Abstract
Passive permeability and active efflux are parallel processes in transcellular flux. Therefore, the observed kinetics of a transporter substrate depends on both of these factors. The transporter expression has been shown to affect both the apparent Km and Vmax values. Kinetic parameters can be obtained from various experimental settings, but these do not necessarily reflect the situation in transcellular flux. Kinetic absorption models need reliable estimates of saturable kinetics when accurate in silico predictions are to be made. The effect of increasing P-glycoprotein expression on apparent transport kinetics was studied using quinidine and digoxin as model compounds. The intracellular concentrations of drugs during the transport process were also measured. A dynamic simulation model was constructed to study the observed data. The apparent Km and Vmax values increased as the P-glycoprotein expression increased. Simulations reproduced the shift in both kinetic parameters as a function of efflux pump expression. In addition, the apparent Km value showed a strong inverse relationship to the passive permeability. In contrast, the apparent Vmax value reached a maximum at intermediate passive permeability and declined above and below this passive permeability. The true Vmax and Km values were never reached. The shift in Km was assigned to a decrease in intracellular concentration at the P-glycoprotein interaction site with both experimental and simulation data. In conclusion, the apparent kinetic parameters in transcellular permeability assays depend on passive permeability and efflux pump activity. Therefore, parameters that are obtained from in vitro assays should be cautiously applied to in vivo predictions.
The determinants of oral drug bioavailability are of major interest in drug research. Absorption, distribution, metabolism, and excretion have been intensively studied, and in vitro-in vivo correlation has become an important issue in these studies. There are several applications for prediction of intestinal drug absorption from in vitro data. The compartmental absorption and transit model (Yu and Amidon, 1999) and other physiologically based computer models (Willmann et al., 2004; Usansky and Sinko, 2005) simulate drug absorption using estimates of permeability and solubility properties of drug molecules and physiological parameters of the gastrointestinal tract. The present models usually work well for passively absorbed drugs, but laborious in vivo experiments are required for accurate simulations on transporter substrates (Kwon et al., 2004).
There are several methods available for creating input parameters for these simulation models. Passive permeability can be measured relatively easily with parallel artificial membrane permeability assay or cell cultures (Hidalgo, 2001) or predicted from the molecular properties of the chemical (Bergström et al., 2003). Kinetic parameters of a transporter, e.g., P-glycoprotein (P-gp), may be evaluated based on binding to the protein (Döppenschmitt et al., 1999), ATP hydrolysis in microsomal preparations (Boulton et al., 2002), inhibition studies (Gao et al., 2001), or permeability experiments with suitable cell monolayers (Horie et al., 2003; Troutman and Thakker, 2003; Balakrishnan et al., 2007). Because the methods usually give different results, the question of which these parameters should be used in a simulation model is raised. To use the kinetic parameters from studies with isolated proteins in a simulation model, one would need a realistic estimate of the concentration at the intracellular interaction site. This is not an easy task because many processes are involved in the intracellular kinetics of drug molecules, as seen, e.g., for paclitaxel (Kuh et al., 2000). Monolayer studies combine the expression level of the transporter of interest and the passive transport, meaning that the kinetic values obtained vary, depending on the cell line. Furthermore, the choice of experimental parameters such as pH conditions and stirring rate influence the passive permeability of drugs and, thus, different kinetic parameters can be obtained despite the use of identical cell lines.
Some previous works described the interesting phenomenon that apparent Km and/or Vmax values in a transport experiment increase when the transporter expression increases. This result has been observed for both influx (Irie et al., 2006; Balakrishnan et al., 2007) and efflux (Zhang and Benet, 1998; Soldner et al., 2000; Horie et al., 2003). Recent computational analyses suggest that transport assays cannot give accurate estimates of the molecular Michaelis-Menten parameters (Bentz et al., 2005; Balakrishnan et al., 2007). A very recent theoretical work using a basic three-compartment transport model suggested that the apparent kinetics of efflux pump substrates are direction-dependent and are also affected by both passive permeability and efflux pump expression (Kalvass and Pollack, 2007). Therefore, there is a need to find useful relationships for kinetic values obtained from in vitro assays and their significance in observed permeability.
