P-glycoprotein (P-gp¹) efflux and cytochrome P450 3A (CYP3A)-mediated intestinal metabolism have been implicated as potential biochemical barriers to the intestinal permeation of their substrates (Watkins, 1997; Benet and Cummins, 2001). Subsequent findings have led to the further hypothesis that P-gp and CYP3A metabolism are functionally linked and may act in concert to limit the passage of intact xenobiotics (including drugs) across the enterocyte. Evidence to support this hypothesis includes, first, the fact that P-gp and CYP3A share a close cellular localization within the enterocyte, and that this spatial relationship may benefit the detoxifying properties of the two proteins (Kolars et al., 1992; Watkins, 1997; Cummins et al., 2001). Second, there is an overwhelming overlap in substrate specificity for P-gp and CYP3A, with the two proteins sharing common inhibitors, inducers, and regulators (Kivistö et al., 1995; Wacher et al., 1995; Schuetz et al., 1996a). Finally, the extent of P-gp activity or expression has a significant influence over CYP3A expression levels (Schuetz et al., 1996b; Schuetz et al., 2000), which is supported by the recent discovery that P-gp and CYP3A expression may be coordinately regulated by ligand binding to nuclear receptors such as the steroid and xenobiotic receptor, SXR (Synold et al., 2001). Therefore, enterocyte-based P-gp and CYP3A appear to form a coordinately regulated alliance to reduce the intestinal permeation and, therefore, the oral absorption of a number of xenobiotics and drugs.

Enterocyte-based CYP3A metabolism has been shown to reduce the intestinal permeability of a number of drugs, including verapamil, tacrolimus, and cyclosporin (Kolars et al., 1991; Fromm et al., 1996; Hashimoto et al., 1998). These compounds are all P-gp substrates, and, therefore, permeability across the enterocyte may also be limited by P-gp-mediated efflux or antitransport. In addition to the individual effects of CYP3A and P-gp on drug permeability, it has further been suggested that P-gp-mediated efflux increases drug exposure to CYP3A within the enterocyte through repeated cycles of efflux and subsequent reabsorption, resulting in increased intestinal residence time and extraction (Benet and Cummins, 2001; Johnson et al., 2001; Cummins et al., 2002).

In an attempt to rationalize the interaction between P-gp and CYP3A within the enterocyte, a diffusion-based simulation model was recently proposed (Ito et al., 1999). In these studies, whereas the fraction of drug absorbed from the lumen into the portal blood was shown to be both P-gp- and CYP3A-dependent, the model did not predict a change to the intestinal extraction ratio in response to a change in P-gp efflux; instead, P-gp efflux reduced only the rate of drug entering the enterocyte. This reduced rate of drug entry subsequently reduced the overall extent of drug absorption due to incomplete gastrointestinal absorption and fecal elimination. In contrast to this theoretical model approach, studies in Caco-2 cells have demonstrated that P-gp can modulate the extent of drug metabolism across the enterocyte (Gan et al., 1996; Hochman et al., 2000; Cummins et al., 2002); however, little work has been performed utilizing either animal or human intestinal tissues, where both metabolic enzymes and efflux transporters are expected to be expressed at physiological levels (for the particular species under examination).

Therefore, to further probe the impact of P-gp on intestinal extraction, the current studies have characterized the bidirectional uptake, transport, and metabolism of the P-gp and CYP3A substrate verapamil, in an in vitro model of intestinal permeability utilizing excised rat jejunal tissue. The relative impact of P-gp and/or CYP3A on verapamil permeability was isolated by the use of selective and nonselective inhibitors of P-gp efflux and CYP3A metabolism. The development of a four-compartment kinetic model describing verapamil transport and metabolism across the intestinal mucosa also facilitated further simulation-based examination of the complex interplay between P-gp and CYP3A. These simulations suggest that although the intestinal extraction ratio of verapamil can be reduced by P-gp inhibition, for this to occur, the increase in drug uptake resulting from P-gp inhibition must result in disproportionate increases in the extent of drug transport relative to the extent of drug metabolism.

