Abstract
In vitro intrinsic metabolic clearance (CLint) is used routinely for compound selection in drug discovery; however, in vitro CLint often underpredicts in vivo clearance (CL). Forty-one proprietary compounds and 16 marketed drugs were selected to determine whether permeability and efflux status could influence the predictability of CL from in vitro CLint obtained from liver microsomal and hepatocyte incubations. For many of the proprietary compounds examined, rat CL was significantly underpredicted using the well stirred model incorporating both fraction of unbound drug in blood and fraction of unbound drug in the microsomal or hepatocyte incubation. Further analysis revealed that the accuracy of the prediction was differentiated by permeability and P-glycoprotein- (P-gp) and mouse breast cancer resistance protein (mBcrp)-mediated efflux. For proprietary compounds with passive permeability greater than 5 × 10−6 cm/s and efflux ratios less than 5 in both P-gp- and mBcrp-expressing cells, CLint provided reasonable prediction. The average -fold error (AFE) was 1.8 for rat liver microsomes (RLMs) and 2.3 for rat hepatocytes. In contrast, CL was dramatically underpredicted for compounds with passive permeability less than 5 × 10−6 cm/s; AFEs of 54.4 and 29.2 were observed for RLM and rat hepatocytes, respectively. In vivo CL was also underpredicted for compounds that were good efflux substrates (permeability >5 × 10−6 cm/s). The AFEs were 7.4 and 8.1 for RLM and rat hepatocytes, respectively. A similar relationship between permeability, efflux status, and human CL prediction reported in the literature was observed for 16 marketed drugs. These data show that permeability and efflux status are determinants for the predictability of CL from in vitro metabolic CLint.
In vitro metabolic clearance (CLint) derived from liver microsomal incubations is used routinely to aid compound selection in drug discovery. Hepatocytes, which possess intact cellular membranes, complete enzymatic systems, and cofactors, have also become a common option for determination of CLint. Successful prediction of in vivo clearance (CL) from CLint in liver microsomes or hepatocytes has been reported for some compounds. Obach (1999) reported that CL was predicted within a 2-fold error for 16 of 29 compounds from CLint in human liver microsomes using the well stirred model incorporating both fraction of unbound drug in blood (fub) and fraction of unbound drug in the microsomal or hepatocyte incubation (fuinc). Likewise, CLint from human hepatocytes provided prediction within a 2-fold error for 11 of 37 (Brown et al., 2007) or 23 of 56 compounds (Riley et al., 2005). However, a tendency toward underprediction was observed for many drugs (Obach, 1999; Ito and Houston, 2004; Riley et al., 2005; Brown et al., 2007; Stringer et al., 2008). Marked underprediction of CL from in vitro metabolic CLint in humans and/or animals has been reported for compounds such as montelukast, troglitazone, and quinotolast (Naritomi et al., 2001, 2003; Riley et al., 2005).
The reasons for a poor prediction of in vivo CL from in vitro CLint are often not explored in depth. Few reports have systematically examined the reasons for the discrepancy between predicted and observed CL, aside from the possible contribution of renal CL. Factors such as extrahepatic metabolism (De Kanter et al., 2004), variations in experimental design (Tang et al., 2008), differences between in vitro and in vivo metabolic enzyme activities, or possible inaccuracy in measurement of plasma unbound fraction (fup) for highly bound compounds could potentially contribute to the poor in vitro and in vivo correlation. It is increasingly recognized that drug transporters such as uptake transporters of the organic anion transport polypeptide family and the efflux transporters P-glycoprotein (P-gp) and breast cancer resistance protein (Bcrp) may play a significant role in the elimination of many drugs (Shitara et al., 2006). Transporters may even affect the CL of compounds that are eliminated primarily as metabolites, such as repaglinide (Bidstrup et al., 2003; Niemi et al., 2005) and cerivastatin (Shitara et al., 2003).
During the lead optimization process for one of our internal projects, a lack of correlation was observed between CLint obtained from rat liver microsomes (RLMs) and in vivo CL observed in rats. Several additional experiments were conducted to gain a better understanding of the reasons for the poor in vitro and in vivo correlation. Initial results suggested that the correlation was not improved when fresh rat hepatocytes were used to determine CLint. Therefore, the quantitative prediction of rat CL from in vitro metabolic CLint in RLM or rat hepatocyte suspensions was investigated in depth using a set of 41 proprietary compounds. Compound selection was based on in vivo CL, CLint in RLM, and structural diversity within the chemical series. For the majority of the compounds examined, rat CL was significantly underpredicted from CLint obtained from either RLM or rat hepatocytes using the well stirred model incorporating fub and fuinc (Ito and Houston, 2004). Preliminary data indicated that several compounds from the project were substrates for P-gp and Bcrp in vitro; several also displayed low permeability. We hypothesized that transporters might contribute to the lack of correlation between in vitro CLint and in vivo CL in the rat. We report here the examination of the relationship between permeability, efflux, and the predictability of CL. The impact of permeability and efflux on the predictability of human CL was also evaluated using 16 marketed drugs.
