Abstract
Accurate prediction of drug target activity and rational dosing regimen design require knowledge of drug concentrations at the target. It is important to understand the impact of processes such as membrane permeability, partitioning, and active transport on intracellular drug concentrations. The present study aimed to predict intracellular unbound atorvastatin concentrations and characterize the effect of enzyme-transporter interplay on these concentrations. Single-pass liver perfusion studies were conducted in rats using atorvastatin (ATV, 1 µM) alone at 4°C and at 37°C in presence of rifampin (RIF, 20 µM) and 1-aminobenzotriazole (ABT, 1 mM), separately and in combination. The unbound intracellular ATV concentration was predicted with a five-compartment explicit membrane model using the parameterized diffusional influx clearance, active basolateral uptake clearance, and metabolic clearance. Chemical inhibition of uptake and metabolism at 37°C proved to be better controls relative to studies at 4°C. The predicted unbound intracellular concentration at the end of the 50-minute perfusion in the +ABT , +ABT+RIF, and the ATV-only groups was 6.5 µM, 0.58 µM, and 5.14 µM, respectively. The predicted total liver concentrations and amount recovered in bile were within 0.94–1.3 fold of the observed value in all groups. The fold difference in total liver concentration did not always extrapolate to the fold difference in predicted unbound concentration across groups. Together, these results support the use of compartmental modeling to predict intracellular concentrations in dynamic organ-based systems. These predictions can provide insight into the role of uptake transporters and metabolizing enzymes in determining drug tissue concentrations.
Introduction
The drug concentration–response relationship at the target determined in vitro, along with plasma drug concentration–time profiles in vivo, are used to predict the pharmacodynamic (PD) effect of a drug. When plasma concentrations are used for in vitro–in vivo correlations, the underlying assumption is that at steady state, the unbound concentration in plasma is equal to the unbound concentration inside the cell as proposed by the free drug hypothesis (Koch-Weser and Sellers, 1976; Israili, 1979; Pardridge et al., 1983). This approach has limitations, particularly for drugs that have poor permeability and are transporter substrates, since the free drug hypothesis is violated in these scenarios. In vivo concentrations in clearance organs and at the target can be significantly different from the plasma concentrations. For example, cimetidine is a poorly permeable compound (log D7.4 = −0.3). Therefore, the concentration of this drug in cerebrospinal fluid is 8- and 5-fold lower than the plasma concentration in dogs (Ziemniak et al., 1984) and humans (Somogyi and Gugler, 1983), respectively. Single-nucleotide polymorphisms in ABCB1 and SLC28A2 genes correlated with the intracellular accumulation of the anti-hepatitis C virus drug boceprevir in peripheral blood mononuclear cells (Cusato et al., 2015). Underprediction of cytochrome P450 (P450)-mediated drug drug interactions (DDIs) is observed when perpetrators are also substrates for uptake transporters (Yoshida et al., 2013). This underprediction is likely due to intracellular perpetrator concentrations higher than plasma concentrations. In some instances, however, transporter effects can be used to design favorable DDIs. Protease inhibitors atazanavir and darunavir are P-glycoprotein (P-gp) inhibitors, and tenofovir is a P-gp substrate. Coadministration of the P-gp inhibitors atazanavir and darunavir increased the intracellular accumulation of tenofovir in peripheral blood mononuclear cells, resulting in decreased HIV-1 RNA to below detectable limits (Lahiri et al., 2015). Since the metabolic machinery for most drugs and the targets for many drugs are located inside the cell, it is the unbound intracellular concentration, and not the plasma concentration, that drives the metabolism and efficacy of many drugs.
Human organic anion transporting polypeptides (OATPs) are uptake transporters of significance in tissue distribution, renal and hepatic clearance, and intestinal absorption of drugs (Nozawa et al., 2005; Maeda et al., 2006; Kalliokoski and Niemi, 2009). Several drug classes susceptible to transport by OATPs include endothelin receptor antagonists (Treiber et al., 2007), cardiac glycosides (Bossuyt et al., 1996), angiotensin II receptor antagonists (Yamashiro et al., 2006) and importantly, hydroxymethyl glutaryl (HMG) CoA inhibitors (statins) (Hsiang et al., 1999; Hirano et al., 2004; Schneck et al., 2004; Kameyama et al., 2005). Uptake transporters like OATPs increase the intracellular concentration of their substrates in the hepatocytes, and these concentrations can be many-fold higher than their plasma concentrations. Polymorphisms in OATPs result in differential hepatocyte versus plasma exposure of substrates and have significant impact on drug pharmacokinetic (PK) (Chung et al., 2005; Lee et al., 2005; Niemi et al., 2005; Katz et al., 2006; Xiang et al., 2006; Zhang et al., 2006). This is crucial especially in the case of statins since rhabdomyolysis is a severe side effect associated with statin use. Cerivastatin was voluntarily withdrawn from the market in 2001 owing to cases of rhabdomyolysis followed by resultant renal failure (Furberg and Pitt, 2001).