Materials and Methods

Materials. Verapamil (5-[(3,4-dimethoxyphenethyl)methylamino]-2-(3,4-dimethoxyphenyl)-2-isopropylvaleronitrile) HCl and norverapamil (5-[(3,4-dimethoxyphenethyl)amino]-2-(3,4-dimethoxyphenyl)-2-isopropylvaleronitrile) HCl were supplied by Sigma-Aldrich (St. Louis, MO) and Alltech Associates (Baulkham Hills, NSW, Australia), respectively. D-617 (2-[3,4-dimethoxyphenyl]-5-methylamino-2-isopropylvaleronitrile), d-620 (5-amino-2-[3,4-dimethoxyphenyl]-2-isopropylvaleronitrile, d-703 (5-[(3,4-dimethoxyphenethyl)-methylamino]-2-(4-hydroxy-3-methoxyphenyl)-2-isopropylvaleronitrile), and the internal standard d-519 (5-[(3,4-dimethoxy phenethyl)methylamino]-2-(3,4-dimethoxyphenyl)-valeronitrile) were a kind gift from Knoll AG (Ludwigshafen, Germany). Ketoconazole was obtained from ICN Pharmaceuticals Biochemicals Division (Aurora, OH), PSC833 was a kind gift from Novartis (Basel, Switzerland), and midazolam was used in the form of Hypnovel injection [5 mg of midazolam (as HCl), 5 mg of NaCl, 1 mg of HCl, NaOH to pH 3.3, water for injection to 1 ml; Roche Diagnostics, Dee Why, NSW, Australia]. All other reagents were analytical grade, all solvents were HPLC grade, and water was obtained from a Milli-Q water purification system (Millipore Corporation, Bedford, MA).

Tissue Preparation. All animal studies were performed in accordance with the guidelines of the Australian and New Zealand Council for the Care of Animals in Research and Teaching, and the study protocol was approved by the institutional animal experimentation ethics committee. The experimental procedure was based on methods described previously (Johnson et al., 2002). Briefly, male Sprague-Dawley rats were anesthetized, and a segment of jejunum was removed, rapidly stripped of its serosal muscle layers, and mounted in side-by-side diffusion chambers (Navicyte Inc., Sparks, NV). Modified Krebs' bicarbonate-Ringer's (KBR) buffer (7 ml) was immediately added to both the mucosal and serosal chambers (Ungell et al., 1992). The modified KBR contained 147.2 mM Na⁺, 5.1 mM K⁺, 1.25 mM Ca²⁺, 1.2 mM Mg²⁺, 115.2 mM Cl^-, 15 mM HCO₃^-, 0.4 mM H₂PO₄^-, 1.8 mM HPO₄^2-, 1.2 mM SO₄^2-, 11.5 mM d-glucose, 4.9 mM l-glutamate^-, 4.9 mM pyruvate^-, and 5.4 mM fumarate^2-. Tissues were allowed to equilibrate for 30 min before the commencement of permeability experiments, when verapamil was spiked into the donor chamber to give a final concentration of 20 μM. During inhibition studies with the P-gp inhibitor PSC833 (20 μM), the CYP3A inhibitor midazolam (20 μM), or the P-gp/CYP3A inhibitor ketoconazole (50 μM), the inhibitors were added to both the mucosal and serosal chambers 15 min before the addition of verapamil. Potent inhibition of the P-gp-mediated transport of digoxin by PSC833 and ketoconazole has previously been demonstrated in rat jejunum in vitro (Johnson et al., 2002).

Study Protocol. Verapamil (20 μM) transport studies were conducted in both the mucosal to serosal (m to s) and the serosal to mucosal (s to m) directions in the presence and absence of the inhibitors described above. This concentration of verapamil is below the K_m of both P-gp-mediated transport and CYP3A-mediated metabolism of verapamil. It was therefore assumed that neither process was saturated during these studies (Makhey et al., 1998; Tracy et al., 1999). Although verapamil and verapamil analogs have been shown to inhibit certain organic cation transporters and the multidrug resistance-associated protein 1, respectively, the current literature suggests that the impact of these transporters on the intestinal permeability of verapamil is of less significance than that of P-gp and is therefore unlikely to have a significant impact on the interpretation of data collected in the current studies (Saitoh and Aungst 1995; Takano et al., 1998; Loe et al., 2000; Dresser et al., 2001). After an initial (t = 0) 100-μl donor chamber sample, additional donor (100 μl) and receptor (200 μl) chamber samples were taken every 30 min for 180 min. Receptor chamber samples were replaced with an equal volume of fresh buffer (including inhibitor if appropriate). Donor chamber samples were not replaced and were diluted with 100 μl of fresh buffer to bring the verapamil concentration into the linear range of the assay. All samples were acidified with 50 μl of 90% methanol in 0.05 M H₂SO₄ to prevent precipitation or adsorption of un-ionized verapamil or its metabolites during analysis. At the conclusion of transport studies (180 min), the tissue segment was removed from the diffusion chamber, cut into small pieces, and frozen for later analysis. Additional studies using 200 μM verapamil in the donor chamber were also conducted, and samples were appropriately diluted for use in the following assay.

