were performed in cultured Madin Darby Canine Kidney epithelial (MDCK) cells overexpressing organic cation (OCT2) or anion transporters (OAT1B1); thus, the later DDI studies also indicated that for highly protein-bound inhibitors, the measured IC50 (the concentration of an inhibitor where the response or binding is reduced by half) measured in 100% human serum was significantly lower than those estimated based on fu p and the intrinsic transporter IC50 assessed in a serum-free medium. 14 Conversely, for drugs with a low degree of albumin binding, the albumin-mediated effects were not observed in these experimental settings. 1,9,17 20 From these observations, the authors concluded that the impact of albumin concentration on the estimation of uptake of drugs in cells might involve mechanisms going beyond the free drug concentration hypothesis. Therefore, it was suggested that the protein-bound drug concentration was also involved in the cellular uptake but for the compounds that are highly bound to albumin particularly in the higher range of albumin concentrations, which supports the notion of albumin-facilitated uptake mechanism(s). The net 1

2 2 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism result of the albumin-facilitated mechanism(s) is that more unbound drug may become available at the cell surface for uptake, that is, drug can be transported in cells in two ways: one way is the unbound form and the other way a cell surface-mediated enhanced dissociation from the bound form. 1,21 The extracellular protein-mediated uptake mechanism(s) of drugs has mostly been observed with the hepatocytes, cardiac myocytes, and cells from kidneys. Similarly, different experimental conditions either performed under in vitro or in vivo conditions with rat and human materials have demonstrated the presence of an albumin-facilitated uptake effect The mechanism(s) was not specific only for serum albumin but also for other serum proteins (e.g., globulins). 4,6,7 In this case, the general trend is that the albumin-facilitated uptake mechanism(s) is predominant for an extracellular binding protein with a net positive charge compared with a net negative charge, and, hence, the experimental studies using alpha-acidglycoprotein (AAG) or negatively charged albumin showed no protein-mediated uptake mechanism(s) compared with positively charged albumin and globulins. 1,6,7,21 However, the exact mechanism by which these binding proteins may facilitate the uptake of a drug in cells is still unknown, but a recent review of literature compiled the mechanisms that have been proposed by diverse authors. 1 To date, the most plausible mechanism seems to be related to the presence of ionic attractions between the protein drug complex and cell surface to result in a marked reduction in the diffusional distance of the extracellular protein drug complex. 1,21 Whichever mechanism explaining the protein-facilitated uptake process, it is almost certain that more drug is delivered into the cells than expected from the conventional ph gradient effect theory of the unbound drug. 1 The additional protein-facilitated uptake mechanism(s) needs, therefore, to be quantified in the in vitro-to-in vivo and in vitro-to-in vitro extrapolation (IVIVE) procedures, as the concentrations of extracellular binding proteins used in cell-based in vitro assays and/or perfused organ studies are usually different from the physiological concentrations of these proteins in vivo, or simply, no proteins are added in these assays, which can be problematic because the intracellular drug concentration may differ between the experimental and in vivo conditions, particularly in the presence of protein-facilitated uptake mechanism(s) Accordingly, inaccurate IVIVEs of the intracellular drug concentration in physiologically-based pharmacokinetic and pharmacodynamic (PBPK/PD) modeling exercises could potentially induce unexpected toxicity and/or inefficacy in preclinical and clinical trials, which can be of concern in pharmacology studies. 1 Moreover, the most recent prediction models of drug disposition only considered the uptake of the unbound drug in cells, and, hence, the uptake effect of the protein drug complex was disregarded. 1,22 27 Poulin and coworkers, 1,17 20 however, have recently proposed diverse equations to quantify the uptake in cells of both the unbound and bound drug forms by combining an albuminfacilitated uptake mechanism and ph gradient effect. The main equation consisted of replacing the measured fu p value with an adjusted fu p value (fu p-adjusted ). This novel parameter considered additional processes that may potentially occur between the experimentally determined in vitro conditions used to estimate fu p and the real in vivo conditions, namely, (1) the ph gradient effect on the unbound drug fraction between plasma and the intracellular compartment, and (2) the albumin-facilitated uptake mechanism(s), which quantifies the uptake of the albuminbound drug fraction at the cell surface. This novel equation was based on the binding isotherm as presented below. 1,17 20 fu p-adjusted = PLR fu p funionized plasma funionized cells 1 + (PLR 1) fu p funionized plasma funionized cells where PLR is the plasma-to-tissue concentration ratio of extracellular binding proteins under in vivo conditions (e.g., albumin), whereas funionized plasma and funionized cell is the corresponding fraction of unionized drug in each matrix. The ionization parameters can be estimated from the Henderson Hasselbalch equations, the pka value of the drug and the physiological ph values on both sides of the membrane (e.g., about 7.0 for cells and 7.4 for plasma). Therefore, the above Eq. (1) converted the protein-bound drug concentration from the plasma to tissue (cell surface) assuming that the protein-bound drug fraction is also available for uptake in cells, and, that, in the case of albumin, this protein was found mainly outside the cells based on literature 18. Eq. (1) also considered that the drug at the cell surface is affected by the ph gradient. In other words, it was assumed that each drug molecule bound to an extracellular binding protein (e.g., albumin) in plasma, and, hence, in the interstitial space, may interact with the cell surface to deliver the additional bound drug into cells. Consequently, the intracellular drug concentration could be substantially higher than expected based solely on the free drug concentration, which is in accordance with the albumin-facilitated uptake mechanism(s). The net result will be that values of fu p-adjusted will be greater than fu p. Eq. (1) is sensitive in the low range of fu p values (i.e., for the highly bound drugs), and the change of fu p-adjusted with fu p is not proportional based on the binding isotherm, which might be of concern in PBPK/PD research as the plasma concentration used as a reference matrix could significantly deviate nonproportionally to the cell concentration. 