Current methods to indirectly determine intracellular free drug concentrations include microdialysis, fluorescence imaging, tomography imaging, capillary electrophoresis, equilibrium dialysis of tissues, microscopic imaging and particle induced photon emission (Chu et al., 2013), secondary ion MS (Dollery, 2013), and differential centrifugation in sandwich-cultured hepatocytes (Pfeifer et al., 2013). Most of these techniques have practical or financial limitations. Thus, experimental measurement of intracellular free drug concentrations remains difficult. Mathematical modeling of cell systems without explicit membrane compartments has been previously reported (Yabe et al., 2011; Ménochet et al., 2012; Shitara et al., 2013). We have previously developed a five-compartment (5-C) model with explicit membranes. Processes such as membrane partitioning (and resulting experimental lag times) and transport out of the membrane can be accurately modeled with the inclusion of explicit membrane compartments (Knipp et al., 1997; Korzekwa et al., 2012).
The goal of this study was to predict unbound intracellular concentration of atorvastatin (ATV) (Baumann et al., 1992) using a 5-C organ model with explicit membrane compartments. Single-pass liver perfusion experiments were conducted in presence and absence of inhibitors of active uptake transport and metabolism to predict the unbound intracellular concentrations of ATV.
Materials and Methods
Materials
ATV calcium trihydrate was obtained from Tokyo Chemical Industry (Portland, OR). Pitavastatin was obtained from Toronto Research Chemicals (Toronto, Canada). Rifampin (RIF) for injection USP was obtained from Sanofi-Aventis (Bridgewater, NJ). 1-Aminobenzotriazole (ABT), dextran, sodium bicarbonate, sodium taurocholate, glucose and dexamethasone were obtained from Sigma-Aldrich (St. Louis, MO). Sodium chloride, potassium chloride, calcium chloride, magnesium chloride, sodium monobasic phosphate, sucrose, potassium phosphate monobasic, and potassium phosphate dibasic were obtained from Fisher Scientific. Sodium sulfate was obtained from EMD Chemicals (Gibbstown, NJ). Male Sprague-Dawley rat plasma was obtained from BioChemed Services (Winchester, VA). Rat liver microsomes (RLMs) were obtained from Corning Life Sciences (Oneonta, NY). HSE UNIPER UP-100 Type 834 perfusion apparatus was obtained from Harvard Apparatus (Holliston, MA). A 96-well equilibrium dialyzer with a mol. wt. cutoff of 5 K and dual-plate rotator was obtained from Harvard Apparatus.
Animals
Male Sprague-Dawley rats (7–9 weeks old) were obtained from Charles River Laboratories, (Malvern, PA) and maintained in the American Association for the Accreditation of Laboratory Animal Care–accredited University Laboratory Animal Resources of Temple University. A normal rodent diet was provided to the animals. Food and water were available to the animals as required. The animals were housed in a standard 12-hour dark/light cycle. Animal studies were approved by the Institutional Animal Care and Use Committee of Temple University.
Liver Perfusion
In situ single-pass liver perfusion was performed with the HSE UNIPER UP-100 Type 834 for isolated organ perfusion in rodents. Although recirculating liver perfusion is more relevant to the in vivo system, single-pass perfusions were performed since both the hydroxy metabolites of ATV are substrates of the rat uptake transporter oatp1b2, and further complications could arise as a result of inhibitory effects of the metabolites on the uptake of the parent (Lau et al., 2006). The perfusion medium consisted of modified Krebs-Henseleit buffer with 3% dextran 40 and 30 µM sodium taurocholate. The perfusion medium was warmed to 37°C or cooled to 4°C, depending on the type of the experiment, and aerated with oxygen containing 5% carbon dioxide throughout the experiment. RIF and ABT were used as inhibitors of active uptake (Lau et al., 2006) and P450- mediated metabolism (Huijzer et al., 1989), respectively. Experiments were conducted with the following treatment groups: ATV alone at 4°C (n = 4), ATV alone at 37°C (n = 3), +RIF (n = 3), and +ABT+RIF (n = 3).
Liver perfusion surgery was performed under anesthesia using EZ-ANESTHESIA apparatus with 3% isoflurane in 2 liters/min oxygen. The inferior vena cava was sutured with a loose knot, and the bile duct was cannulated. Bile was collected on ice for the duration of the experiment. The portal vein was isolated and cannulated using an 18 G IV catheter. The outlet from the perfusion apparatus was connected to the catheter after allowing the blood from liver to drain out to avoid generation of air bubbles. The hepatic vein was cannulated using PE 100 tubing, and the inferior vena cava was tied up. The liver was allowed to equilibrate with blank perfusion medium for 20 minutes.
Passive Diffusion Study in Cold Perfused Liver.
Cold perfusions (4°C) were performed to determine passive diffusion and as a negative control for uptake and metabolism experiments. The perfusion medium was cooled down to 4°C, and the liver was covered with ice to maintain the temperature at 4°C. ATV solution in methanol (1 mg/ml) was added to the perfusate reservoir to give a final concentration of 1 μM. The final concentration of methanol in the perfusion medium was 0.05% v/v.
Passive Diffusion, Uptake, and Metabolism Studies in Perfused Liver at 37°C.
The perfusion medium was warmed to 37°C, and a heating lamp was used to maintain the liver temperature at 37°C. ATV (1 μM) was perfused alone, with either RIF or ABT, or with both. Passive diffusion was also determined at 37°C by combined inhibition of uptake and metabolism. To determine passive diffusion at 37°C, ABT (1mM) was perfused for 20 minutes, followed by switching the reservoir to rifampin (RIF, 20 μM) for 5 minutes, followed by the addition of ATV (1 μM), which was then coperfused with RIF for 50 minutes. For uptake inhibition studies, RIF (20 μM) was perfused 5 minutes before the addition of ATV (1 μM). RIF was then coperfused along with ATV for 50 minutes. Previous studies in multidrug resistance gene 1 - overexpressing Madin-Darby canine kidney epithelial (MDCK) cells have demonstrated that RIF did not inhibit P-gp up to 50 µM (Lau et al., 2006). For metabolism inhibition studies, ABT (1mM) was perfused for 20 minutes, followed by switching of the reservoir solution to 1 μM ATV. It has been previously reported that ABT does not interfere with the uptake transport by human OATP, P-gp, BCRP, and MRP2 up to 1 mM (Plise et al., 2010). Inhibition of metabolism was also tested by coperfusion with ketoconazole (1 μM) along with ATV (data not shown).
Bile was collected on ice over intervals of 0–15, 15–30, and 30–50 minutes. Errors were propagated for the calculation of cumulative amount in bile. Outlet perfusate from hepatic veins was sampled at 0, 5, 10, 20, 30, 40, and 50 minutes. During the entire course of the experiment, the liver was covered with sterile gauze kept moist by occasional spraying of saline. Inlet reservoir concentration was also sampled for analysis. The liver was removed at the end of 50 minutes, flash frozen on dry ice and homogenized with water in the ratio 1:2 before analysis. All perfusate, liver sample, and bile samples were stored at −80°C until further analysis.
Equilibrium Dialysis
Unbound fractions of ATV were determined in RLMs (fum), liver homogenate (ful), and rat hepatocytes (fuh). Equilibrium dialysis was performed using a 96-well equilibrium dialyzer with mol. wt. cutoff of 5 K and placed in dual-plate rotator set to maximum speed (25 rpm) located in a 37°C incubator with a 5% CO2 atmospheric environment. Dialysis buffer consisted of 0.1 M phosphate buffer containing 3 mM MgCl2, pH = 7.4. The pH was adjusted to 7.4 for all the matrices using 1N HCl or 1N NaOH. RLMs (1 mg/ml), liver homogenates diluted 1:2 in water, and 4.45 × 106 hepatocytes/ml were used for determination of fum, fuL, and fuh respectively. The relevant matrix was spiked with 5 μl of 200 μM ATV stock solution to give 2 μM ATV in pH-adjusted matrix. This solution (200 μl) and an equal volume of buffer were added to the respective sides of the 96-well dialysis plate and allowed to dialyze for 22 hours. After 22 hours of dialysis, 125 μl of buffer and matrix solutions were removed from each side of the dialysis plate and mixed with 125 μl of the opposite matrix in a 96-deep-well plate. Samples were then stored at −80°C for future analyte quantitation. A total of four replicates were used for each matrix.
The unbound fractions thus obtained were corrected for dilution by using eq. 1 (Kalvass et al., 2002):
(1)where D is the dilution factor and fudil is the experimentally obtained unbound fraction.
Sample Preparation
Standard solutions for perfusate were prepared by serial dilution in blank perfusate. Standard solutions for all other matrices were prepared by spiking blank matrices with an equal volume of standard sample in 1:1 acetonitrile: water. To 250 μl of standard perfusate sample, 200 μl of the blank liver homogenate plus 200 μl of standard in water, 200 μl of standard dialysis buffer with RLMs or 20 μl of blank bile plus 20 μl of standard in water, 12.5 μl of 1 mM potassium monobasic phosphate was added along with 6.25 μl of 10 µg/ml pitavastatin (internal standard) and 1 ml of methyl-t-butyl ether. Samples were treated similarly except 200 μl of the sample liver homogenate, and 20 μl of bile samples were added to equal volume of water. The mixture was vortexed for 5 minutes and centrifuged at 10,000 rpm for 10 minutes, followed by immersion in dry ice for 1 hour. The sample was centrifuged again at −5°C at 10,000 rpm for 10 minutes. The organic phase was then separated and evaporated to dryness at 20°C under nitrogen. The sample was reconstituted in 25 μl or 50 μl for perfusate and liver homogenate, respectively, or with 20 μl for dialysis buffer with RLMs and bile samples, of 1:1 acetonitrile: water. Sample (5 μl) was injected for analysis into the liquid chromatography tandem mass spectrometry (LC-MS/MS) system.
LC-MS/MS
An Agilent series 1100 high-performance liquid chromatography system coupled to an ABSciex API 4000 triple-quadrupole tandem mass spectrometer with electrospray ionization (ESI) source operating at 400°C was used for analysis. The mobile phase for high-performance liquid chromatography separation consisted of water with 0.1% formic acid as the aqueous phase (A) and acetonitrile with 0.1% formic acid as the organic phase (B). The gradient used for elution started with 90% A at 0 minute to 5% A maintained from 0.5 minute to 3 minutes, followed by gradual increase to 90% A at 6.5 minutes maintained until 9 minutes for ATV. For 2- and 4-hydroxy ATV, the gradient started with 60% A at 0 minute to 5% A maintained from 0.5 minute to 3 minutes, followed by gradual increase to 60% A at 6.5 minutes maintained until 9 minutes. A ZORBAX extend C-18, 5 μm, 4.6 × 50-mm column (Agilent Technologies) maintained at 20°C and protected by a ZORBAX SB C-18, 5 μm, 4.6 × 12.5 mm guard column (Agilent Technologies) was used. The flow rate was 300 μl/min.
The LC-MS/MS method was validated, and the coefficient of variation for interday and intraday validation accuracy and precision was less than 15% for the range of concentrations tested.
Mathematical Modeling and Simulation
The data obtained from the liver perfusion experiments were modeled using the 5-C model depicted in Fig. 1 to predict the intracellular unbound concentration of ATV. This model has been described previously (Korzekwa et al., 2012) for cell-based systems. In the present study, perfusate and bile flow rates were included from the experiment, and metabolic pathways were modeled. Apical efflux was assumed to occur from the apical membrane (e.g., P-gp).
A five-compartment model with explicit membrane compartments for rat liver VA, VB, VC, VAM, VBM: volumes of apical, basolateral, cellular, apical membrane and basolateral membrane compartments, respectively; CA, CB, CC, CAM, CBM: concentration in apical, basolateral, cellular, apical membrane, and basolateral membrane compartments, respectively; CLi, CLo: diffusional clearance into and out of the membrane respectively; CLbu, active basolateral uptake clearance into the cell; CLae, apical efflux clearance into the apical compartment; Q, perfusion flow rate; Qbile, biliary flow rate; Cin, inlet concentration; Cout, outlet concentration.
Wolfram Mathematica (version 10.1) was used for modeling. The five compartments in Fig. 1 represent a rat liver. The basolateral compartment represents the blood compartment, and the apical compartment represents the biliary compartment within the liver. Hepatocytes have been further divided into apical membrane, basolateral membrane, and cytosolic compartments. The volume of membranes was assumed to be 10% of the total cell volume, with the apical and basolateral membranes each comprising 5% (Korzekwa et al., 2012). Of the total liver volume, 10.6% was assumed to be the sinusoidal volume (Blouin et al., 1977). The volume of hepatocytes in liver was calculated by using a rat liver hepatocellularity of 120 million cells per gram of liver (Sohlenius-Sternbeck, 2006) and a volume of 3.9 µl for a million hepatocytes (Reinoso et al., 2001). Volume of the intrahepatic biliary tree was set to 50 µl (Blouin et al., 1977; Masyuk et al., 2001). An initial estimate for the apical efflux clearance (CLae) was iteratively optimized with the +ABT + RIF data set to obtain the observed rate of biliary excretion with concentration in the apical membrane as the driving concentration. CLae was then scaled based on hepatocellularity for subsequent modeling. Initial estimates for metabolic clearance (CLm), diffusional influx clearance into the membrane (CLi), and active basolateral uptake clearance (CLbu) were obtained from the literature and scaled up to the organ level (Lau et al., 2006; Watanabe et al., 2010; Korzekwa et al., 2012). Backdiffusion from the bile into the liver was assumed to be minimal. Intracellular unbound concentration was the driving concentration for metabolism and apical membrane concentration was the driving concentration for apical efflux. This model allowed us to simulate perfusate, membrane, unbound intracellular, and biliary concentrations.
The interaction between various compartments within the model can be described by eq. 2–6:
(2)
(3)
(4)
(5)
(6)
The ratio of CLi and CLo, defined as the Kp of the drug into the membrane, is given by eq. 7:
(7)where, fum, VA, and Vm represent the unbound fraction in the membrane, volume of the partitioning incubation, and volume of the membrane, respectively. In 1 ml of microsomal solution with 1 mg/ml microsomal protein concentration, the volume of lipids is assumed to be 0.7 µl (Nagar and Korzekwa, 2012). Therefore, the ratio of Vm to VA was set to 0.0007. Since Vm was obtained from the amount of microsomes used in the equilibrium dialysis experiment, and the membrane volume was defined as 10% of cell volume in the model, fum used in this exercise was not corrected for dilution during the modeling procedure. Table 1 summarizes the experimental values and initial estimates of model parameters.
Initial estimates for model parameters
The NonlinearModelFit function in Mathematica was used with 1/Y weighting to parameterize various clearances. To get numeric solutions of the ordinary differential equations, the NDSolve function was used with MaxSteps → 100,000, and PrecisionGoal → ∞. To check the robustness of the optimizations, the initial estimates were varied 10-fold from the final parameter estimates.
For model development, the following stepwise optimization procedure was used. Average outflow perfusate concentration data from experiments with ABT + RIF were used to parameterize CLi. CLi estimation was also performed using the data from cold perfusions at 4°C. For CLi estimation, CLm and CLbu were set to 0. Upon obtaining model-fitted estimates for CLi, CLbu was parameterized next using average outflow perfusate concentration data obtained from the +ABT experiments with CLm set to 0. Average outflow perfusate concentration data from experiments with ATV alone were then used to parameterize CLm using the previously optimized values of CLi and CLbu. Finally, intracellular ATV concentrations were predicted using all the optimized clearance values.
Estimation of Kp
Kp for the entire liver was estimated using eq. 7. Kp obtained from +ABT + RIF data were compared with Kp obtained from 4°C data since metabolism and uptake are inhibited in both groups and the steady-state liver concentrations would, therefore, be equal to the Kp of the drug. The fum used was corrected for dilution. VA was the volume of the basolateral compartment, and Vm was the total membrane volume in the cell, which is 10% of the cell volume in the +ABT + RIF group.
Statistical Analyses
Statistical analyses were conducted using SigmaStat software, version 3.5. A t test was used to compare two groups. One-way analysis of variance was used for comparison among multiple groups. Post hoc analysis was conducted using Tukey’s test. A P value less than 0.05 was considered statistically significant.
Results
Passive Diffusion of ATV in the Perfused Liver.
The outflow perfusate concentrations of ATV upon inhibition of active transport and metabolism are shown in Fig. 2A. The outflow perfusate ATV concentration in 4°C studies achieved steady state by 10 minutes after the start of the perfusion, with an average steady-state outflow concentration of 1–1.5 μM (essentially equal to the inlet concentration). The outflow perfusate concentrations in the +ABT + RIF group were generally lower than the 4°C group and were significantly lower at 5 (P < 0.001), 10 and 20 (P < 0.05) min. The mean liver concentration at the end of the 50-minute perfusion was 12.3 μM in the 4°C group, which was significantly lower than 56.8 μM observed in the +ABT + RIF group (P < 0.001) (Fig. 2B). The total amount of the parent drug excreted in bile was 0.00012 μmol in 4°C studies. Both the hydroxy metabolites were undetectable in outflow perfusate as well as bile in 4°C studies. The liver concentration of 2-hydroxy ATV at the end of the 50-minute perfusion was 0.005 μM, whereas 4-hydroxy ATV was undetectable in liver in 4°C studies. The average bile flow was 5-fold lower than the +ABT + RIF group at 37°C. Based on these results and modeling efforts described subsequently, the +ABT + RIF group was selected as the control group for all subsequent studies.
Passive diffusion experiments. (A) ATV outflow perfusate concentrations normalized to inlet concentration and (B) ATV total liver concentrations normalized to inlet concentration, in the absence of active transport and metabolism; data are presented as mean ± S.E.M.; ATV 4°C: n = 4, ATV + ABT + RIF: n = 3; A t test was performed to compare the two groups. ***P < 0.001.
Uptake and Metabolism of ATV in the Perfused Liver.
The outflow perfusate concentrations in the +ABT + RIF group demonstrated a continued increase over 50 minutes when both uptake and metabolism were inhibited at 37°C and were generally higher than the other groups at 37°C (Fig. 3A). The average outflow concentrations were lowest in the ATV alone group and ranged between 0.001 μM at the start of the perfusion and 0.03 μM at the end of the 50-minute perfusion (Fig. 3A). Inhibition of uptake transport on RIF coperfusion significantly increased the ATV outflow perfusate concentrations compared with the ATV-alone group (P < 0.05 at 30 and 40 minutes). The outflow concentrations were significantly higher than the ATV-alone group on inhibition of metabolism by ABT during the first half of the perfusion (P < 0.01 at 0, 5, 10 minutes; Fig. 3A) and significantly lower than the ATV group at 50 minutes (P < 0.01; Fig. 3A); however, no significant difference in concentrations was seen at 30 and 40 minutes.
ATV data collected with single-pass liver perfusions. (A) ATV outflow perfusate concentrations normalized to inlet concentration; inset shows ATV only and +ABT data on a magnified scale. (B) ATV total liver concentrations normalized to inlet concentration. t Tests were performed to compare the ATV alone group with +RIF or with +ABT and to compare +ABT with +ABT + RIF; **P < 0.01. (C) Cumulative amount of ATV excreted in bile. Data are presented as mean ± S.E.M. ATV: n = 3, ATV + RIF: n = 3, ATV + ABT: n = 3, ATV + ABT + RIF: n = 3.
The ATV liver concentrations at the end of the 50-minute perfusion in various groups are shown in Fig. 3B. The mean ATV liver concentrations at the end of a 50-minute perfusion in the ATV-alone group were 108.9 μM. A decrease in ATV liver concentration was observed on treatment with RIF; however, the decrease was not statistically significant. Inhibition of metabolism resulted in a 1.3-fold increase in liver concentration. The lowest ATV liver concentration was observed in the +ABT + RIF group when both uptake and metabolism were inhibited. A statistically significant decrease (P < 0.01) was observed in the liver concentrations between the +ABT group and the +ABT + RIF group. The cumulative amount excreted in bile at the end of 15, 30, and 50 minutes is shown in Fig. 3C.
A steady increase in the amount excreted over time was observed in all groups. The amount in bile and the rate of appearance of ATV in bile were higher in all groups without RIF treatment. There was a lag in the appearance of drug in the bile in all groups. The percentage of total dose recovered in the bile as parent drug was 28.7%, 11.2%, 26.6%, and 10.7% in the ATV alone, +RIF, +ABT, and +ABT + RIF groups, respectively.
The concentrations of 2-hydroxy ATV and 4-hydroxy ATV in the outflow perfusate were below the limit of quantitation. The highest liver metabolite concentrations were observed in the ATV group (Fig. 4, A and B). RIF treatment decreased the mean liver concentration of 2-hydroxy ATV by 2.9-fold and 4-hydroxy ATV by 3.62-fold. Similarly, a 2.9- and 4.7-fold decrease was observed on inhibition of metabolism by ABT in the mean liver concentrations of 2-hydroxy ATV and 4-hydroxy ATV, respectively. The lowest metabolite liver concentrations were observed in the +ABT + RIF–treated group (Fig. 4, A and B). The cumulative amount of metabolites in bile over time is shown in Fig. 4, C and D. The ATV group demonstrated the highest amount of metabolites recovered in bile, with 2-hydroxy ATV and 4-hydroxy ATV accounting for 4.8% and 5.06% of the parent in bile, followed by the +RIF group. The amount of metabolites in bile in the +ABT group and +ABT + RIF group was lower than the other two groups but could not be compared statistically owing to high variability in the data. These data were not used for subsequent modeling efforts.
ATV metabolite data collected from single-pass liver perfusion studies. Liver concentrations normalized to ATV inlet concentration for (A) 2-hydroxy ATV and (B) 4-hydroxy ATV. One-way analysis of variance followed by post hoc Tukey’s test was performed; *Groups statistically different from ATV (P < 0.05) are marked. Cumulative amount excreted in bile of (C) 2-hydroxy ATV and (D) 4-hydroxy ATV. (E) Metabolite-to-parent ratio in liver; t tests were performed to compare each treatment group to ATV alone; *P < 0.05. Data are presented as mean ± S.E.M.; ATV: n = 3, ATV + RIF: n = 3, ATV + ABT: n = 3, ATV + ABT + RIF: n = 3.
The total metabolite-to-parent ratios calculated as the ratio of sum of the two hydroxy metabolite concentrations to the total parent concentration in the liver at the end of the 50-minute perfusion are presented in Fig. 4E. The metabolite-to-parent ratio in the +RIF group was 37.7% of that in the ATV alone group, and the difference was statistically significant (P = 0.012). Inhibition of metabolism by ABT significantly decreased the metabolite-to-parent ratio as expected, to 19.3% of that in the ATV alone group (P = 0.012). This is in agreement with the previous observations by De Montellano and coworkers that preincubation of RLM with 1 mM ABT for 30 minutes resulted in a 77% loss of P450 activity (Ortize de Montellano and Mathews, 1981). No significant difference was found in the metabolite to parent ratio between +ABT and +ABT + RIF groups.
A 1.7-fold increase was observed in the liver-to-bile ratio of ATV owing to inhibition of metabolism by ABT (Table 2). The liver-to-bile ratio of ATV in the RIF-treated group was 2-fold higher than the ATV group. The liver-to-bile ratio of metabolites was not significantly different across different groups.
Liver to bile ratio of atorvastatin (ATV) and total metabolites; data are presented as mean ± S.D.
A t test was performed to compare the ATV alone group with each of the other treatment groups.
Equilibrium Dialysis.
The unbound fraction for ATV in RLM (1 mg/ml microsomal protein) was 0.84 ± 0.074. The ATV unbound fractions in the liver based on RLM, hepatocytes, and liver homogenates corrected for dilution were 0.019 ± 0.006, 0.015 ± 0.0004, and 0.017 ± 0.004, respectively. The calculated unbound fractions in the liver calculated from RLM, liver homogenate, and hepatocyte studies were similar, suggesting that membrane binding is the major component of drug binding at organ level. ATV unbound fraction in perfusion medium was 0.99 ± 0.052. Binding of the drug to the components of the perfusion medium was minimal.
Modeling.
The average outflow perfusate concentrations from 4°C perfusions were used to calculate the passive diffusion of ATV; however, since the average steady-state concentrations were greater than the inlet concentration (1 μM), CLi could not be estimated. Therefore, subsequent CLi estimation was performed using the data from +ABT + RIF experiments.
The 5-C model fits the mean Cout data for +ABT + RIF with an R2 of 0.93 (Fig. 5A), with an estimated CLi of 0. 195 ± 0.045 liter/min. The predicted total liver concentration of 72 µM at the end of 50 minutes was obtained using this CLi estimate (Table 3), and the predicted steady-state liver concentration was 94 µM. Similarly, the predicted drug amount recovered in bile was 0.14 µmol. These predictions compared well with the observed values presented in Table 3. The predicted unbound intracellular concentration at the end of 50 minutes was 0.58 µM (Fig. 5A). Next, the 5-C model was fit to the +ABT Cout data (Fig. 5B), with the CLi fixed from the previous step, and a CLbu of 26.09 ± 6 liters/min was estimated. The predicted total liver concentration obtained using the estimated CLi and CLbu values was 137.4 µM, compared with the mean observed value of 146.5 µM (Fig. 3B). The predicted total amount of drug recovered in bile was 0.263 µmol. The predicted unbound intracellular ATV concentration in this group was 6.5 µM at the end of the 50-minute perfusion (Fig. 5B). Finally, the parameter estimate for CLm was obtained using the ATV only dataset. Because of the variability associated with the experimental data, estimation of CLm was associated with high errors (the standard error was greater than the parameter estimate and the R2 associated with the fitting procedure was 0.46). Therefore, simulations were performed using the previously optimized values for CLi and CLbu, and CLm varied between 0.001 liter/min to 0.01 liter/min. With CLm set to 0.00125 liter/min (Fig. 5C), the simulation predicted a total liver concentration and the amount of drug recovered in bile of 108.6 µM and 0.282 µmol, respectively, comparing well with the experimental mean values of 109 µM and 0.287 µmol, respectively. The predicted unbound intracellular concentration at the end of 50 minutes was 5.14 µM (Table 3). The predicted total liver concentration and amount recovered in bile in all groups were within 0.94- to 1.3 fold of the observed values.
Modeling the liver perfusion data with the five-compartment explicit membrane model. Model fitted outflow perfusate-concentration time profiles (blue) and simulated unbound intracellular concentration-time profiles (red) for (A) ATV + ABT + RIF group and (B) ATV + ABT group. Inset shows data on magnified scale. Model-fitted parameter ± standard error are listed, along with the R2 of the fit. (C) Simulated outflow perfusate concentration-time profile (blue) and cell concentration-time profile (red) for ATV group. Inset shows fitted outflow perfusate concentration-time profile, with R2 = 0.46. The CLm value was obtained from the simulation as detailed in Results. Mean experimental data are depicted as solid circles, and model-fitted or simulated results are depicted as solid lines.
Predicted and observed total atorvastatin (ATV) liver concentration, total amount of ATV recovered in bile, predicted intracellular unbound concentration at 50 minutes, and at steady state
Model Validation.
The 5-C model has been previously validated for a set of diverse compounds in a transwell experimental set up to study passive permeability and apical efflux in MDCK and multidrug resistance gene 1 - MDCK cells, respectively (Korzekwa et al., 2012; Korzekwa and Nagar, 2014; Nagar et al., 2014). In this study, the 5-C model was expanded to include perfusate and bile flow and drug concentrations and metabolic pathways. Although validation of this model will require studies with additional compounds, the resulting model was internally consistent for all data (perfusate outflow, bile amounts, and total liver concentrations).
Discussion
Experimental measurement of intracellular drug concentrations is difficult (Chu et al., 2013). The currently available methods have certain disadvantages and cannot address the complexities underlying intracellular drug concentrations (e.g., enzyme-transporter interplay). Most published models that study the effect of transporters on cell concentrations do not include membrane compartments (Sun and Pang, 2008; Ménochet et al., 2012) and hence ignore explicit membrane partitioning of the drug. Further, lag times for cell permeability and efflux out of the membrane cannot be modeled without explicit membrane compartments (Nagar et al., 2014). The goal of the present study was to predict the intracellular unbound ATV concentration in presence of transport and metabolism using the 5-C model with explicit membrane compartments and isolated perfused rat liver system.
Estimation of passive diffusion is widely performed at 4°C since the proteins relevant to drug disposition, enzymes, and transporters are functionally inactive at this temperature; however, changes in membrane fluidity at this temperature (Kandušer et al., 2008) can significantly affect the estimates for passive diffusion obtained at 4°C. In fact, there was minimal drug partitioning into the liver at 4°C (Fig. 2B). The experimental steady-state liver concentration (12.3 μM) was much lower than the predicted Kp using eq. 7 (Kp = 85.1). On the other hand, the predicted steady-state liver concentration in the +ABT + RIF–treated group (94 μM) was similar to the predicted Kp within the limits of experimental variability. We therefore suggest that estimation of passive diffusion for liver perfusion studies at 37°C with concomitant use of inhibitors of enzymes and transporters is a better approach than the use of 4°C data.
The liver perfusion experiments demonstrated that inhibition of hepatic uptake of ATV increased the outflow perfusate concentrations, and decreased the liver concentrations and amount of parent drug eliminated in the bile (Fig. 3). Conversely, inhibition of metabolism had little effect on the outflow perfusate concentrations, but it increased the liver concentration of ATV. This finding is in agreement the previous reports in which inhibition of OATP had a profound effect on the plasma concentration of OATP substrates, whereas inhibition of metabolism had a lesser impact on increasing the plasma concentration of the drug (Kantola et al., 1998; Mazzu et al., 2000; Maeda et al., 2011; Chang et al., 2014). This is because uptake is the rate determining step in the elimination of ATV from liver (Watanabe et al., 2010). These results further highlight the importance of determining the effect of transporter enzyme interplay on intracellular unbound concentrations, since inhibition of metabolism of an uptake limited drug may not result in increased parent plasma concentrations.
The decrease in liver concentrations of ATV in the presence of RIF was not significant. Moreover, the decrease in formation of metabolites in the RIF-treated group (2.9-fold for 2-hydroxy ATV and 3.62-fold for 4-hydroxy ATV) was greater than the decrease in the parent liver concentrations (1.24-fold) as illustrated by the metabolite to parent ratios (Fig. 4E). This agrees with literature reports that RIF inhibits ATV metabolism (Lau et al., 2006). Because of the confounding effect of RIF on ATV metabolism, it was not possible to model the data from +RIF studies. The decreased cumulative amount of drug excreted in the bile upon RIF treatment could be explained by inhibition of uptake into the cell and decreased intracellular accumulation.
Pretreatment with ABT did not completely inhibit ATV metabolism; however, there was a significant decrease in the liver concentration of 2-hydroxy ATV at the end of the 50-minute perfusion compared with the ATV group (Fig. 4A). The liver concentration of 4-hydroxy ATV was not significantly different than the ATV group (Fig. 4B). The metabolite-to-parent ratio in the liver decreased significantly with ABT treatment, indicating inhibition of metabolism (Fig. 4E). Similarly, the cumulative amounts of metabolites excreted in bile were considerably lower after ABT treatment (Fig. 4, C and D).
The liver-to-bile ratio of metabolites was not significantly different across different groups, indicating that efflux of the metabolites is unaffected by RIF, ABT, or a combination of both (Table 2). Inhibition of metabolism, however, significantly increased the liver/bile ratio of ATV. A similar fold increase was observed in the RIF-treated group, presumably owing to inhibition of metabolism. The increase in ATV liver-to-bile ratio upon inhibition of metabolism could be a result of saturation of efflux transport resulting from elevated intracellular ATV concentrations. Although this increase in the ATV liver-to-bile ratio is consistently observed in the +ABT group (where the liver concentrations are higher than ATV group), an increase is also observed in the +ABT + RIF or +RIF groups (where the liver concentrations are lower than the ATV group). Inhibition of apical efflux by ATV cannot by itself explain these results. This phenomenon appears to be associated with inhibition of metabolism instead of apical efflux transport saturation by ATV. Another possibility is that the concentration of an endogenous efflux transporter substrate is increased by P450 inhibition. It is possible that the apical efflux transport of ATV is inhibited by the endogenous substrate. Since P-gp has been reported to have multiple binding sites (Ruth et al., 2001; Mayer et al., 2002; Ledwitch et al., 2016), cooperative binding of endogenous substrates or metabolites could be involved.
The total liver concentrations and total amounts of drug recovered in the bile during perfusion in all groups could be successfully predicted by the model within 0.94- to 1.3-fold of the experimental value. Also, since the R2 for fitting the outflow perfusate concentrations was > 0.8, the 5-C explicit membrane model can reasonably account for the variance across the experimental groups. The predicted unbound intracellular concentrations in the +ABT + RIF group at the end of 50 minutes and at steady state are 0.58 µM and 0.734 µM, respectively, compared with a 1 µM inlet concentration. This result is in agreement with the idea that for drugs with low passive permeability, in absence of any transporter activity, the unbound intracellular concentration might be less than the plasma concentration at steady state. The modeling effort suggested that the unbound intracellular concentration at the end of 50 minutes was 11.2-fold higher in the +ABT group compared with the +ABT + RIF group. This further highlights the importance of predicting intracellular unbound concentrations since the total liver concentration in the +ABT group is only 2.6-fold higher than the +ABT + RIF group. The unbound intracellular concentration in presence of active uptake and metabolism was 5-fold higher than the inlet concentration. Hence, for uptake transporter substrates, unbound concentrations inside the cell should not be assumed to be equal to that in plasma.
Unbound intracellular concentrations cannot currently be measured directly by most available methods, including the present study. However, accurate model predictions of the experimental Cout, total liver concentrations, and cumulative biliary ATV amounts increase confidence in the prediction of unbound intracellular concentrations in this study. Organelle binding of drugs to calculate unbound intracellular concentrations has been reported for ritonavir, rosuvastatin (Pfeifer et al., 2013), and metformin (Chien et al., 2016). The complex interplay between uptake, metabolism, and efflux and its effect on unbound intracellular concentrations can be addressed with a number of approaches (e.g., sandwich cultured hepatocytes) (Pfeifer et al., 2013). Our approach uses a simple compartmental model parameterized with complex experimental data from the liver perfusion studies. It remains to be determined whether a simple 5-C model will adequately predict unbound intracellular concentrations or additional intracellular organelle compartments are needed.
For poorly permeable drugs or for uptake transporter substrates, these predicted unbound intracellular concentrations may provide better DDI as well as PD predictions compared with the use of unbound plasma concentrations. Further, this work underlines the utility of the 5-C explicit membrane model in characterizing dynamic organ-specific drug disposition. The approach used here (i.e., organ perfusion data sets along with compartmental modeling) can be applied to the study of other organs (e.g., brain, kidney, and intestine), especially when the interplay between transport and metabolism is important. Similar approaches to model dynamic organ drug disposition can be used with other whole-body PK and PD models such as PBPK, hybrid-PBPK, and PK-PD models (Gertz et al., 2014; Yang et al., 2015; Ramakrishnan et al., 2016).
In conclusion, the modeling approaches described here can accurately reproduce complex organ perfusion kinetic data. These results suggest that similar approaches can be used to model dynamic organ drug disposition in the presence of transport or metabolism. Accurate reproduction of tissue, bile, and organ outflow concentrations for a perfused organ experiment lends credibility to using these models to predict unbound intracellular concentrations.
Authorship Contributions
Participated in research design: Kulkarni, Korzekwa, Nagar.
Conducted experiments: Kulkarni.
Performed data analysis: Kulkarni, Korzekwa, Nagar.
Wrote or contributed to the writing of the manuscript: Kulkarni, Korzekwa, Nagar.
Footnotes
- Received June 6, 2016.
- Accepted July 20, 2016.
This work was supported by National Institutes of Health National Institute of General Medical Sciences [Grants R01GM104178 and R01GM114369].
Abbreviations
- ABT
- 1-aminobenzotriazole
- ATV
- atorvastatin
- CA
- CB, CC, CAM, CBM, concentration in apical, basolateral, cellular, apical membrane and basolateral membrane compartments respectively
- Cin
- inlet concentration
- Cout
- outlet concentration
- CLae
- apical efflux clearance into the apical compartment
- CLi
- CLo, diffusional clearance into and out of the membrane respectively
- CLbu
- active basolateral uptake clearance into the cell
- 5-C
- five-compartment
- DDI
- drug-drug interactions
- HMG CoA
- hydroxymethylglutaryl-CoA
- Kp
- partition coefficient
- LC-MS/MS
- liquid chromatography tandem mass spectrometry
- MDCK
- Madin-Darby canine kidney
- OATPs
- organic anion transporting polypeptides
- P450
- cytochrome P450
- PBPK
- physiologically based pharmacokinetic
- PD
- pharmacodynamics
- P-gp
- P-glycoprotein
- PK
- pharmacokinetics
- Q
- perfusion flow rate
- Qbile
- biliary flow rate
- RIF
- rifampin
- ABT
- 1-aminobenzotriazole
- RLM
- rat liver microsome
- VA
- VB, VC, VAM, VBM, volumes of apical, basolateral, cellular, apical membrane and basolateral membrane compartments, respectively
- Copyright © 2016 by The American Society for Pharmacology and Experimental Therapeutics
References
In this issue
Intracellular Unbound Atorvastatin Concentrations