Verapamil and Metabolite Assays. Diffusion chamber samples were analyzed for verapamil, and metabolites (norverapamil, d-617, d-620, and d-703) were analyzed by gradient HPLC based on a previously described method (von Richter et al., 2000). The HPLC consisted of a Waters 712 WISP autosampler, 610 Fluid Unit and Valve Station, and 600 Controller (Waters, Milford, MA) and a PerkinElmer LC 240 fluorescence detector (PerkinElmer, Beaconsfield, UK) with excitation and emission wavelengths set to 280 and 310 nm, respectively; and data were collected on a Shimadzu C-R5A integrator (Shimadzu, Kyoto, Japan). A 100-μl portion of acidified sample was injected onto an Xterra MS C₁₈ 5 μM 3.9 × 150 mm column (Waters), and mobile phase was delivered at 1 ml · min^-1. Mobile phase A consisted of 90% 5 mM ammonium acetate, pH 4.2, and 10% acetonitrile, and mobile phase B was 100% acetonitrile. The initial proportion of mobile phase B was 13%, which was increased to 30% over 12 min, held at 30% for 3 min, increased to 32% over an additional 2 min, returned to 13% over 1 min, and finally held at 13% to facilitate an 8-min re-equilibration period. The retention times of verapamil, norverapamil, d-617, d-620, and d-703 were 19.7, 19.0, 6.6, 6.0, and 10.0 min, respectively. The assay was validated using quadruplicate quality control (QC) standards prepared in KBR on four separate occasions, at high (10 μM), medium (1 μM), and low (0.1 μM) concentrations for verapamil, and medium (1 μM) and low (0.1 μM) concentrations for the metabolites. Accuracy (of all analytes) was within 3% of target at high and medium concentrations and within 10% of target at low concentrations. Precision, expressed as the %CV, was less than 5% for the high and medium QC standards, and less than 10% for the low QC standards. Interday accuracy and precision varied by less than 10%. The LOQ, defined as the minimum quantifiable concentration with appropriate accuracy and precision (within 15% of target and CV <15%) was 0.01 μM for verapamil and norverapamil, and 0.05 μM for d-617, d-620, and d-703.

Tissue samples were analyzed for verapamil and metabolites using a modification of a previously described extraction method (Hendrikse et al., 1998). Frozen samples were rapidly thawed, and 1 ml of 0.05 M NaOH and 50 μl of internal standard solution (20 μM d-519) were added and briefly vortexed. Tissues were subsequently homogenized for 2 min (Polytron PT 1200; Kinematica, Basel, Switzerland) before the addition of 1 ml of methyl t-butyl ether (EM Scientific, Gibbstown, NJ). The samples were vortexed for 1 min and then centrifuged at 2500g for 10 min. The supernatant was transferred using a glass pipette to a clean 1.5-ml tube containing 150 μl of 0.5 M H₂SO₄. After vortexing and centrifugation (5000g for 10 s), 100 μl of the aqueous portion was removed and diluted with an equal volume of Milli-Q water, prior to injection (100 μl) onto the HPLC. Extracted tissue samples were assayed using the chromatographic conditions described above against a standard curve constructed using blank tissue homogenate spiked with known amounts of verapamil and metabolites, and 50 μl of internal standard solution. The retention time of the internal standard (d-519) was 7.9 min. The tissue homogenization assay was validated utilizing QC standards spiked into blank tissue homogenate at high, medium, and low levels for verapamil (200, 100, and 25 nmol, respectively) and metabolites (4, 2, and 0.5 nmol, respectively). Accuracy (of all analytes) was within 10% of target for the high- and medium-level QC standards, and within 20% of target for the low-level QC. Precision (expressed as %CV) was no more than 15% for all analytes at each QC level. The efficiency of the extraction procedure for each compound was similar over all concentration ranges, and the mean extraction efficiency (recovery) of verapamil, norverapamil, d-617, d-620, and d-703 was 88, 91, 76, 69, and 88%, respectively. The LOQ, defined as the minimum quantifiable amount with appropriate accuracy and precision (within 20% of target and CV <20%) was 0.05 nmol per tissue segment for verapamil and norverapamil, and 0.20 nmol per tissue segment for d-617, d-620, and d-703.

No significant adsorption of verapamil to the diffusion chamber was observed over the 3-h period of the permeability experiments. Mass balance of verapamil and its metabolites across all experiments was 94 ± 6% (mean ± S.D., n = 52). Verapamil was stable in the diffusion chambers at 37°C for >3 h when spiked into fresh KBR or into KBR collected after incubation with jejunal segments for 2 h, indicating that no significant degradation or metabolism of verapamil occurred in the buffer chambers.