Materials and Methods
Materials.
Liver microsomes (RLMs) isolated from male Sprague-Dawley rats were purchased from BD Biosciences (San Jose, CA). Plasma from male Sprague-Dawley rats was purchased from Bioreclamation, Inc. (Westbury, NY). Collagenase was purchased from Sigma-Aldrich (St. Louis, MO). Phenol red-free Dulbecco's modified Eagle's medium (DMEM), DMEM-GlutaMax, l-glutamine, Medium 199, 0.05% and 0.25% trypsin-1 mM EDTA, Hanks' balanced salt solution (HBSS), and heat-inactivated fetal bovine serum (FBS) were purchased from Invitrogen (Carlsbad, CA). Noncollagen-coated 24-well Transwell plates (0.4 μm pore size, 0.7 cm2 surface area) were purchased from Millipore Corporation (Billerica, MA). Forty-one proprietary compounds were synthesized by the Department of Medicinal Chemistry at Amgen (Cambridge, MA). Ritonavir was purchased from Moravek Biochemicals (Brea, CA); bosentan and montelukast were purchased from Bosche Scientific LL (New Brunswick, NJ); and troglitazone was purchased from Cayman Chemical (Ann Arbor, MI). All of the other chemicals were purchased from Sigma-Aldrich.
The parental porcine renal epithelial cells and Madin-Darby canine kidney cells were purchased from American Type Culture Collection (Manassas, VA). Transfection of porcine renal epithelial cells with human MDR1 gene (MDR1-LLC-PK1) and of Madin-Darby canine kidney cells with mouse ABCG2 gene (mBcrp-MDCK) and control vector (MDCK) was conducted at Amgen (Thousand Oaks, CA).
Care and Maintenance of Animals.
All the animal procedures were conducted under protocols approved by the Amgen Institutional Animal Care and Use Committee (Cambridge, MA). Male Sprague-Dawley rats, 300 to 325 g, were obtained from Charles River Laboratories, Inc. (Wilmington, MA). For pharmacokinetic studies, rats were obtained from the vendor with catheters implanted in the femoral artery and vein. The rats were housed in a temperature- and humidity-controlled environment subject to a 12:12-h light/dark cycle and had access to water and a standard laboratory rodent diet ad libitum. Animals were allowed to acclimate for 1 week before use.
Intravenous Pharmacokinetic Studies in Male Rats.
Rats were administered a single dose of test material (0.25 mg/kg) by bolus intravenous injection into the femoral vein catheter, and the catheter was flushed with blank vehicle to ensure the full dose was given. Blood samples were collected from the femoral artery catheter at 0.083 (5 min), 0.25, 0.5 1, 2, 4, 6, 8, and 24 h postdose. Plasma was separated by centrifugation and stored at −80°C. Plasma samples were extracted by the addition of acetonitrile containing 0.1% formic acid and an internal standard. Plasma samples were analyzed using the liquid chromatography/tandem mass spectrometry method (LC/MS/MS) system described below.
Isolation of Rat Hepatocytes.
Rat hepatocytes were isolated using the two-step collagenase method described by Seglen (1973) with some modifications. In brief, the rat was euthanized via CO2 asphyxiation, and the liver was exposed and cannulated within 3 min. The vena cava was cannulated caudal to the hepatic vein using PE-60 tubing (Clay Adams; INTRAMEDIC Polyethylene Tubing, Parsippany, NJ), and the portal vein was bisected. The liver was perfused with calcium-free HBSS supplemented with HEPES, sodium bicarbonate, 1 mM EDTA, and 1% bovine serum albumin (BSA) until it was cleared of blood. The liver was perfused further with HBSS containing 0.05% type IV collagenase and 3 mM calcium chloride until extensive dissociation was evident. The liver was agitated gently to release cells from the capsule into ice-cold phenol-free DMEM supplemented with 10% FBS and 2 mM glutamine. The cells were then filtered through sterile gauze and centrifuged at 50g for 5 min at 4°C. The resuspended cells were mixed with equal volume of Percoll in phosphate-buffered saline and centrifuged at 200g for 10 min. The pellet was washed a final time and resuspended in DMEM supplemented with 2 mM glutamine. The viability was determined using the trypan blue exclusion method, and only cells with a viability of greater than 85% were used.
Determination of CLint in RLM and Fresh Rat Hepatocytes.
CLint were determined in duplicate using the substrate depletion method (Obach, 1999). Test compounds were dissolved in dimethyl sulfoxide as 1 to 5 mM stock solution for in vitro experiments; the final concentration of dimethyl sulfoxide in the incubation was ≤0.1%. For the microsomal experiments, the 400-μl incubation mixture (96-well plates) contained 0.25 mg of microsomal protein/ml, 1 mM NADPH, and 2 mM magnesium chloride in 50 mM potassium phosphate buffer, pH 7.4. Metabolism was initiated by addition of test compounds to the prewarmed (37°C) incubation mixtures to achieve a final concentration of 1 μM. At 0, 10, 20, 30, and 40 min, 35-μl aliquots of incubation mixture were collected into an equal volume of acetonitrile.
For the hepatocyte experiments, cells suspended in DMEM were aliquoted into 24-well tissue culture plates at 0.5 million viable cells/well (0.8 ml/well) and were preincubated under 5% CO2 at 37°C for 15 to 30 min. Metabolism was initiated by addition of 200 μl of DMEM containing 5 μM test compound (37°C, final incubation volume of 1 ml and concentration of 1 μM). The plates were incubated without shaking under 5% CO2 at 37°C. At 0, 10, 20, 30, 45, and 60 min, 50-μl aliquots were transferred to a 96-well plate, which contained 200 μl of acetonitrile with 0.1% formic acid per well. Viability of hepatocytes was 50 to 60% at the end of incubation.
The samples were centrifuged at 3500g for 15 min, and the supernatants were analyzed using the LC/MS/MS method described below. For each compound, peak areas at each time point were converted to the natural log of the percentage remaining relative to the 0-min samples (Obach, 1999). The resulting slope of these values relative to time (k) was converted to in vitro t1/2 where t1/2 = −0.693/k. CLint was calculated using the following relationship: CLint = (0.693/t½) × (1/M), where M is the concentration of microsomes or hepatocytes in the incubation. For the scaling of rat CLint, values of 40 g liver/kg b.wt. (Davies and Morris, 1993), 45 mg of microsomal protein/g liver (Naritomi et al., 2001), and 120 × 106 hepatocytes/g liver (Naritomi et al., 2003) were assumed.
Determination of Fraction Unbound in Rat Plasma, Liver Microsomes, Hepatocytes, and Blood-to-Plasma Concentration Ratio.
The fup was determined in triplicate using an ultracentrifugation technique (Nakai et al., 2004). Plasma was spiked with test compound to achieve a final concentration of 5 μM and was centrifuged at 600,000g for 3 h at 37°C. Aliquots of the middle fraction were transferred into an equal volume of blank plasma and extracted with 5 volumes of acetonitrile. Aliquots of the original spiked plasma were mixed with an equal volume of plasma and 2 volumes of plasma water (ultrafiltrate), and extracted with 10 volumes of acetonitrile.
The fuinc in rat liver microsomal or hepatocyte suspensions was determined in duplicate by equilibrium dialysis using the rapid equilibrium dialysis device (Thermo Fisher Scientific, Waltham, MA). For liver microsomal binding experiments, microsomes (0.25 mg/ml) were suspended in 50 mM potassium phosphate buffer, pH 7.4. For hepatocyte binding experiments, the hepatocyte reaction mixture described above underwent a freeze-thaw cycle and then was treated with aminobenzotriazole (100 μM) to inhibit metabolism. Microsomal or hepatocyte suspensions were spiked with test compounds to achieve a final concentration of 1 μM and were dialyzed against 1.7 volumes of the corresponding buffer for 5 h at 37°C. Preliminary experiments with selected compounds indicated that a 5-h incubation time was sufficient to achieve equilibrium. Aliquots of the dialysate were transferred into an equal volume of blank suspension (microsomes or hepatocytes) and extracted with 2 volumes of acetonitrile. Samples of the dialyzed drug suspensions were mixed with an equal volume of buffer and extracted with 2 volumes of acetonitrile.
All the samples from the binding studies were centrifuged at 3500g for 15 min and analyzed by LC/MS/MS as described below. The fup was calculated using the following relationship: fup = middle fraction peak area/(2 × spiked plasma peak area). The fuinc was calculated using the relationship: fuinc = dialysate peak area/dialyzed drug-suspension peak area.