1,18 To apply Eq. (1) to diverse experimental settings, the PLR value can simply be adjusted to reflect the real concentration ratio of albumin between the perfusate (or buffer) and organ materials. Similarly, the ph values can be adjusted to estimate the drug ionization parameters. Equation (1) was extensively and successfully validated with several drug examples for which their hepatic clearance was governed either by a metabolic, transporter, and/or permeation limitation effect ,23 By considering several drugs that bind mainly to albumin or AAG, the hepatic drug CL in vivo was more accurately predicted by correcting the microsomal and hepatocyte in vitro data with fu p-adjusted compared with fu p. In other words, the novel concept of fu p-adjusted demonstrated a significantly reduced prediction bias compared with the conventional fu p approach, particularly for those drugs highly bound to albumin. Furthermore, it has been demonstrated that the PLR effect (i.e., the correction for the albumin-bound drug effect) was predominant compared with the ph gradient effect according to Eq. (1) Overall, the hepatic in vivo clearance was more accurately predicted by incorporating the calculated value of fu p-adjusted in IVIVE procedures compared with the actual value of fu p measured in vitro. Also, the in vivo tissue plasma and tumor plasma partition coefficients at the whole organ level of several acidic and neutral drugs that bind mainly to albumin, were also accurately predicted by considering a ph gradient (1)

3 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism 3 effect on the unbound drug fraction as well as by considering a conversion of the albumin-bound drug fraction between plasma and tissues from the PLR value; therefore, the concept of fu p-adjusted has also been successfully applied to the prediction of drug tissue distribution in vivo In this context, however, there might still be some doubt on the validity of this novel parameter, especially for the prediction of drug uptake in cells. Therefore, after a second review of literature, the objective of the present study was to perform further validation exercises of the concept of fu p-adjusted by using additional case examples of IVIVEs related to the drug uptake and clearance in cells that covered diverse drug properties and experimental settings with varied albumin concentrations (e.g., perfused liver, isolated and suspended hepatocytes, and cultured cells overexpressing transporters) Eight more drugs were investigated in this study. Overall, a successful validation exercise made with additional case examples will enforce the use of Eq. (1) in PBPK/PD modeling research and the notion of the presence of a so-called albumin-mediated uptake mechanism influencing the IVIVE procedures 1. METHOD The impact of replacing fu p with fu p-adjusted in IVIVEs was again challenged in the estimations of the uptake rates with varied albumin concentrations. Although the cellular uptake of a number of endogenous and exogenous ligands seems to be driven by a so-called albumin-mediated uptake mechanism, the two following conditions should be prioritized according to the literature and published sensitivity analyses: (1) high protein binding in plasma (i.e., a low fu p value), and (2) an important binding to albumin. 1 21,23 Therefore, these conditions were further challenged in this study with different case examples. Case Examples, Datasets, and Assumptions Four robust case examples were investigated in this study, which covered eight drugs with different physicochemical and tissue disposition properties. Rat and human data were investigated. Furthermore, diverse types of cells (hepatocytes and MDCK) and different experimental settings (isolated and suspended hepatocytes, perfused liver, and cultured cells overexpressing transporters) were challenged, which should satisfy the criteria of an additional validation exercise. The datasets used to investigate these four case examples were obtained from the literature 3 14,19,24 34 and are presented in Tables 1 4; however, detailed explanations are also provided below. First, the current drugs were reported to bind mainly to albumin in plasma. 4 10,30 32 Second, as mentioned, it was assumed that each drug molecule (unbound and protein drug complex) may interact with the cell surface in the presence of albuminfacilitated uptake mechanism(s) to become available for uptake in cells. Accordingly, it was also assumed that such facilitated uptake mechanism(s) occurred on 100% of the cell surface in the studied experimental settings. 1 Moreover, it was hypothesized that the facilitated uptake mechanism(s) should also be valid for multiple uptake process predictions because it is applicable as long as the drug gets to the cell surface. The basis of these assumptions is based on literature information because it has been demonstrated that the proposed approach applied to drugs either metabolized, transported, and/or showing a limited permeability in cells Therefore, this present study further investigated these assumptions with additional case examples. Finally, human serum albumin or bovine serum albumin (essentially fatty acid free), or fresh human serum was used in the studied experimental settings. Consequently, it was also assumed that human serum albumin and bovine serum albumin provided the same effect. A concern that the changes in albumin concentration in the perfusate may affect the osmotic pressure of the perfusate (or incubation medium) has also been taken into account in the studied experimental settings; however, the osmotic/oncotic pressure (and drug concentration) was kept constant or was not found to be a cofounding factor. 4 8 Case Example 1 Tsao et al. 4 generated uptake data in vitro in isolated rat liver gently diced for rose Bengal. Hence, the intrinsic ability of hepatocytes to take up rose Bengal was estimated. The first case example consisted of predicting the change in the initial velocity (V 0 ) of uptake of rose Bengal by the isolated rat Table 1. Case Example 1: Rose Bengal Uptake in Isolated Rat Liver Slices Input Parameters Comparisons of Observed and Calculated Values of V 0 Albumin Concentration in Medium (g/dl) fu p PLR a fu a p-adjusted V 0 Observed (pmole/min mg Protein) V 0 Predicted Based on fu p (pmole/min mg Protein) b Fold Error of Deviation c V 0 Predicted Based on fu p-adjusted (pmole/min mg Protein) b Fold Error of Deviation d ± ± ± ± ± a The PLR value was used to estimate fu p-adjusted according to Eq. (1). At the normal physiological albumin concentration in rats (i.e., about 3.8 g/dl), the value of PLR (serum-to-liver concentration ratio of albumin) in rats ranges between 11.6 and ,18,24 ; therefore, the average value of 12.5 was used for the physiological concentration. The PLR value for the varied albumin concentrations in the incubation medium was adjusted by linear interpolation. A standard ph value of 7.4 for plasma was used in the calculation of fu p-adjusted, but for the intracellular fluid the literature indicates a ph value of 7.1 for the hepatocytes that are in contact with buffer or plasma in incubation medium. 34 The acidic pka values considered for rose Bengal are 1.9 and b As explained in Method. c Fold errors of deviation between the values of V 0 observed and V 0 calculated based on fu p. d Fold errors of deviation between the values of V 0 observed and V 0 calculated based on fu p-adjusted.