Data Analysis. The only metabolites of verapamil detected in these studies were norverapamil and d-617. Both metabolites were formed in almost identical quantities and responded equally to CYP3A inhibition by midazolam and ketoconazole (data not shown). Although norverapamil and d-617 were quantified individually, for ease of data presentation and to reduce the number of parameters that required estimation during compartmental modeling, metabolite levels have been pooled. Data were corrected for buffer evaporation over the course of the experiment (7.38 ± 1.62% over 3 h, mean ± S.D. n = 36). The amount of verapamil and metabolite detected in intestinal tissue segments were corrected for verapamil and metabolite present in the adhered buffer volume, via data determined in studies conducted with [³H]polyethylene glycol 4000 (adhered buffer volume was 58.1 ± 9.5 μl · cm^-2, mean ± S.D., n = 9). Apparent permeability coefficients were calculated from both verapamil appearance in the receptor chamber and disappearance from the donor chamber using eqs. 1 and 2, respectively (Martin, 1993). where dX/dt is the rate of change of amount of verapamil transported into the receptor chamber with respect to time (i.e., the flux, nmol · min^-1), C₀ is the initial concentration in the donor chamber, V is the volume of the chamber (7 ml), A is the surface area available for diffusion (1.78 cm²), and k_u is the uptake rate constant (min^-1) determined from the negative slope of a log_e donor chamber concentration versus time plot. The extraction ratio (ER) of verapamil across the tissue was determined at t = 180 min by eq. 3 (Cummins et al., 2001), where the subscripts d, r, and t represent the total amount (nmol) of metabolite or parent compound in the donor, receptor, and tissue compartments, respectively.

Compartmental Modeling. Compartmental kinetic modeling of the diffusion chamber data from control experiments was performed using nonlinear ordinary least-squares regression (ModelMaker 4.0, ModelKinetix; AP Benson, Reading, Berkshire, UK). To adequately fit a closed model to the diffusion chamber data, correction for any discrepancy in mass balance had to be made; therefore, data in each compartment (donor chamber, receptor chamber, or tissue segment) were adjusted proportionally to achieve a mass balance of 100% (requiring 6 ± 6% adjustment, mean ± S.D., n = 52). Regression and parameter estimation was achieved using the Levenberg-Marquardt fitting algorithm with fourth-order Runge-Kutta integration, and models were simultaneously fitted to data from all individual m to s and s to m control experiments only. Model schemes are shown in Fig. 1, where Model I represents the simplest model to describe transport events across the tissue segments, and compartments 1, 2, 3, and 0 represent the mucosal chamber, tissue segment, serosal chamber, and metabolite pool, respectively, and were all assumed to be well stirred. The metabolite pool represents the total amount of metabolite formed at the conclusion of the experiment (mucosal, tissue, and serosal compartments combined). In the interest of parsimony and identifiability, all rate constants in Model I were assumed to be first order in nature, despite k₂₁ possibly being better represented by parallel first-order (passive diffusion) and Michaelis-Menten (capacity-limited P-gp efflux) functions and k₂₀ by a second Michaelis-Menten function (capacity-limited metabolism). Model IIa includes a second tissue compartment, essentially separating the enterocyte (where metabolism occurs) from the submucosal layers of the tissue, and the sum of verapamil in these two compartments was therefore fitted to the tissue data and represented total tissue uptake. To simplify Model IIa, it was assumed that partitioning between the enterocyte (compartment 2) and submucosal layers (compartment 3) was first order in nature and equal in both directions (i.e., k₂₃ = k₃₂); as with Model I, all other rate constants in Model IIa were assumed to be first order. Model IIb is identical to Model IIa, with the exception of k₂₀ being replaced by the Michaelis-Menten function: V_max/(K_m + X₂), where V_max, K_m, and X₂ represent the maximum rate of metabolism (% · min^-1), the Michaelis-Menten constant (%), and the amount of verapamil in compartment 2 (%), respectively. As a means of testing the validity of the Michaelis-Menten metabolism parameters, a simulation of the ER of 200 μM verapamil was compared with experimental data collected using 200 μM verapamil in the donor chamber. This simulation does not conclusively justify the use of the more complex Model IIb; however, it does provide some evidence that the derived metabolism parameters are reasonable estimates. More complicated models, such as those describing the individual disposition of each metabolite in the tissue, mucosal, and serosal compartments, could not be supported by the current data set and resulted in large CVs of these additional parameters (data not shown). The aims of the modeling procedures described above were not compromised by this simplification and still allowed the discrepancy between disappearance and appearance permeability coefficients to be reconciled and examination of the ER under certain theoretical conditions to be examined. The compartmental model analysis was not extended to the PSC833, midazolam, and ketoconazole data sets, because the parameter estimates obtained were complicated by difficulties in accurately defining the individual degree of inhibition of P-gp and/or CYP3A by these inhibitors.