P-gp is of considerable interest because it affects the absorption of several drugs. However, it is not known how the expression level of P-gp affects the apparent Michaelis-Menten constants in a permeability assay. We have previously described P-gp expression levels in three cell lines, Caco-2/WT < Caco-2/hPXR < MDCK-MDR1 (Korjamo et al., 2006). In this report, the changes in the apparent kinetic parameters of the known P-gp substrates quinidine and digoxin were studied with these cell lines expressing various amounts of P-gp. Direct experimental evidence showing that the changes in the intracellular drug concentrations lead to a shift in the apparent kinetic parameters was obtained. The effect of passive permeability on the apparent transport kinetics was investigated with a novel simulation model.
Materials and Methods
Materials. Digoxin and quinidine were from Sigma-Aldrich (St. Louis, MO). [3H]Digoxin and [3H]mannitol were from PerkinElmer Life and Analytical Sciences (Boston, MA). [3H]Quinidine was from American Radiolabeled Chemicals (St. Louis, MO). The P-gp inhibitor GF120918 (Hyafil et al., 1993) was a kind gift from GlaxoSmithKline (Research Triangle Park, NC).
Cell Cultures and Filters. Caco-2/WT (HTB-37; American Type Culture Collection, Manassas, VA) and Caco-2/hPXR (Korjamo et al., 2005) cell lines were cultured as described previously (Korjamo et al., 2005).
MDCKII-MDR1 cells were purchased from The Netherlands Cancer Institute (Amsterdam, The Netherlands). Growth medium was Dulbecco's modified Eagle's medium (Gibco 61965-026; Invitrogen, Carlsbad, CA) with 10% heat-inactivated fetal bovine serum and 100 U/ml penicillin and 100 μg/ml streptomycin. The cells were grown in 75-cm3 flasks at +37°C in a 95% air/5% CO2 atmosphere and subcultured twice a week.
For permeability experiments, 82,000 cells/cm2 (Caco-2/WT and Caco-2/hPXR) or 390,000 cells/cm2 (MDCKII-MDR1) were seeded onto uncoated polycarbonate filters (pore size 0.4 μm, area 1.1 cm2; Corning Life Sciences, Corning, NY). Cell passage numbers were 37 to 46, 25 to 31, and 29 to 46, respectively. For both Caco-2 cell lines, the medium was changed three times per week and always on the day before the experiment. The monolayer was used for the experiments on the 21st day. For MDCKII-MDR1 cells, the growth medium was changed daily and the monolayer experiments were performed on the 4th day. Standard growth medium was used for all the cell lines.
Permeability Experiments. Apparent absorptive and secretory permeability coefficients (Papp) were determined for P-glycoprotein substrates digoxin and quinidine at various donor concentrations using radiolabeled compound (1 μl/ml) in the donor chamber as a marker. The permeability experiments were performed in both apical-to-basolateral (AB) and basolateral-to-apical (BA) directions as described previously (Korjamo et al., 2005). Passive permeability (Ppass) was studied at each drug concentration with assay buffer containing the P-glycoprotein inhibitor, GF120918 (2 μM). The stirring rate was 450 rpm. The experiments were conducted for up to 45 min (quinidine) or 90 min (digoxin). All assays were done in triplicate. Transepithelial electrical resistance values were measured at the beginning and at the end of the assay, and the filters with too low values [<300 Ω·cm2 (Caco-2/WT), <200 Ω·cm2 (Caco-2/hPXR), and <150Ω·cm2 (MDCKII-MDR1)] were discarded. Additionally, [3H]mannitol was used as a control substance for the monolayer integrity in all the assays. Mass balance was calculated as (amount in donor at last time point + amount in receiver at last time point)/amount initially added × 100%.
Unstirred Water Layer. We studied the effect of the unstirred water layer (UWL) on quinidine permeability with Caco-2/WT cells at a 0.3 μM concentration using various stirring rates (250-450 rpm). The assay was performed in both directions with and without 5 μM GF120918 at each stirring rate.
Cellular Retention. After the kinetic parameters were determined, the cellular accumulation of drugs during a permeability experiment was determined at two concentrations (below and above the observed Km value). The concentrations were quinidine Caco/WT 1 and 10 μM, Caco/hPXR and MDCK-MDR1 10 and 100 μM, and digoxin 50 μM (only because of solubility problems). Altogether 12 inserts were prepared, and the permeability experiment was started as described above. At time points 5, 15, 30, and 60 min, three inserts were sampled on both chambers. These same inserts were washed from both sides with ice-cold Hanks' balanced salt solution buffered with 25 mM Hepes. Thereafter, 300 μl of Hanks' balanced salt solution-Hepes containing 1% Triton X-100 was added, and the monolayers were lysed on an orbital shaker for 20 to 30 min. A 100-μl sample was withdrawn to liquid scintillation counting.
Sample Detection and Result Analysis. Radioactivity of the assay samples was measured with a MicroBeta liquid scintillation counter (PerkinElmer Wallac, Turku, Finland) using OptiPhase HiSafe scintillation cocktail (PerkinElmer Wallac). Apparent permeability coefficients under sink conditions were calculated for each drug concentration according to the equation
where J is flux across the cell monolayer (nanomoles per second), A is area of the filter (square centimeters), and C0 is the initial donor concentration (micromolar). Each drug concentration was plotted against the flux (J), and Michaelis-Menten kinetic parameter values Vmax and Km were calculated from these curves according to the equation
where Ppass is the passive permeability in the presence of P-gp inhibitor GF120918 (centimeters per second), A is the area of the filter (square centimeters), and C0 is the initial donor concentration (micromolar). GraphPad Prism software (GraphPad Software Inc., San Diego, CA) was applied for curve fitting. Weighting of 1/concentration was used in the fitting procedure.
Simulations. A novel three-compartment model (Fig. 1) was constructed with Stella 8.1.4 software (isee systems, Lebanon, NH) to simulate mass flow in a permeability experiment. Three-compartment models have been used to simulate in vitro permeability experiments (Ito et al., 1999; González-Alvarez et al., 2005). We introduce an apparent cellular distribution coefficient (K = Mbound/Mfree) to the model. Mfree represents the drug amount in cytosol that is ready to diffuse across cell membranes, and Mbound is the amount of drug that is bound to intracellular structures, which leads to high cellular drug accumulation. This approach can accurately simulate the time course of intracellular accumulation of drug during a permeability experiment (unpublished results). The flux across the basolateral membrane was assumed to be purely passive. In addition to the passive component, an efflux transporter obeying regular Michaelis-Menten kinetics affected the flux across the apical membrane. A virtual efflux pump substrate with high cellular retention (K = 200, corresponding to a fitted value for quinidine, unpublished results) was created. The Km value was fixed to 5 μM. The volumes of the compartments were 0.5 ml (apical), 1.5 ml (basolateral), and 0.0022 ml (cellular assuming cell height of 20 μm). The virtual compounds accumulated into cells according to apparent cellular distribution coefficient. The efflux pump used the “free” cellular concentration as its substrate. The true Vmax and membrane permeabilities were varied 26-fold (80-2100 fmol/s) and 5-fold (50-250 × 10-6 cm/s), respectively. Each simulation was run at eight initial concentrations (0.1-300 μM). For each parameter combination, the simulation with Vmax = 0 was considered as an efflux pump-inhibited experiment. For each simulation, the highest flux (1-min intervals) to the receiver chamber was searched. The initial applied donor concentration was used for calculations. The passive flux (true Vmax = 0) was subtracted from the total flux in the presence of efflux activity. This “active flux” data were then fitted to regular Michaelis-Menten equation
where A is the area of the filter (square centimeters) and C0 is the initial donor concentration (micromolar). GraphPad Prism software was applied for curve fitting. Weighting could be omitted because simulation data lack experimental error.
Statistical Testing. The difference in the intracellular accumulation between P-gp inhibited and uninhibited wells at every time point was tested with Student's t test. The analysis was performed with GraphPad Prism software.
Results
Permeability. The passive permeability of quinidine was unpolarized and of similar magnitude in all cell lines (Table 1). The mass balance of quinidine was 75 to 80% in the AB direction and 87 to 90% in the BA direction. The Ppass of digoxin was moderately lower in the MDCK-MDR1 cells than in the Caco-2 cell lines, and the mass balance was practically 100%. P-gp-inhibited digoxin permeability decreased at concentrations 200 μM in all cell lines in both the AB and BA directions (data not shown), and permeability values obtained at higher concentrations were not used for curve-fitting.
The unstirred water layer forms a considerable permeability barrier for compounds that traverse the cell monolayer very rapidly (Karlsson and Artursson, 1991). Therefore, we determined the Papp values at a low quinidine concentration (0.3 μM) in both directions at various stirring rates. The passive permeability in both directions increased as the stirring became more vigorous (Fig. 2). The uninhibited transport, however, was increased only in the BA direction. Thus, the apparent P-gp-mediated transport (Papp - Ppass) was relatively constant in the BA direction but increased considerably in the AB direction at higher stirring rates. Therefore, the kinetic studies were conducted at the highest stirring rate (450 rpm). The apparent permeability of digoxin did not change at different stirring speeds (data not shown).
Cellular Retention. P-glycoprotein expression affected intracellular drug concentrations. The P-gp inhibitor GF120918 increased the cellular retention of quinidine in Caco/WT at early time points at 1 μM concentration (Fig. 3a). In contrast, no difference in the cellular quinidine amount between inhibited and uninhibited cells was observed when the concentration was increased well above the apparent Km values (10 μM). Caco/hPXR and MDCK-MDR1 cells have sufficient P-gp expression to maintain decreased cellular quinidine levels at 10 μM (Fig. 3, b and c). However, a considerable increase (to 100 μM) in donor concentration removes the difference in cellular retention between inhibited and uninhibited wells. The results were similar in the BA direction even though the time courses of accumulation were somewhat different (data not shown). The overall mass balance (apical + basolateral + cellular) remained constant after initial drop of 3 to 9%, which is probably due to binding to the apparatus (Palmgrén et al., 2006) or small pipetting errors. Similar observations on cellular accumulation were made in the simulation experiments when the efflux pump expression level was altered (Fig. 4).
Cellular retention of digoxin was clearly lower than that of quinidine (Fig. 5), and the mass balance was nearly perfect. P-gp could reduce the cellular accumulation to a similar extent in the two Caco-2 cell lines. However, MDCK-MDR1 cells retained much less digoxin than the Caco-2 cells, and P-gp could only modestly decrease its accumulation.
Experimental Kinetic Parameters. The apparent kinetic transport parameters are presented in Table 1. The Km and Vmax values increased in the order Caco/WT < Caco/hPXR < MDCK-MDR1 with both substrates. However, parameters obtained for digoxin from MDCK-MDR1 cells should be treated with caution because the Km value is 5-fold higher than the highest experimental concentration. A practically linear relationship was obtained when Km and Vmax values of quinidine in the AB direction are plotted against each other (Fig. 6). A similar but not equally linear relationship was seen for both drug molecules in the BA direction. Some literature data are also presented for comparison in these figures.
The total P-gp levels in the cell lines were obtained from the background corrected optical density values of the Western blot figure published recently (Korjamo et al., 2006). Table 2 presents the protein levels and kinetic parameters in relation to Caco/WT cells. The kinetic parameters of quinidine in the AB direction correlate almost perfectly to the protein data. Other apparent kinetic parameters of Caco/hPXR cells change less than expected from the protein data. The Km values from MDCK-MDR1 cells show consistency with protein data, but there is more variation in Vmax.
Simulations. The apparent Km and Vmax values both increased as the efflux pump expression increased. This shift in the apparent Km value increased as the passive permeability decreased (Fig. 7a). At low passive permeability, the apparent Vmax was low (Fig. 7b). When the permeability increased to medium levels, the apparent Vmax reached its peak value. Finally, apparent Vmax decreased again at high passive permeability. Similar to the experimental observations, an increase in efflux pump expression decreased intracellular drug concentration (Fig. 4). The apparent Km was always considerably higher in the AB than in the BA direction (Table 3).
Discussion
The apparent transport kinetics of both influx and efflux transporter substrates depend on the access of the substrate to the interaction site. The increase in transporter expression leads to biased apparent kinetic parameters in monolayer transport assays. These observed shifts have been hypothesized to result either from a decreased concentration in the UWL next to the plasma membrane (Winne, 1977; Balakrishnan et al., 2007) or from a decreased intracellular concentration (Bentz et al., 2005). The previous works have based their hypotheses mainly on computational data. This study is the first that shows direct experimental evidence of how the increase in efflux pump expression decreases the intracellular drug amount, which then leads to larger shifts in the apparent kinetic parameters. Our novel simulation model can reproduce the time course of intracellular accumulation. Simulations show that the passive membrane permeability altered the bias in kinetic parameters. The shift in the apparent Km value is inversely proportional to the passive permeability. In contrast, the apparent Vmax first increases and then decreases as the passive permeability increases.
Digoxin permeabilities up to 500 μM has been reported (Troutman and Thakker, 2003). However, we detected that the Ppass decreased above 200 μM (data not shown), probably due to precipitation of the drug or saturation of some uptake transporter. Indeed, organic anion transporting polypeptide E has been shown to transport digoxin (Mikkaichi et al., 2004). However, its expression is low both in Caco-2 (Hilgendorf et al., 2007) and MDCK cells (Goh et al., 2002). UWL affects the kinetic parameters of absorptive transporters (Winne, 1977; Balakrishnan et al., 2007). It has been hypothesized but not experimentally proven that the apparent Km value increases because the concentration next to plasma membrane decreases as the UWL thickness increases. In the present quinidine experiments, the UWL affected the Papp in the BA direction and the Ppass in both directions (Fig. 2), but the Papp in the AB direction remained fairly constant. Therefore, the UWL influences the kinetic parameters also for efflux transporter substrates. The kinetic parameters in the AB direction may be more sensitive to changes in the stirring rate than those in the BA direction because the P-gp effect (Papp - Ppass) changes considerably in the AB but not in the BA direction. Digoxin permeability and, consequently, the kinetic parameters are fairly insensitive to the stirring rate because its transport is not rate-limited by the UWL.
The apparent kinetic values (Table 1) show that both Km and Vmax increase as the P-gp activity increases and that these values are strongly correlated (Fig. 6). Moreover, there is often a good correlation between the kinetic parameters and the expression of P-gp (Table 2). A plausible explanation for this is that P-gp reduces the drug concentration at the interaction site below saturation concentrations. Therefore, the donor concentration does not reflect the concentration at the P-gp interaction site. Indeed, the total cellular amounts of quinidine (Fig. 3) and digoxin (Fig. 5) are smaller in the absence than in the presence of the P-gp inhibitor when the donor concentration does not saturate P-gp. In addition, the simulation model reproduced the experimental data for quinidine (Fig. 4). Together these data confirm the hypothesis that the reduced intracellular concentration causes the shift in observed kinetic parameters in monolayer permeability assays. The actual concentration at the interaction site may be even more reduced if the efflux pump is able to create a concentration gradient inside the cell.
Recent results from a simulation model suggest that apparent Vmax of P-gp from permeability experiments follows the molecular Vmax and, thus, the protein expression level reasonably well (Bentz et al., 2005). However, the apparent Km value is always expected to be higher than the molecular Km, and it was shown to depend on the passive permeability and the maximal pumping capacity of P-gp. Supporting this result, our simulations show that the decrease in passive permeability increases the observed Km shift when efflux pump expression increases (Fig. 7a). In contrast, the apparent Vmax value at a particular transporter expression level increases up to a certain passive permeability and starts to decline thereafter (Fig. 7b). The apparent Vmax never reached the true Vmax but only attained <50% of this value. The decrease in apparent Vmax at low passive permeability may be caused by the combination of slow entrance and high accumulation into cellular structures. On the other hand, the high passive permeability masks the active transport and the apparent Vmax decreases. Thus, the parallel active and passive transport processes prevent obtaining of true kinetic values of transporters, at least with simple approaches (eqs. 2 and 3) that are used to treat such permeability data. Recently a new kinetic approach for calculating the parameters in transport experiments has been suggested (Kalvass and Pollack, 2007). Nevertheless, the present results give an idea of how different parameters affect the apparent transport kinetics viewed from the donor solution. Actually, the UWL and the plasma membrane both form a diffusion barrier to the interaction site, and the Km shift is expected to be more pronounced for efflux than for influx transporters with an extracellular interaction site (Balakrishnan et al., 2007).
The relationship between P-gp expression and kinetic parameters is not simple, and it can vary, depending on the substrate and transport direction (Table 2). A previous study reported that the apparent Km value is always higher in the AB than in the BA direction (Troutman and Thakker, 2003). This observation was explained by rigidity of the apical membrane, which results in lower permeability than at the basolateral membrane. In this study, digoxin behaved similarly, because the Km value in the AB direction is much higher than 200 μM (Table 1). The fact that our simulations with equal membrane permeabilities gave higher Km values in the AB than in the BA direction (Table 3) suggests that the observed difference may partly be an artifact caused by the Transwell apparatus. Because the donor and receiver volumes are unequal and small, the donor concentration decreases more rapidly in the AB than in the BA direction. Thus, the cellular concentrations remain lower in the AB than in the BA direction at any given donor concentrations, and the Km shift increases. Moreover, the donor concentration is considerably below the initial concentration at steady state in which the maximal flux is taken. The directional difference in the apparent Km value was practically abolished when both apical and basolateral chambers were increased to 5000 ml in simulations (data not shown). Now the sink conditions could be maintained, and the donor concentration remained at initial level. The present experimental data suggest that the apparent Km value may also be higher in the secretory than in the absorptive direction (quinidine) (Table 1). This result requires that the resistance at the apical membrane be considerably lower than that at the basolateral membrane. Our hypothesis is that the applied vigorous stirring decreases the total passive resistance (UWL + cell membrane) at the apical membrane to a lower value than at the basolateral membrane. This hypothesis needs further experimental evidence.
In conclusion, The apparent kinetic parameters shift when a diffusion barrier is placed between the donor and the interaction site. The access to the interaction site may be very different in simple laboratory experiments and in whole organisms. Therefore, scaling in vitro parameters to permeability at different locations in vivo is not straightforward. Further studies are needed to understand the relationships between permeability barriers and observed kinetics. Ultimately, a predictive permeability model may be constructed using simple in vitro parameters reflecting the passive permeability, intracellular distribution, and transporter affinity.
Acknowledgments
Technical assistance from Paula Nyyssönen and Markku Taskinen is gratefully acknowledged.
Footnotes
-
This study was financially supported by the Finnish Funding Agency for Technology and Innovation.
-
doi:10.1124/dmd.107.016014.
-
ABBREVIATIONS: P-gp, P-glycoprotein; WT, wild type; hPXR, human pregnane X receptor; 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; AB, apical-to-basolateral; BA, basolateral-to-apical; UWL, unstirred water layer.
- Received April 3, 2007.
- Accepted May 31, 2007.
- The American Society for Pharmacology and Experimental Therapeutics