4 4 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism Table 2. Case Example 2: Warfarin and Taurocholate Uptake by Rat Perfused Liver Input Parameters Comparison of Observed and Predicted Variations in the Measured CL int Values Drugs Albumin Concentration in Perfusate (g/dl) fu p PLR a fu p-adjusted CL int of the Perfusate (ml/min g Liver) Observed Variations in CL int b Predicted Variations in CL int (Variations in fu p ) b Fold Error of Deviation c Predicted Variations in CL int (Variations in fu p-adjusted /fu p ) b Fold Error of Deviation d Warfarin Taurocholate a The PLR value was used to estimate fu p-adjusted according to Eq. (1). At the normal physiological albumin concentration in rats (i.e., about 3.8 g/dl), the value of PLR (serum-to-liver concentration ratio of albumin) in rats ranges between 11.6 and ,18,24 ; therefore, the average value of 12.5 was used for the physiological concentration. The PLR value for the varied albumin concentrations in the perfusate was adjusted by linear interpolation. A standard ph value of 7.4 for plasma was used in the calculation of fu p-adjusted, but for the intracellular fluid the literature indicates a ph value of 7.1 for the hepatocytes that are in contact with buffer or plasma in incubation medium. 34 The acidic pka value considered for Warfarin and Taurocholate is 5.1 and 1.9, respectively. 17,29 b As explained in Method. Variations between each albumin concentration and serum-free perfusate (i.e., 0 g/dl). For Taurocholate, the serum-free perfusate is replaced by the perfusate having the lowest albumin concentration (i.e., 0.5 g/dl). c Fold errors of deviation between the variations in CL int and variations in fu p. d Fold errors of deviation between the variations in CL int and variations in fu p-adjusted /fu p. Table 3. Case Example 3: Mibefradil, Ozazepam, Midazolam, and Bosentan in IVIVE Studies Made from Human Hepatocytes Suspensions Input Parameters Comparison of Observed and Predicted CL int Values in 100% Human Serum Drugs CL int In Vivo in 100% Human Serum (ml/min kg) Unbound CL int In Vitro in Serum-Free fu p in 100% Medium Human (ml/min kg) Serum PLR for 100% Plasma a fu p-adjusted fu inc in Serum-Free Medium Calculated CL int Based on fu p /fu b inc Fold Error of (ml/min kg) Deviation c Calculated CL int Based on fu p-adjusted /fu b inc Fold Error of (ml/min kg) Deviation d Mibefradil Oxazepam Midazolam Bosentan a The PLR value was used to estimate fu p-adjusted according to Eq. (1). At the normal physiological albumin concentration in 100% human serum (i.e., about 4.5 g/dl), the value of PLR (serum-to-liver concentration ratio of albumin) in humans ranges between 11.6 and ,18,24 ; therefore, the average value of 12.5 was used for 100% human plasma. A standard ph value of 7.4 for plasma was used in the calculation of fu p-adjusted, but for the intracellular fluid the literature indicates a ph value of 7.1 for the hepatocytes that are in contact with buffer or plasma in incubation medium. 34 Oxazepam and Midazolam are mostly nonionized at the physiological ph, whereas an acidic pka value of 5.5 and a basic pka value of 9.5 were used for Bosentan and Mibefradil, respectively. b Calculated either as unbound CL int x fu p /fu inc or unbound CL int x fu p-adjusted /fu inc. c Fold errors of deviation between the CL int determined in 100% human serum (first column) and unbound CL int scaled based on fu p /fu inc. d Fold errors of deviation between the CL int determined in 100% human serum (first column) and unbound CL int scaled based on fu p-adjusted /fu inc. liver cells with varied albumin concentrations in the incubation medium (ranging from 0 to 14.5 g/dl). Human serum albumin (essentially fatty acid free) was used. V 0 was measured at a rose Bengal concentration (C t ) of 11.5 :M and at each albumin concentration. The value measured in the albuminfree medium was compared with the value of V0 calculated with the following equation for each albumin concentration; V 0 = [(V max fu p C t )/(K m +fu p C t ) + D app fu p C t ], where C t is the drug concentration added in the buffer (11.5 :M), and the parameters for the uptake of unbound rose Bengal were determined separately such as the Michaelis Menten affinity constant (K m = 10.1 :M), maximal velocity constant (V max = 216 pmol/(min mg protein)), and transport maximum rate (D app = 1.15 :L/(min mg protein)). In the higher range of albumin concentrations, the measured values of V 0 are much greater than the values calculated based on fu p of the incubation mediums, which violates the assumption that only the unbound drug concentration is available for uptake. Therefore, the albuminbound drug fraction was probably also available for uptake. To test this hypothesis, V 0 was also calculated by replacing the value of fu p with fu p-adjusted calculated with Eq. (1). A standard ph value of 7.4 for plasma was used in the calculation of fu p-adjusted, but for the intracellular fluid, the literature indicates a ph value of about 7.1 for the hepatocytes that are in contact

5 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism 5 Table 4. Case Example 4: Rifampicin in Inhibition Interaction Studies that Were Made in Cultured MDCK Cells Overexpressing OATP1B1 Transporter Incubated in 100% Human Plasma or a Plasma-Free Medium Input Parameters Comparison of Observed and Predicted Variations in the Measured IC50 Values Drugs IC50 in 100% Human Serum IC50 in Serum-Free Medium fu p Measured in 100% Human Serum Estimated fu p-adjusted in 100% Human Serum a fu inc in Serum-Free Medium b Observed Variations in IC50 c Predicted Predicted Variations Variations in IC50 in IC50 (Variation in (Variation Fold Error of fu p-adjusted / Fold Error of in fu p /fu inc ) d Deviation e fu inc ) d Deviation e Rifampicin inhibiting E17$G a As explained in Method. It was assumed that both the bound and unbound drug fractions can directly interact with the transporter at the cell surface in these specific experimental settings. In other words, the real fu p at the cell surface, and hence, fu p-adjusted should equal to unity. b As explained in Method. Assumed equals to 1 based on Rifampicin lipophilicity and ionization. c Variation in IC50s between the experimental setting using 100% human serum and a serum-free medium. d Variation between fu p and fu inc or between fu p-adjusted and fu inc. e Fold errors of deviation between the observed and predicted variations in IC50s. with buffer or plasma in incubation mediums. 34 The dataset is presented in Table 1. Case Example 2 The second case example consisted of predicting the kinetics analysis of removal of Warfarin and Taurocholate by perfused rat liver with varied albumin concentrations in the perfusate (i.e., ranging from 0 to 4.6 g/dl for Warfarin and from 0.5 to 5 g/dl for Taurocholate). 6 8 Bovine serum albumin (essentially fatty acid free) was used. The CL int (ml/s g liver) was determined at each albumin concentration and in the albumin-free perfusate. The concentration of the drug added in the perfusate was kept constant. This study demonstrated that the measured values of total CL int decreased, whereas the albumin concentration increased in the perfusate (and, hence, fu p decreased). Therefore, although only the unbound drug concentration is available for uptake in cells, the change in total CL int between two albumin concentrations should equal the change in the corresponding fu p values. However, this condition was not respected particularly in the higher range of albumin concentration (i.e., the change in fu p values between two perfusates of different albumin concentrations is much larger that the corresponding change in CL int values). Hence, assuming that the albumin-bound drug complex is also available for uptake in cells, higher CL int values are expected than based on fu p values. Consequently, it was assumed that in the higher range of albumin concentrations, the albumin-bound form availability is the main factor governing CL int of these highly bound drugs compared with the unbound drug. To verify this assumption, the contribution of the albumin-bound drug was highlighted compared with the unbound drug fraction. Accordingly, the measured fu p value of each perfusate was corrected to also get a measure of the albumin-bound drug level at the cell surface and the ph gradient effect in vivo, which are not taken into account by the actual measured fu p values in vitro compared with the calculated fu p-adjusted. The correction was made simply by calculating the ratio of fu p-adjusted /fu p ; thus, the later parameter principally quantifies the contribution of the albumin-bound drug and ph gradient effect compared with only the unbound drug effect. Overall, the change in total CL int values with varied albumin concentrations in the perfusate was then compared with the corresponding change in the actual fu p values (a measure of the unbound drug) and calculated ratios of fu p-adjusted /fu p (a measure of the effect of the protein-bound drug faction and ph gradient compared with the unbound drug). Again, fu p-adjusted was calculated with Eq. (1). The dataset is presented in Table 2. Case Example 3 Blanchard et al. 9 generated data on CL int in cryopreserved human hepatocyte suspensions that were incubated in 100% human serum or serum-free incubation mediums for some highly bound drugs, namely, Mibefradil, Midazolam, Oxazepam, and Bosentan Therefore, the third case example consisted of scaling the CL int determined in serum-free medium to get an estimate of the true CL int in 100% human serum. Physiologically based direct scaling was used to scale the in vitro CL int to the in vivo value in ml/(min kg). 9 The CL int values that were determined in 100% human serum were very close to the real in vivo values, and, hence, the in vivo conditions were considered. The unbound CL int determined in vitro in the serum-free medium was then scaled to the in vivo conditions by using two correction factors: (1) CL int fu p /fu inc, or (2) CL int fu p-adjusted /fu inc, where fu inc is the unbound fraction that has been determined in the plasma-free incubation medium. 19 The scaled CL int values were compared with the CL int values determined in 100% human serum (when incubations were performed in 100% serum, it was assumed that fu p /fu inc equals to the unity). Note that the original CL int values for the plasma-free and 100% plasma mediums were back calculated from the corresponding hepatic clearance values obtained with the well-stirred model as presented for each experimental condition by Blanchard et al. 9 The dataset is presented in Table 3. Case Example 4 The last case example was used to challenge the albuminfacilitated uptake mechanism(s) in cells other than hepatocytes (i.e., cultured MDCK cells). Recently Jahic et al. 14 determined the IC50 values in DDI studies for Rifampicin (inhibiting E17ßG) and Dolutegravir (inhibiting metformin). These

6 6 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism studies were performed in cultured MDCK cells overexpressing the transporter OAT1B1 (for Rifampicin) or OCT2 (for Dolutegravir). Again, assuming that only the unbound drug concentration is available for inhibition at the cell surface, the change in IC50s between the cells incubated in 100% human serum and in a serum-free medium should equal the corresponding change in the corresponding unbound fraction values. However, this is not the case, that is, for these two drugs, the variation in IC50 values between the cell system using 100% serum and the serum-free medium is much lower than the corresponding variation in the unbound fractions (Table 4). In other words, these data suggest that for these two highly protein bound drugs, the real free drug concentrations in the presence of transporters and other drug-binding proteins could be substantially higher than the calculated free drug concentration in human serum based on the unbound fraction values. Moreover, for Rifampicin, the measured IC50 values are identical in the two experimental settings (i.e., 1.61 for 100% human serum and 1.51 for the serum-free medium), as the two corresponding curves for the transporter activity as a function of drug concentration are mostly superimposed. 14 Therefore, the later observation suggests that both the unbound and bound drug fractions seem to interact with the transporter cell surface for more inhibition than expected based on the unbound fraction. In these specific experimental settings where 100% human serum surrounded the cells in the vials (i.e., due to a loss of organ integrity in cultured cells), it was assumed that both the unbound and albumin-bound drug fractions directly interacted with the transporters at the cell surface in the presence of a so-called albumin-mediated uptake mechanism by contrast to the serum-free medium. Hence, the real value of fu p at the cell surface in the vials using 100% human serum should simply equal the unity. In this case, fu p-adjusted will become equal to unity. Overall, it was assumed that the change in IC50 values between cells incubated in 100% human serum and a serumfree medium should rather equal the corresponding change in the corrected unbound fraction values [i.e., fu p (and fu p-adjusted ) assumed equal to unity in 100% human serum divided by fu inc for the unbound fraction in the serum-free medium]. The fu inc represents a loss of substrate in the serum-free vials, and, consequently, it needs also to be taken into account. For Rifampicin, however, the value of fu inc (a measure of nonspecific drug binding to cells and plastic material in the serum-free experimental setting) was set equal to unity as explained below. Rifampicin is highly ionized in the incubation medium at ph 7.4 (acid pka = 1.7 and basic pka = 7.9), and, hence, its log D (n-octanol:buffer partition coefficient corrected for drug ionization) is low at Therefore, it was assumed that fu inc is equal to unity in the absence of an experimental value. However, for the neutral drug Dolutegravir, which is tightly bound to plasma proteins (fu p = 0.005) compared with Rifampicin (fu p = 0.1), it has been reported in the literature that the low solubility of Dolutegravir in buffered solutions and its nonspecific binding to materials present significant challenges in the absence of serum proteins. 36,37 Accordingly, the log P (n-octanol:buffer partition coefficient) is 2.2 for this drug suggesting a relatively high lipophilicity. 38 Consequently, the value of fu inc cannot be roughly estimated (even though fu inc should be very low based on the extent of its extremely high plasma protein binding). The problem with this drug is not to calculate fu inc by considering the binding to the biological material, although considering the binding to the plastic material cannot be predicted without experimental data. Therefore, Dolutegravir was not further investigated. The dataset is presented in Table 4. Relative Impact of Each Contributing Factor of fu p-adjusted There are two changes relatively to the traditional fu p theory: the ph gradient effect and the PLR effect. This exercise was made to verify which of the ph gradient effect or PLR effect was the most predominant effect in the calculations of fu p-adjusted, and, thus, of the uptake rate. Therefore, the calculations of fu p-adjusted from Eq. (1) were also made either by considering separately or together the effect of ph and PLR. To verify the effect of ph only, the PLR value was set equal to unity, whereas to verify the effect of PLR only, the ph of cells and plasma was set equal to the same value. The calculated fu p-adjusted values were compared with the measured fu p values. Note that, however, such an exercise has already been made in previous calculations of hepatic drug clearance, and the results indicated that the PLR effect (and hence the adjustment made to consider the albumin-bound drug effect) was the predominant effect compared with ph to get an accurate estimate of the in vivo clearance data. This aspect was further verified in this present study. Evaluation of Predictive Performance The statistical analyses performed in the present study were the same as already described in the literature. 39 Briefly, the predicted value (either from fu p or fu p-adjusted ) was compared with the observed value to determine the predictability of each method using standard statistical techniques for accuracy, precision, and correlation. The absolute average fold error and root mean-squared error were calculated and are presented. Furthermore, the concordance coefficient of correlation is presented, which evaluates the global degree to which pairs of predicted and observed data fall on the line of unity passing through the origin. Specific fold errors of deviation between the predicted and observed values (percentage of fold error 2, 3, and 10) were also calculated. Finally, the plot of observed versus predicted values was also made. RESULTS A total of four more case examples were evaluated in the present study for eight drugs. The predicted and observed values are presented in Tables 1 4, whereas the overall comparative assessment is listed in Table 5. The plot of observed versus predicted values is shown in Figure 1. Furthermore, Table 6 shows the comparison between the measured fu p values and the values of fu p-adjusted calculated by considering the effect of ph and PLR separately or together. Case Examples 1 4 On the basis of all statistical parameters, the proposed IVIVE method based on fu p-adjusted was the best-performing prediction method of the uptake rate (or clearance) as a function of protein binding compared with the conventional method based on the fu p theory. This is particularly true for the highly bound drugs in the higher range of albumin concentrations ( 0.72 g/dl) (Tables 1 5). This is reflected by the fold errors of deviation between the observed and predicted values, which are closer to unity for the IVIVE methods based upon fu p-adjusted compared with fu p, and this can be viewed graphically (Fig. 1).