Simulations based on Models IIa and IIb were conducted to further examine the impact of P-gp on CYP3A metabolism by calculating the model-simulated extraction ratio under various conditions (ER, calculated as per eq. 3 from simulated data at 180 min). These simulations were only performed in the m to s direction. The model parameter k₂₁ was assumed to encompass P-gp-mediated efflux that removed drug from compartment 2 to compartment 1. The rate constant k₂₁ was varied from 20 to 180% of its estimated parameter value to simulate the impact of inhibiting and stimulating P-gp efflux on the ER. Simultaneous to changes in k₂₁, k₂₀ in Model IIa was varied to simulate changes to intestinal intrinsic clearance, and the K_m of metabolism in Model IIb was varied to simulate pseudo-first-order and pseudo-zero-order metabolism conditions. The percentage change in ER observed with each of these conditions with respect to the control ER (model predicted) was then calculated.

Statistical Analysis. Comparisons within experimental data sets (e.g., permeability coefficients, tissue uptake, extraction ratios) collected with or without the inhibitors midazolam, ketoconazole, and PSC833 were compared by one-way analysis of variance using either parametric (general linear model) or nonparametric (Kruskal-Wallis analysis of variance on ranks) methods. Post hoc multiple comparisons between groups were made using Tukey (parametric) or Dunn's (nonparametric) tests, where p < 0.05 was considered significant. Statistical comparisons between alternative pharmacokinetic models were made by comparison of the residual sums of squares (RSS), the %CV of the parameter estimates, bias or systematic deviation of standardized residuals, and the F ratio test and Akaike Information Criterion (AIC) as tests for parsimony (Boxenbaum et al., 1974; Landaw and DiStefano, 1984).

Results

Impact of Inhibitors on the Uptake and Transport of Verapamil. The bidirectional transport profiles of verapamil in the presence and absence of the inhibitors midazolam, ketoconazole, and PSC833 are shown in Fig. 2. The permeability coefficients of verapamil, determined from verapamil appearance in the receptor chamber and disappearance from the donor chamber, in the presence and absence of the inhibitors, are summarized in Table 1. Polarized transport (P-gp efflux) of verapamil was evident, and significant differences between m to s and s to m appearance permeability coefficients were observed in control, PSC833, and midazolam groups (p < 0.002), but not in the presence of ketoconazole. Appearance permeability or flux of verapamil in the m to s direction was significantly increased from control in the presence of midazolam and ketoconazole (p < 0.05) as a result of reduced metabolic clearance and reduced metabolic clearance and P-gp efflux inhibition, respectively. Surprisingly, the permeability of verapamil was not altered by PSC833, which is a putative P-gp inhibitor. Although midazolam increased the s to m transport of verapamil when compared with control, the steady-state flux (as used for calculation of the permeability coefficient) was not significantly increased (Fig. 2A). Figure 2B depicts the transport of verapamil in the absence of significant metabolism (plus midazolam) and in the absence of significant metabolism and P-gp inhibition (plus ketoconazole). The m to s and s to m transport profiles under P-gp inhibition (plus ketoconazole) are midway between those in the absence of P-gp inhibition (plus midazolam), consistent with inhibition of an apically localized drug efflux process.

Permeability coefficients measured from verapamil disappearance from the donor chamber were not significantly different between m to s and s to m data sets, or significantly affected by any of the inhibitors (Table 1). Interestingly, permeability coefficients calculated on verapamil disappearance from the donor chamber were approximately 10-fold higher than those calculated on verapamil appearance and, apparently, completely insensitive to the effects of the metabolic and/or P-gp inhibition. These data suggest substantial tissue uptake of verapamil during the course of the experiment, which is consistent with the significant accumulation of verapamil and metabolite within the tissue segment observed at the conclusion of transport experiments (Table 2). The amount of verapamil in the intestinal tissue segment at the conclusion of these transport studies was not significantly altered by the direction of the transport study or the presence of PSC833, midazolam, or ketoconazole.

Impact of Midazolam, Ketoconazole, and PSC833 on Metabolite Formation. As reported previously (Johnson et al., 2001, 2002), d-617 and norverapamil are the two major verapamil metabolites formed in the rat intestine, and in the current studies, they were formed in almost identical quantities (data not shown). For ease of data presentation, the metabolite data described in Fig. 3 represent the sum of both metabolites. In control m to s experiments, approximately twice as much metabolite was detected in the mucosal chamber compared with intact verapamil detected in the receptor chamber, reinforcing the likelihood that P-gp is partly responsible for the large amount of metabolite seen in this donor chamber. The appearance of metabolite in this chamber was nonlinear with respect to time, with metabolite formation significantly reduced by midazolam, totally abolished by ketoconazole, and generally unchanged in the presence of PSC833. The specificity of midazolam to reduce metabolism but not efflux was evident from the significant difference in m to s and s to m appearance permeability coefficients for verapamil in the presence of midazolam (i.e., evidence of a lack of P-gp inhibitory activity, Table 1) at the same time as a significant reduction in metabolite formation (Fig. 3). The amount of metabolite recovered from the tissue segment after 180 min was significantly decreased by midazolam and ketoconazole, but unaffected by PSC833 (Table 2).