Blood-to-plasma concentration ratio was determined in vitro after incubation of compound with fresh rat whole blood. Blood was warmed to 37°C, and test compound was spiked to a concentration of 1 μM. The blood samples were processed for plasma after 1-h incubation at 37°C. Reference plasma was also spiked with test compound to a concentration of 1 μM. Samples were extracted with 6 volumes of acetonitrile containing internal standard, centrifuged, and analyzed on the LC/MS/MS system described below. The blood-to-plasma concentration ratio was calculated by dividing the peak area observed in the reference plasma (representing nominal blood concentration) by the peak area observed in the treated plasma (representing plasma concentration).
Transport across MDR1-LLC-PK1, mBcrp-MDCK, and MDCK Cells.
MDR1-LLC-PK1 cells were grown in Medium 199 supplemented with 10% FBS in the presence of vincristine (640 nM), which was used to maintain high P-gp activity over passages. MDCK and mBcrp-MDCK cells were maintained in DMEM-GlutaMax supplemented with 10% FBS and 1 mM sodium pyruvate. Transport studies were conducted according to the method described by Polli et al. (2001) with some modification. Cells were seeded onto Transwell filter membranes (surface area = 0.7 cm2) at a density of 200,000 cells/well. Compound incubations (in duplicate) were performed 5 days postseeding. To determine efflux in MDR1-LLC-PK1 cells, compounds were tested at 5 μM in the presence of 0.1% BSA (our in-house standard incubation condition for P-gp screening that provides good analytical sensitivity and recovery for majority discovery compounds). For mBcrp-MDCK cells, compounds were tested at 1 μM in the absence of BSA, which is a more optimal incubation condition for compounds having good analytical sensitivity and low nonspecific binding to plastics and/or cells. Passive permeability (PappA→B) was determined in MDCK cells; compounds were tested at 5 μM without BSA in the presence of the known P-gp inhibitor GF120918 (2 μM) to eliminate the contribution of endogenous P-gp. In the absence of BSA, low recovery prevented the accurate estimation of Papp values for a few compounds; therefore, those compounds were retested in the presence of 0.1% BSA, which tends to improve recovery for many hydrophobic compounds. Transport studies were conducted at 37°C in a humidified incubator with shaking (70 rpm) for 120 min. Samples were analyzed by LC/MS/MS as described below.
The Papp, recovery, and efflux ratios were calculated according to the following equations (Polli et al., 2001): where A is the membrane surface area, C0 is the donor drug concentration at t = 0, dQ/dt is the amount of drug transported within a given period, CAt and CBt are the concentrations in the apical and basolateral chambers at time (t), VA and VB are the volumes of the apical and basolateral chambers, and VD is the volume of the donor. PappB→A and PappA→B are the Papp values in the basolateral to apical and apical to basolateral direction.
LC/MS/MS Analysis.
The LC/MS/MS system consisted of two LC-10AD high-performance liquid chromatography pumps and a DGU-14A degasser (Shimadzu, Columbia, MD), a CTC PAL autoinjector (LEAP Technologies, Carrboro, NC), and either an API3000 or API4000 mass spectrometer (Applied Biosystems, Foster City, CA), according to the requirements of the compound. For each analyte, the mass spectrometer electronics were tuned to the most intense mass transition (see supplemental figures for representative LC/MS/MS spectrum and chromatogram).
Samples from in vitro experiments were injected onto a Sprite Armor C18 analytical column (20 × 2.1 mm, 10 μm pore size; Analytical Sales and Products, Pompton Plains, NJ) with a 0.5-μm PEEK guard filter (Analytical Sales and Products). Analytes were separated using a gradient solvent system consisting of two components: solvent A (0.1% formic acid in water) and solvent B (0.1% formic acid in acetonitrile). The percentage of solvent B was increased in a linear fashion from 2 to 95% over 2 min; flow rate was 0.4 ml/min.
Samples from in vivo studies were injected onto a Waters (Milford, MA) YMC ODS A column (20 × 2.0 mm, 5 μm pore size) and separated using the mobile phase gradient described above. Raw data were collected using Analyst 1.4 (Applied Biosystems). The area under the concentration versus time curve from time 0 to infinity (AUC0-inf) was calculated in Small Molecules Discovery Assay Watson (version 7.0.01; InnaPhase Corp., Waltham, MA) using the linear/log trapezoidal method. CL was calculated in Small Molecules Discovery Assay Watson according to the equation: CL = Dose/AUC0-inf.
Prediction of Rat Hepatic Clearance.
Predicted CLh was calculated according to the well stirred model (Ito and Houston, 2004) as follows: where CLh is hepatic blood clearance, fub is the fraction of drug unbound in blood, CLint is the in vitro intrinsic clearance obtained from rat liver microsomes or hepatocyte suspensions, fuinc is the fraction of unbound drug in the microsomal or hepatocyte incubations, and Qh is hepatic blood flow in rat (55.2 ml/min/kg) (Davies and Morris, 1993). The overall accuracy of clearance prediction was determined by average -fold error (AFE) as follows (Obach, 1999):
Calculated Physicochemical Properties.
Physicochemical descriptors were calculated using commercial software packages: Daylight (Daylight Chemical Information Systems, Aliso Viejo, CA) for polar surface area (PSA) and cLogP calculation, and Pipeline Pilot (Accelrys Software Inc., San Diego, CA) for calculating molecular weight and the number of hydrogen bond donors and acceptors.
Results
Intrinsic Clearance of Proprietary Compounds in RLM and Hepatocyte Suspensions.
The compound set displayed a wide range of CLint from <5 to 865 μl/min/mg in RLM and <5 to 126 μl/min/million cells in hepatocytes (Table 1). When corrected for the binding of drug to microsomes or hepatocytes, unbound CLint determined from RLM and hepatocytes was comparable for the majority of the compounds (Fig. 1). However, a significant difference in unbound CLint between RLM and hepatocytes was observed for some compounds. The unbound CLint of compounds 28, 30, 36, 39, and 40 was more than 3-fold higher in hepatocytes than in RLM. In contrast, the unbound CLint of compounds 14, 21, and 37 was more than 3-fold lower in hepatocytes than the very high CLint observed in RLM.
Plasma Protein, Microsomal, and Hepatocyte Binding of Proprietary Compounds.
In general, the compound set was highly protein bound in rat plasma, with fup ranging from 0.001 to 0.097 (Table 1). Overall, hepatocyte and microsomal binding of the compound set was comparable under the experimental conditions (microsomal concentration of 0.25 mg/ml and hepatocyte density of 0.5 million cells/ml) (Table 1). Most compounds had fuinc of greater than 0.5, whereas a few compounds had fuinc of less than 0.5. Compound 41 was highly bound to both microsomes and hepatocytes; the fuinc for this compound was 0.07 in both microsomal and hepatocyte incubations.
Prediction of Rat CL.
The in vivo plasma CL of the test compounds ranged from 3 to 7964 ml/h/kg (Table 1). Blood-to-plasma concentration ratio was determined for a few compounds, and the tested compounds did not distribute into blood cells extensively. Based on this finding, a blood-to-plasma partition ratio of 1 was assumed for all the compounds; therefore, plasma CL was considered to be equal to blood CL, and fup was assumed to be equal to fub. For 28 of 41 compounds examined, CLint derived from RLM underpredicted CL by more than 3-fold, with the underprediction as great as 541-fold (Fig. 2A). The AFEs using RLM data were 7.6. Although CLint from hepatocytes provided better prediction than RLM for some compounds, CL remained underpredicted for the majority. CL was underpredicted by more than 3-fold for 31 compounds, with underprediction up to 70-fold (Fig. 2B). The AFE using rat hepatocytes was 7.5.
Passive Permeability and Efflux of the Proprietary Compounds.
Passive permeability was estimated from MDCK cells, a cell line commonly used for in vitro permeability tests (Irvine et al., 1999). Under our experimental conditions, the Papp values for the paracellular marker compound atenolol and the transcellular marker compound propranolol were ≤2 and 28.5 × 10−6 cm/s, respectively. The passive permeability of the test compounds ranged from <3.2 to 46 × 10−6 cm/s, with the majority displaying passive permeability greater than 5 × 10−6 cm/s (Table 1).
Because P-gp species difference in substrate specificity between human and mouse is minimal (Feng et al., 2008) and the ABCG2 gene is highly conserved (Robey et al., 2009), it was assumed that there was no species difference in P-gp- and Bcrp-mediated transport for the test compounds. P-gp-mediated transport was evaluated using MDR1-LLC-PK1 cells, which express human P-gp. Loperamide (1 μM), a known P-gp substrate, exhibited an efflux ratio greater than 20. Bcrp-mediated transport was evaluated using mBcrp-MDCK cells, which express mouse Bcrp. Prazosin, which was used as a control for the Bcrp assay, displayed an efflux ratio greater than 15. Many of the compounds in the set were good substrates for both P-gp and mBcrp. The efflux ratios ranged from 0.6 to 46 in MDR1-LLC-PK1 cells and from 1.1 to 41 in mBcrp-MDCK cells.
Classification and Physicochemical Properties of the Proprietary Compounds.
The 41 proprietary compounds were classified into three categories based on passive permeability and efflux ratios in MDR1-LLC-PK1 and mBcrp-MDCK cells. The 12 type I compounds were characterized by passive permeability values greater than 5 × 10−6 cm/s and efflux ratios less than 5 in both MDR1-LLC-PK1 and mBcrp-MDCK cells. The 20 type II compounds displayed passive permeability greater than 5 × 10−6 cm/s and efflux ratios greater than 5 in either cell line. The remaining nine compounds were categorized as type III, which had passive permeability values less than 5 × 10−6 cm/s in MDCK cells and efflux ratios ranging from <1 to 26.6 (Tables 1 and 2).
The accuracy in predicting CL from CLint varied significantly among the three classes of compounds. For type I compounds, in vivo CL ranged from 3 to 2660 ml/h/kg. In vitro CLint derived from either RLM or hepatocytes predicted in vivo CL reasonably well, with AFEs of 1.8 and 2.3, respectively (Fig. 3A; Table 2). The in vivo CL of type II compounds ranged from 15 to 7600 ml/h/kg. CL was underpredicted from RLM by more than 3-fold for 15 of 20 type II compounds with an AFE of 7.4. A similar degree of underprediction was observed when hepatocytes data were used to predict CL for this class of compounds (Fig. 3B; Table 2). All the type III compounds exhibited high CL; for the majority, CL exceeded hepatic blood flow (Table 1). In vitro CLint dramatically underpredicted CL for type III compounds. The AFE was 54.3- and 29.2-fold using RLM and rat hepatocytes, respectively (Fig. 3C; Table 2).
Selected physicochemical properties of the 41 proprietary compounds used in the present study are shown in Table 3. Molecular weight ranged from 337 to 523 mol/g, and PSA ranged from 56 to 137. Most compounds were hydrophobic and neutral, whereas some were basic. Statistically significant differences (P < 0.01) were observed in the physiochemical properties among the three types (Table 4). Some physical chemical properties may be correlated to each other. All the compounds that displayed PSA less than 100 contained no hydrogen bond donors, whereas most of the compounds with PSA greater than 100 contained one to two hydrogen bond donors. In vitro CLint provided better prediction for compounds with lower PSA and fewer hydrogen bond acceptors and donors (Table 5).
Permeability, Efflux Status, Physical Chemical Properties, and Classification of 16 Marketed Drugs.
Sixteen marketed drugs that exhibited a wide range of human in vivo clearance (0.3–19.2 ml/min/kg) and varying degrees of underprediction of CL from in vitro metabolic CLint (Obach, 1999; Riley et al., 2005) were selected for permeability and efflux determinations (Table 6). Nine compounds, including buspirone, clozapine, diclofenac, and imipramine, displayed high passive permeability and low efflux ratios in both MDR1-LLC-PK1 and mBcrp-MDCK cells that are characteristics of type I compounds. In vivo clearance of these type I drugs was well predicted using CLint derived from either human hepatocytes or human liver microsomes. The AFE for these compounds was 1.7. The in vivo CL of bosentan, cyclosporine, montelukast, and troglitazone in humans was dramatically underpredicted by in vitro CLint. These four drugs exhibited poor permeability and could be classified as type III compounds. The type II compounds indomethacin, prazosin, and ritonavir, for which human clearance was not well predicted, had high passive permeability and were identified as P-gp and/or mBcrp substrates.
Selected physicochemical properties of the 16 marketed drugs (see supplemental table for structure) are shown in Table 7. Similar to the proprietary compounds examined, drugs that were classified as type I tended to have lower molecular weight and PSA and fewer hydrogen bond acceptors and donors than type II and type III drugs.
Discussion
To understand the reasons for the lack of correlation between in vitro CLint and in vivo CL, we investigated the quantitative prediction of rat CL from in vitro CLint using RLM and hepatocytes for 41 proprietary compounds. Consistent with previous reports (Obach, 1999; Riley et al., 2005; Brown et al., 2007; Stringer et al., 2008), the current results showed that in vitro CLint from RLM or hepatocytes was predictive of CL for some compounds but significantly underpredicted CL for many others using the well stirred model. A similar degree of underprediction was also observed if the parallel tube or the dispersion model (Ito and Houston, 2004) was used.
To investigate potential causes for the underprediction, we examined the relationship between predictability, passive permeability, and efflux status. Based on the data generated, the 41 compounds were grouped into three categories that exhibited different degrees of underprediction. For type I compounds, which exhibited good passive permeability and low efflux ratios in both MDR1-LLC-PK1 and mBcrp-MDCK cells, in vivo CL in rats was predicted reasonably well from in vitro CLint. Type II compounds, which displayed good permeability and high efflux ratios in either cell line, showed a wide range of CL. CL was predicted within 3-fold error for only a few compounds; the AFE for compounds in this group was 7 to 8. In contrast to type I and II compounds, all the type III compounds were poorly permeable (Papp <5 × 10−6 cm/s), and most displayed CL that exceeded hepatic blood flow. In vitro CLint obtained from RLM dramatically underpredicted in vivo CL for type III compounds. The magnitude of the underprediction was reduced to some extent when CLint derived from hepatocyte data was used to predict CL; however, none of these compounds' CL was predicted within 3-fold error.
Preliminary data indicated that these type III compounds were stable in rat plasma and did not distribute extensively into blood cells (data not shown). Therefore, plasma instability and partitioning into blood cells were unlikely to be the reasons for the underprediction of CL. Tang et al. (2008) have suggested that the sampling scheme used in pharmacokinetics studies can have a profound effect on the calculation of in vivo CL. The addition of a 2-min time point was reported to reduce CL by more than 2-fold. The in vivo CL data for our compound set were collected from discovery pharmacokinetic screening studies, in which the first blood sample was collected at 5 min postdose. Further studies are being conducted to evaluate the potential overestimation of in vivo CL for some compounds in the current set.
Physicochemical properties, such as lipophilicity, hydrogen bonding ability, and PSA, can affect the permeability and efflux status of a compound (Egan and Lauri, 2002; Stouch and Gudmundsson, 2002; Mälkiä et al., 2004). Analysis of the calculated physicochemical properties revealed statistically significant differences in some properties among the three groups of compounds. Type I compounds had lower molecular weight, smaller PSA, and fewer hydrogen acceptors than either type II or type III compounds. None of the type I compounds contained any hydrogen bond donors. In contrast, most of type II and III compounds contained one to two hydrogen bond donors. PSA and the number of hydrogen bond acceptors and donors underlying permeability and efflux status appear to be distinguishing factors in determining whether in vitro CLint is predictive of in vivo CL for the compound set.
Sixteen marketed drugs were selected to determine whether permeability and efflux could contribute to the varying degrees of underprediction of human clearance from in vitro data reported in the literature. The human predictions cited in our report were mainly taken from the literature by Riley et al. (2005): either CLint in human hepatocytes from their laboratory or the mean values from multiple laboratories were used for the prediction. As shown by Riley et al., whereas variability ≤3-fold was observed for some compounds such as diclofenac, imipramine, and tolbutamide, variability was more significant for some other compounds such as diltiazem and zolpidem, suggesting that the accuracy of human clearance prediction may differ depending on possible interlaboratory variability of in vitro data.
The 16 drugs were also classified into three categories based on permeability and efflux data. Similar to the proprietary compounds examined, type I-marketed drugs tend to have lower molecular weight, PSA, and fewer hydrogen bond acceptors and donors than type II- and III- marketed drugs. Despite the potential impact of interlaboratory variability of in vitro data on prediction, the trend observed with our internal data in rats appears to be applicable to the human predictions reported in the literature. Consistent with accurate prediction of rat clearance, the human CL for the type I-marketed drugs could be predicted well from in vitro metabolic CLint (Obach, 1999; Riley et al., 2005). As was observed for the rat, metabolic CLint underpredicted human CL for type II- and III-marketed drugs such as bosentan, montelukast, and prazosin (Riley et al., 2005). Our results suggested that the underprediction of CL for these drugs was in part because of the involvement of transporters in their elimination. Type II compounds were good substrates for P-gp and/or Bcrp, which are expressed not only in the liver but also in the kidney and intestine (Shitara et al., 2006). Therefore, biliary and intestinal secretion could play a significant role in the elimination of some type II compounds. Imatinib, which is a substrate for P-gp and Bcrp and displays good permeability, is excreted primarily as metabolites in mice (Dai et al., 2003; Burger et al., 2004). After intravenous administration of imatinib, systemic exposure was approximately 60% higher and fecal excretion was 5-fold lower in Mdr1a/1b(−/−)/Bcrp(−/−) mice versus wild-type mice (Oostendorp et al., 2009). Studies in our laboratory showed that the biliary excretion of compound 18 was reduced by 90% in Bcrp knockout mice (data not shown). P-gp-mediated secretion from the systemic circulation into the intestinal lumen has also been reported for several drugs, including apafant (Mayer et al., 1996; Sparreboom et al., 1997; Leusch et al., 2002). Apafant has passive permeability of approximately 10 × 10−6 cm/s and an efflux ratio of 9 in Caco-2 cells. After an intravenous administration of [14C]apafant to bile duct-catheterized mice, intestinal excretion of radioactivity was 54.9 and 6.8% of the dose in wild-type and Mdr1a(−/−) mice, respectively (Leusch et al., 2002). The interplay between transporters and metabolic enzymes could influence the predictability of CL from in vitro metabolic CLint as well. In Caco-2 cells expressing CYP3A4, P-gp inhibition reduces the extent of metabolism of K77, which is a substrate for both CYP3A4 and P-gp. In contrast, the metabolism of felodipine, which is a substrate for CYP3A4 only, is not affected by P-gp inhibition (Cummins et al., 2002), supporting a role for P-gp in modulating metabolism.
The drastic underprediction of CL from metabolic CLint for type III compounds suggested that permeability might have the greatest influence on disposition. Our results were consistent with the trend described by Wu and Benet (2005) that the predominant routes of elimination for poorly permeable compounds were renal and/or biliary excretion of unchanged drug rather than metabolism. Using metabolic CLint in the well stirred model to predict in vivo CL relies on the assumption that rapid equilibrium between blood and hepatocytes can be reached, an assumption which may be valid only for compounds that exhibit high passive permeability. Poorly permeable compounds may need to use uptake and efflux transporters to cross cellular membranes efficiently. Bosentan and montelukast, which are mainly excreted into bile as metabolites in humans (Balani et al., 1997; Weber et al., 1999) and displayed poor passive permeability, have been shown to be substrates for uptake transporters (Treiber et al., 2007; Mougey et al., 2009). Hepatic uptake could be a rate-determining step in their elimination. Further investigation is required into the significance of uptake transporters in the elimination of type III compounds and the prediction of CL from in vitro data.
In summary, the accuracy of predicting CL from in vitro metabolic CLint could be differentiated by permeability and P-gp- and mBcrp-mediated efflux for the compound set. Clearance could be predicted reasonably well from in vitro metabolic CLint for compounds that displayed high passive permeability (>5 × 10−6 cm/s) and that were not good substrates (efflux ratio <5) for efflux transporters such as P-gp and mBcrp. For compounds with poor passive permeability (<5 × 10−6 cm/s) or for compounds that were good P-gp and/or mBcrp substrates (efflux ratio >5) and exhibited high passive permeability, in vitro metabolic CLint was unlikely to be predictive of in vivo CL because of the significant role that transporters might play in their elimination. These data show that passive permeability and efflux must be taken into consideration when attempting to predict in vivo CL from in vitro CLint, and additional models need to be developed.
Acknowledgments.
We thank the Department of Medicinal Chemistry, Amgen (Cambridge, MA) for compound synthesis. We also thank Earl Moore for analyzing some in vitro samples, Dr. Alan Cheng for calculation of the physical chemical properties of marketed drugs, and Dr. Zhiyang Zhao for insightful scientific review and comments of the manuscript.
Footnotes
-
Article, publication date, and citation information can be found at http://dmd.aspetjournals.org.
doi:10.1124/dmd.109.029066.
-
↵ The online version of this article (available at http://dmd.aspetjournals.org) contains supplemental material.
-
- CLint
- in vitro intrinsic metabolic clearance
- CL
- clearance
- fub
- fraction of unbound drug in blood
- fuinc
- fraction of unbound drug in the microsomal or hepatocyte incubation
- fup
- fraction of unbound drug in plasma
- P-gp
- P-glycoprotein
- Bcrp
- breast cancer resistance protein
- RLM
- rat liver microsome
- DMEM
- Dulbecco's modified Eagle's medium
- HBSS
- Hanks' balanced salt solution
- FBS
- fetal bovine serum
- MDR1-LLC-PK1
- porcine renal epithelial cells transfected with human MDR1 gene
- mBcrp-MDCK cells
- Madin-Darby canine kidney cells transfected with mouse ABCG2 gene
- MDCK
- Madin-Darby canine kidney cells transfected with control vector
- LC/MS/MS
- liquid chromatography tandem mass spectrometry system
- BSA
- bovine serum albumin
- Papp
- apparent permeability value
- GF120918
- N-(4-[2-(1,2,3,4-tetrahydro-6,7-dimethoxy-2-isoquinolinyl)ethyl]-phenyl)-9,10-dihydro-5-methoxy-9-oxo-4-acridine carboxamide
- AUC0-inf
- area under the concentration versus time curve from time 0 to infinity
- AFE
- average -fold error
- PSA
- polar surface area.
- Received June 18, 2009.
- Accepted October 27, 2009.
- Copyright © 2010 by The American Society for Pharmacology and Experimental Therapeutics