7 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism 7 Table 5. Comparison of Predicted and Observed Values Obtained from Tables 1 to 4 Scenarios n %, Twofold %, Threefold %, 10-Fold AAFE RMSE CCC Predictions based on fu p-adjusted Predictions based on fu p n, number of predicted values per approach; AAFE, absolute average fold error; RMSE, root-mean-square error; CCC, concordance correlation coefficient. Therefore, the results show two levels of accuracy: (1) in the relatively higher range of albumin concentrations ( 0.72 g/dl) where the use of fu p-adjusted compared with fu p was a significant benefit for the highly bound drugs, and inversely, (ii) in the lower range of albumin concentrations (<0.72 g/dl) where the use of fu p-adjusted represented no benefit, particularly for those drugs less bound to albumin. The later observations are in accordance with the results obtained for Taurocholate, as this drug is relatively moderately bound compared with the other drugs (its fu p value varied only from 0.11 to 0.57 in the range of albumin concentrations tested), and, for this reason, the use of fu p-adjusted slightly improved the prediction compared with fu p (Table 2). Finally, the variation in the IC50 values as a function of protein binding was accurately captured for Rifampicin with the proposed approach (Table 4). Relative Impact of Each Contributing Factor of fu p-adjusted Figure 1. Comparison between the predicted and observed values. The x-axis represents the values predicted either based on a correction from fu p-adjusted or fu p for the case examples 1 4, whereas the y-axis represents the corresponding observed (measured) values obtained from the literature. The higher range of albumin concentrations was considered (i.e., ranging from 0.72 to 14.5 g/dl). The solid line indicates the best fit (unity). Dashed lines either side of the unity include a factor of two and three, respectively. Red circles and blue squares indicate the values that were predicted from fu p-adjusted (R 2 = 0.95) and fu p (R 2 = 0.18), respectively (n = 13). Raw data were obtained in Tables 1 4. Table 6 confirms that the PLR effect is the predominant process governing fu p-adjusted compared with the ph gradient effect. This is because by considering a ph gradient effect only, the calculated values of fu p-adjusted were inferior or similar to the fu p values measured in vitro by contrast to when the PLR effect is considered. In other words, the ph effect was relatively minor compared with the PLR effect for the current drugs; therefore, the PLR effect explains why the calculated values of fu p-adjusted were significantly superior to the measured fu p values, which provided more accurate predictions of drug uptake in cells. Again, the results of this exercise additionally support the assumption that albumin-bound drug concentration is a major contributor to intracellular concentration. DISCUSSION A novel parameter (fu p-adjusted ) has been proposed to improve the IVIVE procedures, and, again, this parameter gave the most robust predictions of the uptake rate (clearance) as a function of protein binding for additional drug datasets. Therefore, Table 6. Calculations of fu p-adjusted According to Eq. (1) Either by Considering the Effect of ph Gradient and PLR Separately or Together, and Comparison with the Measured fu p Values Calculated fu p-adjusted a Drugs Albumin Concentration (g/dl) Measured fu p ph Effect Only PLR Effect Only PLR and PH Effects Rose Bengal Warfarin Taurocholate Mibefradil Oxazepam Midazolam Bosentan a From Eq. (1). To verify the effect of ph only, the PLR value was set equal to unity, whereas to verify the effect of PLR only, the ph of cells and plasma was set equal to the same value.

8 8 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism this present study confirms the utility of fu p-adjusted compared with the fu p theory in the IVIVEs of drugs highly bound to albumin. The outcome is then similar to previous studies of Poulin and coworkers, 1,17 20 and, hence, the mechanistic basis of fu p-adjusted that has been proposed by these authors is well supported. The current results confirm that the unbound drug concentration is not high enough to explain the uptake rates observed in the current experimental settings; thus, the cell surface appears to interact also with the protein drug complexes. In other words, this supports the notion that considering additional mechanisms such as the albumin-facilitated uptake mechanism(s) and ph gradient effect yields greater prediction successes. Thus, the experimental data that are generated in cell-based assays and/or perfused organ systems without albumin or with varied albumin concentrations will need corrections to reflect more accurately the in vivo conditions. Four case examples were presented in this present study where the experimental data have been successfully corrected with fu p-adjusted. In a companion manuscript, the novel theory proposed in this study was further confirmed for Bisphenol A with newly obtained experimental data. Moreover, the PLR effect (i.e., the contribution of the albumin-bound drug) was predominant compared with the ph gradient effect according to Eq. (1) for the current drugs, which is in line with previous sensitivity analyses recently published in the literature Overall, the proposed simple equation that replaced fu p with fu p-adjusted in the IVIVE procedures could already be used in PBPK/PD modeling research in drug discovery and development, as it is only based on readily available input parameters (i.e., fu p and pka). The results showed that the albumin-facilitated uptake mechanism(s) (i.e., the use of fu p-adjusted ) becomes important in the higher range of albumin concentrations for the highly bound drugs, which is also in total accordance with the literature. 6 8 Therefore, correcting the experimental data with fu p-adjusted compared with fu p will be relevant in the range of physiological albumin concentrations for a drug highly bound to albumin. Accordingly, the results confirm the ability of the proposed Eq. (1) to cover a large range of albumin concentrations in the calculation of the input PLR value, and, hence, of fu p-adjusted (Tables 1 and 2). This will be important in the case of hyper- or hypoalbuminemia. Remember that in Eq. (1), the original PLR value represents the ratio of albumin concentration between plasma and the organ under in vivo conditions, 24,25 and, hence, PLR was adjusted for the diverse albumin concentration observed in the plasma (or perfusate) of the studied experimental settings (Tables 1 4). The difference of albumin concentration between plasma (or perfusate) and the organ material was related to the difference in the albumin-bound drug fraction available for uptake. Accordingly, Tsao et al. 6,7 have also investigated this aspect for Warfarin by using an IPRL system. In the lowest range of albumin concentrations (e.g., <0.7 g/dl), the level of albumin drug complex is low and this has probably no apparent effect on the cellular uptake of drugs, which is also confirmed in the present study. However, in the physiological albumin concentration range, the uptake can mostly be explained by the bound drug concentration. Conversely, in the very high range of albumin concentrations, the albumin-bound drug is almost constant, and the increase of albumin concentration may inhibit the uptake of albumin-bound drug because of the competition of free albumin with the albumin drug complex for the binding site on the cell surface. The later observations were based on a hypothetical mathematical model developed by Tsao et al. 6,7 ; however, in the present study, data determined with albumin concentrations ranging from 0.72 to 14.5 g/dl indicated that the impact of the albumin-bound drug complex was predominant (Tables 1 4). Conversely, Cui and Walter, 40 however, indicated that the uptake of the albumin-bound organic anion sulfobromophthalein complexes by the human hepatocyte transporters occurs only from the pool of unbound ligand. However, the albumin concentrations used in their study were very low compared with the physiological albumin concentration (lower by 100-fold), which can be a rational explanation. In addition, it seems that when the uptake rate (clearance) of a drug is low compared with the plasma flow in liver, the uptake can be considered to represent mainly the uptake ability of the liver, and, hence, of the albumin-facilitated uptake mechanism(s). Thus, the compounds showing a low uptake rate and/or CL int were more accurately covered from fu p-adjusted compared with fu p. 5,17 20 Despite many attempts to identify a mechanism(s) for the mediated uptake, it remains largely unknown. Recently, Poulin et al. 1 reviewed the proposed mechanisms. As said in the Introduction section, at present, the most plausible mechanism(s) seems to be related to the presence of ionic attractions between the protein drug complex and cell surface to result in a marked reduction in the diffusional distance of the extracellular protein drug complex. Hence, there is an excellent correlation between the uptake rates and extracellular protein isoelectric points (pi) when the unbound ligand concentration is constant to keep events in the unstirred water layer constant. Accordingly, Burczynski et al. 21 reported that palmitate clearance in hepatocytes was greatest when lysozyme albumin (pi = 11.0) was used as the extracellular binding protein and lowest in the presence of AAG (pi = 2.7). The same trend was observed when varying the pi value of albumins. The present study and the literature indicate that the extracellular binding proteins with a net positive charge (e.g., albumin and globulins) showed an increased hepatic elimination of drugs compared with the proteins with a net negative charge (e.g., AAG) because of a reduced uptake by the hepatocytes. 1,4,6,7,17 20 It is evident that the uptake rates of drugs are much higher than those predicted based on the unbound fractions in the presence of either albumin or globulins, 4 and this is demonstrated further in this present study. However, the impact of other extracellular binding proteins that might be implicated in a protein-facilitated uptake mechanism such as intracellular fatty acid-binding protein and lipoproteins has not yet been investigated. Nevertheless, although another protein is suspected to be implicated (e.g., Globulins), 4,7 the value of fu p-adjusted can be calculated the same way we have carried this out for albumin. At present, it becomes imperative to investigate the binding of diverse protein drug complexes to the surface of different types of cells to better understand the role of the extracellular binding proteins in the cellular uptake of small molecules. Association between albumin ligand complexes (and ligand-free albumin) and hepatocyte cell surface has already been studied but only for rose Bengal. 5 It has been suggested on this basis that the binding of albumin to the cell surface is functionally equivalent to the process that makes bound ligand available for hepatic uptake. In this context, it seems that the types of cells might also be a critical aspect in the albumin-mediated uptake; nevertheless, the current literature and this present study suggests the existence of a so-called albumin-mediated uptake

9 RESEARCH ARTICLE Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism 9 mechanism at least in hepatocytes and cardiac myocytes as well as in MDCK cells overexpressing OATP transporters Other studies reported the existence of albumin-mediated uptake mechanism(s) also in perfused kidney studies, but the drugs used in these experimental settings are not highly bound to albumin (i.e., fu p > 0.30), 12,13 which may explain the relatively lower impact of albumin compared with other studies performed with hepatocytes for much more highly bound drugs. 1 13,21 Therefore, such mechanism(s) and cells from other organs as well as other transporters deserve further exploration. Furthermore, human serum albumin or bovine serum albumin has been used in the in vitro cell assays and perfused liver studies, but in light of the published experimental studies, 3 13 and of the present study, the albumin-facilitated uptake mechanism(s) have been observed with both sources of albumin. Nevertheless, we agree that the human and bovine albumin may differ to some extent, and, hence, this may relatively influence the past and current comparative assessments. The presence of potential interactions between the protein drug complexes and cell surfaces may also explain the results observed in the DDI studies of Rifampicin (inhibiting E17ßG) 14 (i.e., the case example 4 in Table 4). The inhibition studies were performed in suspensions of MDCK cells overexpressing the OATP1B1 transporter; thus, it might be that the protein drug complex interacted with the transportes present at the cell surface in these experimental settings. Rifampicin inhibition effect could be predicted in this present study; therefore, it seems that for this drug both the bound and unbound drug fractions interacted with the OATP1B1 transporter at the cell surface as hypothesized, and, this effect has been accurately quantified. One thing is sure is that, for Rifampicin, the inhibition potency (e.g., IC50) is almost identical in the two experimental settings (i.e., using 100% human serum and serum-free medium), as the two corresponding curves for the transporter activity as a function of drug concentration are mostly superimposed. 14 Therefore, this observation supports the presence of albumin-facilitated uptake mechanism(s) at the transporter cell surface in the inhibition studies. Accordingly, the literature indicates that the family of OATP transporters interact mostly with drugs bound to albumin, which is the case of Rifampicin. 31,41 43 Overall, the results for Rifampicin support the notion that both the unbound and bound drug fractions also governed the IC50s. However, we were not able to investigate the results of the DDI study made with Dolutegravir inhibiting Metformin, as the essential experimental value of fu inc for the serum-free medium was not available for this specific experimental setting, and also fu inc cannot easily be predicted. In this case, it was difficult to make a rough estimate of fu inc for Dolutegravir because it is reported in the literature that this drug has a low solubility in buffered solutions, and, hence, a relevant nonspecific binding to plastic materials represents a significant challenge in the absence of serum proteins. 36,37 Another DDI study is in accordance with the results obtained for Rifampicin. In this case, the time-dependent inhibition of the cytochrome P450 (isoform 3A4) was investigated in two different experimental settings by other authors (i.e., human liver microsomes incubated in a plasma-free medium, and human hepatocytes suspended in 100% human plasma). 16 The inactivation constant (K i ) (for the Crizotinib Midazolam interaction) differed between the two experimental conditions (3.08 : M in plasma-free medium, and 9.57 : M in 100% human plasma). The same is true for the unbound fraction of the inhibitor (fu inc = 0.12 in plasma-free medium, and fu p = in 100% human plasma). However, the K i values varied by 3.1-fold between the settings using 100% plasma and a plasma-free medium, whereas the corresponding fu p and fu inc values varied only by 0.78-fold. Therefore, the change in K i values not followed the change in the unbound drug concentration, which again violates the fu p theory. Interestingly, the change in K i values can be relatively well predicted by using the same approach as proposed for Rifampicin (i.e., simply by fixing fu p or fu p-adjusted sets equal to unity as stated in the method); therefore, the estimated change in the unbound fractions will be about 8.3 (i.e., 1/0.12), which is in closer agreement with the change in Ki values. This could be another case example where adjusting the unbound fraction of the vials containing 100% plasma to consider additional albumin-facilitated uptake mechanism(s) represents an advantage; however, the authors agree that two different experimental settings were compared in this DDI study (plasmafree human liver microsomes and 100% human hepatocyte suspensions), and, hence, the use of only hepatocyte data should also be investigated. In other words, fu inc needs to be determined also in the plasma-free hepatocyte system compared to microsomes. CONCLUSION This present study confirms the validity of using fu p-adjusted compared with fu p in IVIVE methods, which again supports the notion that considering additional mechanisms such as the albumin-facilitated uptake mechanism(s) and ph gradient effect yield greater successes in the prediction of uptake rate as a function of protein binding for small molecules. These observations have been confirmed by using additional data obtained in different experimental settings for eight drugs, and the main conclusion of this study is in total accordance with the previous literature. 1 24,44 The protein-facilitated uptake mechanism(s) could be another paradigm shift in PBPK/PD modeling research in addition to a previous paradigm shift related to the ph gradient effect. 1,44 The consideration of an albumin-facilitated uptake mechanism was important for drugs highly bound to albumin but only in the higher range of albumin concentrations (e.g., the physiological albumin concentration range) as observed in this study and the literature One of our previous related manuscripts presents a sensitivity analyse that indicated that for those drugs highly bound to albumin (i.e., a measured fu p value lower than 0.01) and having a low plasma CL in vivo (i.e., lower than 5% of the liver blood flow rate) the impact of using fu p-adjusted is more pronounced 18, which is in accordance with the present study. It is encouraged to continue the research in that domain as the presence of significant extracellular protein-mediated uptake mechanism(s) could change the way we are estimating the intracellular drug concentration in the IVIVE procedures, which is related to efficacy and toxicity. Accordingly, intracellular drug concentration could potentially be predicted in PBPK/PD modeling studies, and, hence, in toxicity and/or efficacy studies, as recently suggested by Poulin et al. 1 Overall, for drugs highly bound to plasma proteins (e.g., albumin), it is suggested to estimate the uptake kinetic parameters with cell-based assays incubated in 100% human serum or to make a correction while the experimental data are generated either without albumin or with varied albumin concentration,