Determination of Verapamil Extraction Ratio. Utilizing eq. 3, the ER of verapamil across rat jejunal tissue was determined at 180 min (Table 3). The ER in the m to s direction was 0.31 ± 0.04 and was halved when experiments were performed in the s to m direction. Inhibition of CYP3A by midazolam, as expected, reduced the ER of verapamil by 80%, and no extraction of verapamil was observed in the presence of ketoconazole (i.e., 100% reduction). The ER was not significantly altered by the P-gp inhibitor PSC833.

Compartmental Modeling. Predicted data from Models I, IIa, and IIb are shown in Fig. 4, and a summary of the parameter estimates and model selection criteria is presented in Table 4. Model I does not adequately predict the difference in the total amount of metabolite formed between m to s and s to m experiments, creating bias in the tissue and metabolite residuals, and therefore predicts an insignificant difference in the m to s and s to m ER that is inconsistent with the experimental data (Table 4). The inclusion of a second tissue compartment (Model IIa) did not significantly increase the %CV of parameter estimates and provided a superior fit to the data over Model I, as evidenced by a decrease in the RSS, less bias in the tissue and metabolite residuals (see Fig. 4), a lower AIC, and a significant F ratio test between Models IIa and I (p < 0.025). The inclusion of the second tissue compartment predicts that the amount of verapamil in the enterocyte compartment is approximately 2-fold higher in m to s experiments when compared with s to m. This difference, in addition to the potential impact of P-gp efflux, was adequate to explain the differences in m to s and s to m metabolism observed in these studies and predicts an ER that is in excellent agreement with the observed data (Tables 3 and 4). Model IIb is structurally identical to Model IIa, with the exception of the replacement of k₂₀ (the first-order metabolism parameter) with a Michaelis-Menten function. This increase in model order lowered the RSS but was not considered a statistically better model than Model IIa, due to a higher AIC, a nonsignificant p value for the F ratio test (between Models IIb and IIa), and large CVs obtained for the additional metabolism parameters (Table 4). However, notwithstanding the high CVs of the metabolism parameters in Model IIb, the predicted ER for verapamil in m to s and s to m experiments were still consistent with the experimental data (Table 4). Furthermore, using Model IIa, the m to s ER of verapamil at 200 μM donor chamber concentrations was (as expected for an entirely linear model) the same as that predicted with 20 μM donor chamber concentrations (i.e., 0.31). In contrast, Model IIb predicted that the ER at 200 μM verapamil donor chamber concentrations would drop to 0.063, which is in excellent agreement with experimental data collected from 200 μM m to s verapamil permeability studies where the ER was found to be 0.071 ± 0.018 (mean ± S.D., n = 9).

Simulations on the Impact of P-gp on Intestinal ExtractionRatio. To probe the interaction of P-gp and CYP3A in a theoretical setting, the influence of P-gp on the modulation of verapamil metabolism (as described by the ER) was first investigated by varying k₂₁ in Model IIa to simulate P-gp inhibition and stimulation. However, in an entirely linear model (such as Model IIa), changes in k₂₁ cannot result in changes to the m to s ER, and the ER only varies significantly with changes to intrinsic clearance or k₂₀ (Fig. 5A). Therefore, Model IIb was developed to introduce the potential for nonlinear verapamil metabolism, where k₂₀ was replaced by a Michaelis-Menten function (described above). Simulations of the impact of varying k₂₁ (varying efflux) were then performed using Model IIb as a mechanism to illustrate the potential impact of P-gp on the extent of metabolism in a nonlinear system.

The degree of “linearity” of metabolism within the model can be varied by adjustment of the K_m in Model IIb, where the higher the K_m, the more “linear” the metabolism becomes (i.e., the more difficult to saturate, and the model tends toward Model IIa: entirely linear). Simulations performed using Model IIb and a high K_m show a reduced “P-gp dependence” on the ER, consistent with simulations utilizing Model IIa, where no P-gp dependence on the ER was observed. Simulations performed using a low K_m (essentially creating pseudo-zero-order or saturated metabolism conditions) resulted in an increased dependence of the ER on P-gp efflux (k₂₁). The above findings are illustrated in Fig. 5B, where the initial model-predicted ER of verapamil is plotted centrally, and changes in P-gp activity (via k₂₁) and linearity (via K_m) are indicated on their respective axes as the percentage change in the original parameter value from Table 4. As shown in Fig. 5B, increasing P-gp efflux activity (k₂₁) leads to an increase in ER, and for those compounds with a lower K_m, the ER is most sensitive to a change in efflux, i.e., changes to P-gp efflux are most likely to have the greatest impact on the ER of compounds where intestinal metabolism is most readily saturated.

Discussion

In these experiments, the appearance of verapamil in the receptor chambers displayed significant polarity, confirming verapamil to be a substrate for an efflux transporter in this model, a fact sometimes obscured by verapamil's high intrinsic permeability (Polli et al., 2001). In the presence of midazolam, a CYP3A inhibitor, the cellular metabolism of verapamil was reduced, and this increased verapamil flux regardless of the direction of transport, i.e., m to s, or s to m (Tables 1 and 3; Figs. 2 and 3). Verapamil transport under conditions of metabolic inhibition (i.e., plus midazolam, inhibition of metabolism alone; and plus ketoconazole, inhibition of metabolism and P-gp) showed that P-gp inhibition could significantly reduce the efflux of verapamil (Fig. 2B). Surprisingly, however, the P-gp inhibitor PSC833 had little impact on verapamil transport, despite it being an effective and selective inhibitor of verapamil efflux in situ (Johnson et al., 2003) and of digoxin efflux in the rat intestine in vitro (Johnson et al., 2002). A possible reason for this discrepancy is that PSC833 cannot adequately compete for the P-gp efflux pump in the presence of the large amounts of verapamil retained in the tissue in the present in vitro studies [6.4% · cm^-2 compared with 0.3% · cm^-2 in the previous in situ studies (Johnson et al., 2003)]. This higher retention of verapamil in the tissue of the in vitro model is presumably a result of reduced partition into the aqueous transport buffers. This suggestion is consistent with previous studies in which an increase in verapamil transport across Caco-2 monolayers was observed when human plasma was employed in the receptor chamber (Chung et al., 2001). Unfortunately, the employment of lower concentrations of verapamil and higher concentrations of PSC833 were impractical in the current studies due to limitations of analytical sensitivity and solubility, respectively.

Although ketoconazole significantly increased the m to s permeability of verapamil through inhibition of both P-gp and CYP3A, the s to m permeability of verapamil was unchanged in the presence of ketoconazole when compared with control. This most likely reflects the fact that the effects of P-gp and CYP3A inhibition on the s to m transport of verapamil run counter to each other; i.e., P-gp and CYP3A inhibition should reduce and increase the s to m flux of verapamil, respectively; and, therefore, inhibition of both of these processes may result in no net change in s to m transport (Fig. 2D).

In contrast to the verapamil appearance data, the disappearance of verapamil from the donor chamber in m to s or s to m experiments was more rapid and was not significantly altered by the addition of the P-gp and/or CYP3A inhibitors. The insignificant changes in disappearance permeability are most likely due to the small attenuation of verapamil uptake resulting from P-gp efflux, relative to the significant tissue accumulation of verapamil observed. It is evident, therefore, that permeability assessment made only using drug disappearance data may show less sensitivity to the potential impact of P-gp and/or CYP3A inhibition on drug absorption. This is consistent with data obtained in human jejunal perfusion studies where the presence or absence of ketoconazole in the jejunal perfusate had no impact on the disappearance permeability coefficient of verapamil, but led to a significant increase in the absorptive flux and bioavailability when compared with control (Sandström et al., 1999). The disappearance permeability coefficient of verapamil in this in vitro model was also significantly higher (approximately 10-fold) than that calculated on verapamil appearance in the receptor chamber, another observation consistent with the substantial tissue uptake of verapamil and the affinity of lipophilic compounds for the tissue segment in such in vitro experiments. Such an observation may therefore cast doubt on the validity and usefulness of in vitro permeability experiments in which permeability has been assessed on drug appearance data alone, and may also mask the impact of P-gp (or P-gp inhibitors) on drug permeability.

In an effort to examine the impact of P-gp on the intestinal metabolism of verapamil more thoroughly (Johnson et al., 2001), a mechanistic assessment of the interaction between P-gp efflux and CYP3A metabolism was undertaken utilizing a compartmental kinetic model. Model I describes the most parsimonious system with which to describe the in vitro permeability data. However, this simple three-compartment model led to under- and overestimation of the extent of verapamil metabolism in m to s and s to m transport experiments, respectively (Fig. 4A), and in the model-predicted ER (Table 4). A four-compartment model was therefore proposed to allow inclusion of a submucosal tissue compartment (Model IIa), and application of this model resulted in a superior fit to the data. Model IIa predicts an approximately 2-fold higher verapamil concentration in compartment 2 (the putative metabolizing compartment) in m to s studies compared with s to m, which, in addition to the potential impact of P-gp, may explain the higher ER observed in m to s experiments when compared with s to m. Permeability data analysis using compartmental kinetics is not commonly performed; however, the current analysis has demonstrated that the interaction between uptake, efflux, metabolism, and transport processes occurring within the intestine may be best examined using such techniques. This type of analysis allowed the discrepancy between disappearance and appearance permeability coefficients, and the attainment of steady-state tissue levels of verapamil to be reconciled in a valid and mechanistic manner.

Using Model IIa as an initial framework, the impact of P-gp on the extent of metabolism was further probed by running a number of theoretical simulations. Assuming that P-gp only influences passage of verapamil across the apical membrane and, therefore, that P-gp inhibition results only in a lowered value of k₂₁, simulation studies performed with Model IIa (an entirely linear model) demonstrate that lowering or raising the value of k₂₁ cannot result in a significant change in the m to s ER (Fig. 5A). Rather, alterations to k₂₁ lead to proportional changes in tissue uptake, transport, and metabolism, and therefore, the ER (a ratio of these proportionally varying components) is relatively unchanged. However, experimental studies similar to those described here, performed with the P-gp/CYP3A substrates K11777 and indinavir across Caco-2 cells expressing CYP3A4, are inconsistent with these simulations (Fig. 5A) and demonstrate that P-gp inhibition can lead to a reduction in the apical to basolateral ER (Hochman et al., 2000; Cummins et al., 2002). The changes in ER evident in these previous Caco-2 studies can be most simply explained by the inclusion of a nonlinear function to describe drug metabolism within the enterocyte (Model IIb). This inclusion uncouples the directly proportional relationship between the extent of metabolism and the extent of verapamil transport inherent in Model IIa, consistent with conclusions made by Hochman et al. (2000). From the simulations in Fig. 5B, it can be seen that the greater the metabolizing capacity (higher K_m of metabolism), the smaller the impact k₂₁ (or P-gp) has on the ER and, conversely, that under pseudo-zero-order metabolism (low K_m), the largest P-gp effect is observed.

Given the complex nature of drug disposition in the current in vitro system, Fig. 6A may depict a more physiologically representative model. This model allows for P-gp and CYP3A to act on different drug pools within the enterocyte, consistent with the suggestion that P-gp extrudes drug from the apical membrane, whereas CYP3A draws on its substrates from the cytoplasm (Raviv et al., 1990; Stein, 1997). Fitting the model described in Fig. 6 to the current data set is beyond the realm of identifiability; however, simulations run using the rate constants obtained in Model IIa as estimates clearly show that substantial changes in ER would be expected if compartment 4 (the metabolizing compartment) is capacity limited, even if all rate constants (including k₄₀) are first-order. Such a situation might occur, for example, on saturation of a cytoplasmic binding protein, a suggestion that is consistent with observations made in previous in situ studies (Johnson et al., 2003).

In summary, the m to s appearance permeability of verapamil was significantly increased by CYP3A and CYP3A/P-gp inhibition, indicating the significant influence of these processes on verapamil transport in vitro. A substantial discrepancy between permeability coefficients calculated on verapamil disappearance from the donor chamber and appearance in the receptor chamber was well captured by a compartmental model, and allowed a comprehensive description of the data gathered in the current in vitro studies. From the compartmental modeling described, it is evident that P-gp-mediated changes in the intestinal ER may be stimulated via changes to the intracellular drug concentration and, therefore, saturation (or desaturation) of the metabolizing enzyme. Alternatively, P-gp may influence the ER via a currently unidentified mechanism whereby the accumulation and transport of verapamil across the enterocyte are disproportionately related to the extent of metabolism. In either case, under appropriate kinetic conditions, P-gp efflux inhibition may result in disproportionate changes to the extent of drug metabolism relative to drug transport and directly impact on the intestinal extraction ratio of P-gp/CYP3A substrates.

Appendix

The differential equations used to fit Models I, IIa, and IIb to the control data set are shown below. As described in Fig. 1, all rate constants were assumed to be first-order (min^-1), except k₂₀ in Model IIb, where k₂₀ was replaced by a Michaelis-Menten function (V_max/(K_m + X₂), where the dimensions of V_max are % · min^-1 and K_m and X₂ are percentage.

Model I:

Model IIa:

Model IIb: