Experimental Procedures

Materials.

Baculovirus/insect cells, cDNA-expressed CYP3A4 and control microsomes (Supersomes), 7-benzyloxyquinoline (BQ), 7-benzyloxy-4-trifluoromethylcoumarin (BFC), and dibenzylfluorescein (DBF) were obtained from GENTEST Corporation (Woburn, MA). 7-Benzyloxyresorufin (BzRes) was obtained from Molecular Probes (Eugene, OR). The test chemicals and their suppliers were as follows: itraconazole and cisapride (Janssen Biotech, Flanders, NJ); nifedipine, cyclosporin A, erythromycin, terfenadine, clotrimazole, (±)-verapamil, tamoxifen, testosterone, troleandomycin, progesterone, quercetin, ethynylestradiol, α-naphthoflavone, carbamazepine, (±)-miconazole, quinidine, nicardipine, astemizole, diazepam, and propofol (Sigma-Aldrich, St. Louis, MO); midazolam, 1′-hydroxymidazolam, ketoconazole, and oxidized nifedipine (Ultrafine Chemicals, Manchester, UK); digitoxin, nimodipine, and haloperidol (ICN Biochemicals, Inc., Aurora, OH); and mibefradil (a generous gift of Dr. Rudolfo Gasser, Roche Pharmaceuticals, Basel, Switzerland). All other chemicals were obtained from Sigma-Aldrich.

Flourometric Enzyme Inhibition Assays.

Incubations were conducted in a 200-μl volume in 96-well microtiter plates (Catalog 3915, Corning Costar, Cambridge, MA) based on the method described on the GENTEST Corporation website (www.gentest.com) and as described below. Serial dilutions were performed using a Multiprobe II liquid handling station (Packard Instruments, Downers Grove, IL). A cofactor/serial dilution (C/SD) buffer was prepared in 50 mM potassium phosphate, pH 7.4. This buffer contained 2.6 mM NADP⁺, 6.6 mM glucose 6-phosphate, 0.8 U of glucose-6-phosphate dehydrogenase/ml, and 0.1 mg/ml microsomal protein prepared from wild-type baculovirus-infected insect cells. C/SD buffer (100 μl) containing twice the upper concentration of inhibitor was added to the first well in each row. In the second well, 50 μl of that solution was added. In the second well and all remaining wells, 100 μl of C/SD buffer that contained the solvent vehicle but lacked test compound was added. Fifty microliters of the inhibitor solution from the second well in each row was then transferred into the third well and serially diluted 1:3 through the eighth well. Wells 9 and 10 contained no inhibitor, and wells 11 and 12 were used as controls for background fluorescence (enzyme and substrate were added after the reaction was terminated). The final concentration of the inhibitors in the first well varied between 1 and 1000 μM, depending on the solubility characteristics or potency of the inhibitor. All inhibitors were dissolved in acetonitrile for addition to the incubations. The plate was then prewarmed at 37°C for 10 min, and the reaction was initiated by the addition of 100 μl of prewarmed enzyme/substrate (E/S) mix. The E/S mix contained 0.35 M potassium phosphate buffer (pH 7.4), cDNA-expressed P450 in insect cell microsomes, substrate, and wild-type baculovirus insect cell microsomes to give the final assay concentrations in a reaction volume of 200 μl. Reactions were terminated after various times (see Table1) by addition of 75 μl of a 4:1 acetonitrile:0.5 M Tris base solution, except for DBF, in which the reaction was terminated by the addition of 2 N sodium hydroxide. Fluorescence in each well was measured using a BMG Labtechnologies Inc. FLUOstar model 403 fluorescence plate reader (Durham, NC). The DBF metabolite, fluorescein, was measured by using an excitation wavelength of 485 nm and an emission wavelength of 538 nm. The BzRes metabolite, resorufin, was measured by using an excitation wavelength of 530 nm and an emission wavelength of 590 nm. The BFC metabolite 7-hydroxy-4-trifluoromethylcoumarin and the BQ metabolite 7-hydroxyquinoline were measured by using an excitation wavelength of 410 nm and an emission wavelength of 538 nm. Production of the products of each assay was proportional with time and protein concentration. Data were exported and analyzed using an Excel spreadsheet. The IC₅₀ values were calculated by linear interpolation.

Testosterone, Nifedipine, and Midazolam Assays.

Incubations were conducted twice in duplicate in a 200-μl volume in 96-deep-well polypropylene microtiter plates (Waters Corporation, Milford, MA). A cofactor solution was prepared in 50 mM potassium phosphate, pH 7.4. This buffer contained 2.6 mM NADP⁺, 6.6 mM glucose 6-phosphate, 0.8 U glucose-6-phosphate dehydrogenase/ml, and 0.1 mg of microsomal protein/ml prepared from wild-type baculovirus-infected insect cells. For sample and control wells (substrate added after the reaction was terminated), 100 μl of cofactor solution was added that contained twice the final concentration of inhibitor. The final concentration of inhibitors examined was 10 μM. Reference wells contained 100 μl of cofactor buffer and contained solvent vehicle but lacked test compound. The plate was prewarmed at 37°C for 10 min and the reaction initiated by the addition of 100 μl of prewarmed E/S mix. The E/S mix contained 0.35 M potassium phosphate buffer (pH 7.4), cDNA-expressed P450 in insect cell microsomes, substrate, and wild-type baculovirus insect cell microsomes to give the final assay concentrations in a reaction volume of 200 μl. The final concentration of cDNA-expressed P450 was 7.5, 5, and 0.5 nM for testosterone, nifedipine, and midazolam, respectively. Total microsomal protein was 0.25 mg/ml. Substrate concentrations (chosen to approximate theK_m) were 50, 10, and 3 μM for testosterone, nifedipine, and midazolam, respectively. Reactions were terminated after 5 min for all assays by addition of 100 μl of acetonitrile (testosterone), 40 μl of 96:4 acetonitrile-glacial acetic acid (nifedipine), or 50 μl of 96:4 acetonitrile-glacial acetic acid containing 0.2 μM diazepam (midazolam). Diazepam was used as an internal standard. After stopping the reactions, microtiter plates were centrifuged at 2750g for 10 min at 4°C (model 5810r, Eppendorf, Cologne, Germany). A portion of the supernatant was analyzed by HPLC/absorbance (testosterone, nifedipine) or LC/MS (midazolam). Substrate utilization for these assays was 16% or less.

Analytical.

HPLC/UV

The HPLC system consisted of a Waters 2690 separations module. A C18 reversed-phase column (Zorbax, Hewlett-Packard, obtained from VWR, Boston, MA; 4.6 × 250 mm, 5 μ) was used to chromatograph the testosterone and nifedipine samples. For testosterone, the gradient was linear from 53% B to 58% B to 8 min and increased to 100% B over 0.1 min, and was held until 14 min before returning to the starting conditions (solvent A: 10% methanol; solvent B: 100% methanol; total flow rate, 1 ml/min). The 6β-hydroxytestosterone metabolite was monitored at a wavelength of 254 nm and eluted at ∼6.1 min. The metabolite was quantified using a 6β-hydroxytestosterone standard curve. For nifedipine, solvents were held isocratic at 62% D from 0 to 6 min. The gradient was linear from 62% D to 100% D from 6 to 7 min and held at 100% until 10.1 min before returning to starting conditions (solvent C: HPLC-grade water; solvent D: 100% methanol; total flow rate, 1 ml/min). The pyridine metabolite was monitored at a wavelength of 254 nm and eluted at ∼5.7 min. The metabolite was quantified using a standard curve of oxidized nifedipine.

LC/MS.

The HPLC system consisted of a Waters 2790 separations module and was operated in sequential mode. A C8 reversed-phase column (Waters, Symmetry, 4.6 × 50 mm, 3.5 μ) was used to chromatograph the midazolam samples. The gradient was linear from 0% F to 100% F in 2 min and held at 100% F for 30 s before returning to the starting conditions (solvent E: 10% acetonitrile, 5 mM ammonium acetate, pH 3.4; solvent F: 100% acetonitrile). The column was equilibrated 1 min before injecting the next sample. The flow rate was 1.5 ml/min with the first 1.1 min diverted to waste before directing the effluent to the mass spectrometer for 1.1 min. The mass spectrometer was a Micromass ZMD single quadrupole (Manchester, UK) using positive atmospheric pressure chemical ionization (APCI) with selected ion-monitoring detection of 1′-hydroxymidazolam (m/z 342) and the internal standard, diazepam, (m/z 285). The APCI needle was set at 2.6 kV, the cone at 37 V, and the APCI heater probe to 600°C. Under these conditions, 1′-hydroxymidazolam eluted at 1.5 min and diazepam eluted at 2.0 min. Quantification standards of 1′-hydroxymidazolam were run at the start and end of each sample set.

Results

Assay Design.

The assay design was similar to that originally reported by Crespi et al. (1997) in that IC₅₀ values were determined using eight inhibitor concentrations generated by serial 1:3 dilutions. However, several aspects were modified in the present study to minimize the differences among the assays. In particular, to control for the effects of any nonspecific binding of the compound to the microsomes (Obach, 1997), the final microsomal protein concentration was standardized to 0.25 mg/ml by the addition of control microsomes (prepared from insect cells infected with wild-type virus). ApparentK_m values were measured under this standardized protein concentration (Table 1). These values were not significantly different from previous determinations without protein standardization. Representative kinetic plots are provided in Fig.1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Representative kinetic plots for metabolite formation for BzRes (A), BQ (B), DBF (C), and BFC (D).

Incubation times and enzyme concentrations were as specified underExperimental Procedures, except that the BFC assay incubation time was 10 min.

To detect competitive inhibitors with comparable efficiency, BQ and DBF were used at concentrations equal to the apparent S₅₀ and K_m values, respectively. BzRes was used at the same concentration reported earlier (Crespi et al., 1997), which was slightly higher than the apparentK_m as determined in Table 1. BFC metabolism was linear with respect to BFC concentration up to 100 μM. Above 160 μM concentration, a precipitate was clearly evident. A BFC concentration of 50 μM was used for routine inhibition analysis.

To facilitate the testing of compounds with limited aqueous solubility and to mitigate inhibitor loss caused by binding to the surface of the microtiter plates, control microsomes (0.1 mg/ml) were present in the serial dilution buffer. Finally, the preparation of the plates for the assay was conducted on a robotic liquid-handling system. In the aggregate, the assays were performed in a way similar to that used in a cytochrome P450 inhibition assay screening program in support of drug lead optimization. However, these experiments differ from a true lead optimization program in that the 27 compounds tested were not all of the same structural series.

Responses and IC₅₀ Values for Four Fluorometric Substrates.

The mean IC₅₀ values for 27 compounds is shown in Table 2, and graphical representations of typical curves are contained in Fig. 2. When an IC₅₀ was not determined, 50% inhibition did not occur at the highest concentration tested. In some cases, this was accompanied by activation, although activation sometimes occurred before inhibition for curves exhibiting inhibition by 50% or more (e.g., midazolam in Fig. 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Representative experiments showing the effects of eight concentrations of the test compound with the four substrates.

Different substrates are indicated by symbols in the lower right plot. Each compound tested is identified at the top of each plot.

The mean and standard deviation for the replicate IC₅₀ values were calculated for each determination. For each inhibitor/substrate pair considered individually, the mean within-day coefficient of variation (CV)/between-day CV was 0.19/0.21, 0.11/0.16, 0.12/0.19, and 0.11/0.17 for BzRes, BQ, DBF, and BFC, respectively. This level of variability is consistent with the number of liquid transfer steps in the assay and the precision of the liquid-handling station (1–2% error per transfer) with a small component of other sources of error. The variability in the assay was similar across probe substrates. In contrast, the mean CV for each inhibitor, considering the substrates collectively, was 0.98. This observation is one indicator of substrate-dependent effects in inhibition potency beyond the effect of experimental error. When the means of all four IC₅₀ determinations (duplicates on 2 days) were compared for each inhibitor and the four substrates, the range in IC₅₀ values in Table 2 varied from 2.1-fold (propofol) to 195-fold (cyclosporin A). The mean range was 29-fold. As with the above CV comparison, the large range in IC₅₀ values is indicative of substrate-dependent differences in IC₅₀ values and not experimental error.

The graphical presentations of the data in Fig. 2 indicate that for many compounds the responses for pairs of substrates tended to track together. However, the pairing of the substrates was often different among inhibitors. For example, BFC and DBF tracked together for the azole antifungal compounds ketoconazole, miconazole, itraconazole, and clotrimazole but not for mibefradil and erythromycin. There were several examples of qualitatively different responses (i.e., activation of the metabolism of some substrates, inhibition of the metabolism of other substrates, or no inhibitory effect). For several of these compounds (tamoxifen, midazolam, terfenadine, haloperidol, and carbamazepine), the deviations from control activity levels (inhibition or activation) often occurred in the same compound concentration ranges. For other compounds (testosterone, progesterone, and α-naphthoflavone), activation (BzRes and BFC) was observed at lower concentrations relative to inhibition (DBF). Finally, several examples of incomplete inhibition were observed. However, only for BQ/cyclosporin A did this have a large effect on the IC₅₀ value. This observation explains the very large range in IC₅₀ values for cyclosporin A.

Overall Sensitivity of the Substrates to Inhibition.

For each inhibitor, the relative IC₅₀ values for each probe substrate was calculated (Table 2). The relative IC₅₀ values for BQ, BzRes, DBF, and BFC were 1.66, 1.34, 0.79, and 0.31, respectively. Therefore, within the context of this screening paradigm, BFC appears to be the most sensitive substrate and BQ the least sensitive. The greater sensitivity of BFC is further illustrated in Fig. 2, where BFC is always in the more sensitive group for compounds causing inhibition to two or more substrates.

As indicated in Table 1, BFC was used at a concentration at which the rate of metabolism was linear with respect to substrate concentration, while the other substrates were used at a concentration near the apparent K_m or S₅₀. For competitive inhibitors and Michaelis-Menten kinetics, the IC₅₀ is 2 times theK_i and the IC₃₃equals the K_i. IC₃₃values were calculated from the inhibition curves (Cheng and Prusoff, 1973). Corresponding mean IC₃₃ values (relative to the mean IC₅₀ value) for BzRes, BQ, and DBF were 0.48, 0.58, and 0.32, respectively. These are only slightly larger than the 0.31 value for relative BFC IC₅₀. Based on assumptions of Michaelis-Menten kinetics and competitive inhibition, and when the inhibitors are considered as a whole, the inherent sensitivity of the four substrates is similar. However, as shown in Table 1, not all of the substrates conform to Michaelis-Menten kinetics. In addition, all inhibitors are not competitive (e.g., ethynylestradiol and troleandomycin are mechanism-based inhibitors). Detailed analyses of the modes of inhibition were beyond the scope of this study. Therefore, there are limits to the above relative sensitivity comparison, and the numbers should be regarded as guidelines.

Correlation among the Substrates.

Three correlation analyses were performed to assess the degree of similarity among the substrates. The approach of Kenworthy et al. (1999) was applied to our data with the fluorometric substrates. Specifically, the percent inhibition at a single inhibitor concentration (10 μM) was calculated by either linear interpolation or extrapolation (for potent inhibitors). In addition, three traditional substrate probes, testosterone 6β-hydroxylase (TS), midazolam 1′-hydroxylase (MDZ), and nifedipine oxidase (NF), were used to measure the extent of inhibition at 10 μM inhibitor. These results are presented in Table 3. Linear regression analysis was performed based on the percent inhibition. Compound/substrate pairs that exhibited activation were treated as 0% inhibition. The results are presented in Table4, part A. With this approach, BQ, DBF, and BFC were found to correlate with each other (correlation coefficients between 0.80 and 0.85) and with both TS and MDZ (correlation coefficients between 0.81 and 0.93). BQ, BFC, and DBF correlated more weakly with NF (correlation coefficients between 0.51 and 0.78). BzRes also correlated more weakly with the other fluorometric substrates (correlation coefficients of 0.49–0.68) and traditional substrates (correlation coefficients of 0.51–0.76).

We have also performed linear regression analysis on the IC₅₀ and IC₃₃ values for the fluorometric substrates. These results are presented in Table 4, part B. The results of these analyses were consistent with the correlation observed in Table 4, part A (i.e., BQ, BFC, and DBF correlated with each other, whereas BzRes correlated more poorly with the other substrates). However, with the exception of BQ/BzRes IC₅₀ correlation (0.15), the difference was modest. In addition, the IC₃₃ (and BFC IC₅₀) values showed higher correlation coefficients than the IC₅₀ values. As expected, the IC₃₃ and IC₅₀ values for a probe substrate correlated with each other (correlation coefficients between 0.98 and 0.87), although the extent of correlation was lower than might initially be expected. This reduction was caused by the influence of several partial inhibitors that tended to elevate IC₅₀ values relative to the IC₃₃.

The correlation analysis of Kenworthy et al. (1999) primarily detects differences in response to inhibitors of intermediate potency (i.e., potent inhibitors and weak inhibitors are all grouped together). A linear regression analysis of IC₅₀ values is most sensitive to differences for the weakest inhibitors. To provide some assessment of the trend across all inhibitor concentrations, we performed a linear regression analysis on the log transformed IC₃₃ and IC₅₀ values. The results are presented in Table 3, part C. While the correlation coefficients are generally higher, the same trend is evident: BQ, BFC, and DBF tend to correlate with each other, whereas BzRes is different.

We have also analyzed the variability in IC₅₀values based on the overall potency of inhibition. The 21 compounds that were inhibitory to three or more substrates were placed into three groups based on the overall IC₅₀ value (Table 1) and the average CV compared among groups. The most potent inhibitors (cisapride, clotrimazole, itraconazole, ketoconazole, miconazole, nicardipine, and mibefradil), the least potent inhibitors (astemizole, carbamazepine, cyclosporin A, erythromycin, haloperidol, propofol, and quinidine), and the inhibitors of intermediate potency (ethynylestradiol, midazolam, nifedipine, nimodipine, troleandomycin, terfenadine, and verapamil) gave mean CV values of 0.96, 0.89, and 1.00, respectively. Therefore, the variability in the responses among the substrates was not dependent on inhibitor potency.

Analysis of Inhibition Potency as a Function of Substrate Concentration.

BQ dealkylation demonstrates sigmoidal saturation kinetics (Hill coefficient of 1.4). In the two-substrate model when the substrate concentration = S₅₀, the inhibitor interacts with a heterogeneous population of enzyme with a single substrate present (ES) and enzyme with two substrates present (ESS) (Korzekwa et al., 1998). Therefore, the relationship between IC₅₀ and K_i is complex, and a simple relationship such as 2 ×K_i = IC₅₀ orK_i = IC₃₃ may not apply. Indeed, a fixed relationship between IC₅₀and K_i is unlikely because any given inhibitor may interact more strongly with either ES or ESS.

As substrate concentration is reduced below the S₅₀, ES should predominate over ESS. To estimate the interaction of the inhibitor with ES, we examined the relationship between BQ concentration and IC₅₀. Inhibition of BQ dealkylation by ketoconazole was examined at ten BQ concentrations spanning 2.5 to 100 μM. The results appear in Fig.3. As expected, the IC₅₀ decreases as BQ concentration decreases. Between 40 and 2.5 μM BQ, the IC₅₀ value decreases 1.8-fold. Below 10 μM BQ, an apparent plateau was reached. At 10 μM BQ, the signal to noise ratio in the assay was approximately 2, which was just adequate for reliable, routine quantitation of IC₅₀ values.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Effect of BQ concentration on ketoconazole IC₅₀ value.

Values are the mean of quadruplicates. Incubation time and enzyme concentration were as specified under Experimental Procedures.

To expand the analysis, seven additional, structurally diverse inhibitors (nifedipine, midazolam, cisapride, cyclosporin A, itraconazole, verapamil, and terfenadine) were examined using three BQ concentrations (10, 20, and 40 μM). For comparison purposes, these same inhibitors were tested using 10, 20, and 50 μM BFC and 0.25, 0.5, and 1 μM DBF. The results appear in Table5.

The IC₅₀ values of nifedipine, midazolam, and cisapride were essentially independent of BQ concentration. Itraconazole and ketoconazole showed a modest <2-fold reduction in IC₅₀ when BQ concentration decreased from 40 to 10 μM. In contrast, verapamil and cyclosporin A exhibited a marked 14- to 20-fold decrease in IC₅₀, while terfenadine exhibited a surprising increase in IC₅₀ of more than 5-fold. Surprisingly, substituting the IC₅₀ values for 10 μM BQ for either the IC₅₀ values or the IC₃₃ values reduces the correlation coefficients in Table 4, part B.

Relative to BQ, BFC and DBF were better behaved. Given that the BFC concentration is well below any apparent K_mand that IC₅₀ and K_ishould converge as substrate concentration is reduced below theK_m, a very modest dependence of IC₅₀ on BFC concentration was expected. Nifedipine, midazolam, cisapride, itraconazole, terfenadine, and ketoconazole IC₅₀ values were essentially independent of either BFC concentration. While the BFC IC₅₀ values for cyclosporin A and verapamil decreased with BFC concentration, the magnitude of the effects was about 3-fold. Given that the DBF concentration =K_m, the IC₅₀ = 2 × K_i; therefore, as DBF concentration was reduced below the K_m, the IC₅₀ should decrease 2-fold for competitive inhibitors. Indeed, for DBF, IC₅₀ values were unchanged or decreased for all inhibitors examined. The magnitude of the effect varied from 0.9- to 3.4-fold, with an average reduction in IC₅₀ value of 2.2-fold.

Discussion

CYP3A4 inhibition was measured using four fluorometric probe substrates. All substrates were benzyl ethers, but they varied in molecular weight from 235 (BQ) to 512 (DBF). Quantitative inhibition potency (IC₅₀) and qualitative response (inhibition, activation, and no effect) were observed to be substrate-dependent for CYP3A4. When IC₅₀ values are compared, BFC dealkylation was the most sensitive to inhibition, whereas BQ dealkylation was more than 5-fold less sensitive. BzRes and DBF dealkylations were of intermediate sensitivity. If the IC₃₃ values for the substrates that demonstrated saturation kinetics were compared with the IC₅₀values for BFC, the differences in overall sensitivity among the substrates was less than 2-fold.

BzRes dealkylation was observed to be the most sensitive to activation (10 of 27 compounds), whereas activation of BQ dealkylation was rarely observed and activation of DBF dealkylation was not observed. For several compounds, only DBF dealkylation was inhibited. These qualitative differences in response clearly contribute to the overall variability among substrates. However, activation potential fails to explain several of the most substantial differences in response observed. For example, troleandomycin, erythromycin, astemizole, cisapride, and ketoconazole exhibited a substrate-dependent range of IC₅₀ values between 21- and 46-fold in the absence of CYP3A4 activation.

For individual substrate/inhibitor pairs, the assays were found to be quite reproducible with within-day and between-days CV values of less than 0.22. Experiments were designed and performed to minimize experimental variables that may affect IC₅₀values. The final protein concentration was standardized, and each inhibitor was tested with all four substrates side by side. While protein concentration was standardized, the need to achieve an acceptable signal to noise ratio in the assay and the need to avoid substrate depletion resulted in the use of different enzyme concentrations and incubation times. Differences in the amount of CYP3A4-mediated inhibitor metabolism (depletion or formation of an inhibitory metabolite) may account for part of the quantitative differences. However, the experimental design and several trends discussed under Results argue against a large systematic bias caused by inhibitor depletion. For example, the enzyme concentrations were low (1–15 nM). In contrast, 1 mg/ml liver microsomal protein contains on average 100 nM CYP3A4 (Shimada et al., 1994) in addition to other enzymes that may contribute to inhibitor depletion. In addition, the incubation time for DBF was one half of the incubation time for BFC (same enzyme concentration), yet the mean DBF IC₅₀ and IC₃₃ values were higher. BQ had a 1.5-fold lower enzyme concentration than BzRes (same incubation time), yet the mean IC₅₀ was higher. However, these overall trends do not eliminate the possibility of substrate depletion affecting the results for some specific compounds. In addition, tight-binding inhibitors may interact with the enzyme stoichiometrically. This appears to be the case in the present study with clotrimazole. In this instance, substrate-dependent responses are certainly affected by the concentration of the enzyme (Gibbs et al., 1999).

Substrate-dependent effects were comparable regardless of the potency of the inhibition. In spite of these effects, all four substrates identified cisapride, clotrimazole, itraconazole, ketoconazole, mibefradil, and miconazole as submicromolar inhibitors. Therefore, any of the four probe substrates appear capable of identifying compounds that will have broad-based effects on CYP3A4. In addition, within the two small analog series contained in the set of 27 compounds, the rank order of potency was similar regardless of probe substrate. For the three calcium channel blockers, the rank order of potency of inhibition was nicardipine > nimodipine > nifedipine for all four substrates. For the four azole antifungals, the rank order of potency was ketoconazole ≃ clotrimazole > miconazole ≃ itraconazole. These observations suggest that a small number of CYP3A4 substrates can be used to choose the compounds within a series with the lowest CYP3A4-inhibition potential.

Substrate-dependent effects on CYP3A4-inhibition potential have been observed previously. For example, the flavonoid α-naphthoflavone, although well known to activate CYP3A4 (Schwab et al., 1988), may also inhibit (Yun et al., 1992) the enzyme, depending on the CYP3A4 substrate. We found a similar pattern of substrate dependence with α-naphthoflavone in the present study (e.g., activation or inhibition). Interestingly, the substrate-dependent pattern of response of α-naphthoflavone was essentially identical to that exhibited by both progesterone and testosterone. Wang et al. (2000) have reported that the CYP3A4 (cDNA-expressed and human liver microsomal) inhibition as measured by terfenadine, testosterone, midazolam, and nifedipine metabolism was substrate-dependent. As with the present study, partial inhibition was often observed, and testosterone and α-naphthoflavone were observed to activate some probe reactions while inhibiting others.

Kenworthy et al. (1999) examined the effects of a 30 μM concentration of 34 compounds on 10 different substrates of CYP3A4 (cDNA-expressed). Correlation analysis indicated that seven of the substrates could be categorized into distinct groups where the extent of inhibition was correlated. This analysis was applied to the data from the fluorometric substrates and a data set with the same inhibitors tested with three traditional substrates (MDZ, NF, and TS).

Fourteen inhibitors and four substrates (BzRes, MDZ, NF, and TS) were in common in both the present study and in the study byKenworthy et al. (1999). However, there are minor differences. For example, Kenworthy et al. (1999) used 30 μM inhibitor (versus 10 μM in the present study), and the system to express the CYP3A4 enzyme was different [lymphoblastoid cells in Kenworthy et al. (1999) and baculovirus/insect cells in the present study].

For MDZ, NF, and TS, our results correlated well with those ofKenworthy et al. (1999). For the 14 inhibitors in common, the correlation coefficients for the percent inhibition values in Kenworthy et al. (1999) Table 1 and our data in Table 3 were 0.82, 0.85, and 0.84 for MDZ, NF, and TS, respectively. Many of the differences appear to be as expected due to the 3-fold difference in inhibitor concentration. Therefore, the agreement between the two studies is likely higher than indicated by the calculated correlation coefficients. Moreover, our correlation coefficients between these substrate pairs for the entire set of inhibitors were similar to those reported by Kenworthy et al. (1999) [MDZ/TS, MDZ/NF, and TS/NF of 0.86, 0.78, and 0.78 versus 0.83, 0.63, and 0.76 in Kenworthy et al. (1999), respectively]. We also observed that BzRes was found to be quantitatively and qualitatively the least concordant when compared with the other substrates tested. However, our correlation coefficients with BzRes were higher than those of Kenworthy et al. (1999). This may be due to differences in the inhibitor set.

BQ, BFC, and DBF correlated well with each other and also strongly and equally well to MDZ and TS. Our data do not clearly place any of these substrates in either the MDZ or TS groupings as described by Kenworthy et al. (1999). However, the response of these fluorometric probes appears closely related to traditional probes such as MDZ and TS and more distantly related to the response of NF.

In the present study, several other correlation analyses indicated that the response of BQ, DBF, and BFC tended to track together, whereas BzRes was substantially different. Note that in our regression analyses, IC₅₀ (or IC₃₃) values for inhibitor/substrate pairs that show no inhibition or activation were unavailable and thus were not considered. Therefore, overall concordance among the substrates is less than the correlation coefficients might suggest.

Another approach to address the agreement among the substrates is to exclude the data for individual substrates and examine the effect on the range in IC₅₀ values. Excluding BzRes from the data set reduced the mean range in IC₅₀values from 29- to 13-fold. In contrast, excluding BQ, DBF, or BFC reduced the mean ranges from 29- to 16-, 26-, or 25-fold, respectively. The larger effect on the range upon eliminating the BzRes data provides another indication that the BzRes responses are different from the other substrates. For comparison purposes, the mean ranges for the independent IC₅₀ determinations were 1.79-, 1.60-, 1.42-, and 1.30-fold for BzRes, BQ, DBF, and BFC, respectively.

Atypical CYP3A4 metabolite formation and inhibition kinetics (e.g., activation, partial inhibition, mutual inhibition, sigmoidal kinetics, and substrate inhibition) has been explained by using a multiple conformer model (Koley et al., 1995), a cooperativity model (Ueng et al., 1997), and a two-substrate model (Korzekwa et al., 1998). We have elected to interpret some of our results in the context of this latter model because it represents a simple and attractive model to explain the response of CYP3A4. However, these models are not mutually exclusive, and the other models can not be excluded based on our observations. The observation that a CYP3A4 mutant designed to possess a smaller active site exhibits hyperbolic kinetics typical of an enzyme accommodating a single substrate supports the two-substrate model (Harlow and Halpert, 1998). It may be quite informative to use this mutant CYP3A4 in a study identical to the present one.

The substrate-dependent effects on CYP3A4 inhibition apply to both drug molecules (Kenworthy et al., 1999; Wang et al., 2000) and model compounds with a fluorometric endpoint. The four fluorometric substrates used in the present study each have specific advantages and disadvantages. From a purely practical point of view, the higher excitation and emission wavelength for BzRes and DBF are less susceptible to assay interference caused by fluorescence or quenching properties of the test compound. In addition, the lower enzyme and incubation time requirements for BFC and DBF mitigate the potential for inhibitor depletion. Based on the results of the present study, the results with BFC and DBF are easier to interpret than the results with BzRes or BQ. BzRes is more prone to demonstrate activation, and while this is clearly an indication that the test compound is interacting with the CYP3A4 active site, there is no current framework for interpreting significance of such a result to an in vivo effect. In some cases, activation occurs in a concentration range similar to that which causes inhibition of the metabolism of other substrates, and in some cases activation begins at lower concentrations. Our substrate concentration analyses (Table 5) indicate that the IC₅₀ values for some compounds are highly BQ concentration-dependent and can decrease or increase with decreasing BQ concentration. This has been interpreted as differences in affinity between the inhibitor and ES versus ESS. Our data can be interpreted that cyclosporin A and verapamil preferentially inhibit ES, whereas terfenadine preferentially inhibits ESS. A more modest substrate concentration dependence was also observed for cyclosporin A and verapamil with BFC as substrate. This surprising observation suggests that even though the rate of metabolite formation was linear with respect to substrate concentration, which implies a single ES species, the situation within the active site may be more complex. Finally, partial inhibition is more commonly observed with BQ and BzRes. In contrast, the inhibition curves for BFC and DBF were normal (Fig. 2).

In the aggregate, BFC and DBF appear to be the preferred fluorometric substrates, whereas BzRes and BQ seem more suited for a secondary role (for example, acquiring more information for compounds with a profile like α-naphthoflavone in which DBF is inhibited and BFC is activated). The tendency for BzRes and BQ to give partial inhibition indicates that these substrates are not the most suitable for testing inhibition potential at a single inhibitor concentration. The observations of substrate and substrate concentration-dependent effects for CYP3A4 also indicate that follow-up studies of CYP3A4 inhibition with likely comedications are necessary. In addition, given the large number of drugs that are substrates for CYP3A4, a comprehensive analysis cannot be practical for all compounds.

The unexpectedly large changes in IC₅₀ values as substrate concentrations were reduced below the apparentK_m (Table 5) indicates that simple assumptions for relating IC₅₀ toK_i do not generally apply to all CYP3A4 substrates. For some substrates, the common practice of testing for inhibition using a substrate concentration only at theK_m or S₅₀ may be undesirable if circulating drug concentrations are well below the apparent K_m (or S₅₀). A more appropriate and conservative approach would be to use a substrate concentration below the apparentK_m (but within the limits of assay sensitivity).

Eight-concentration point curves for 27 compounds were generated with four substrates in duplicate with two independent experiments on separate days. The data generation took one person about one week. Analysis in 96-well plates required a total of 60 min of instrumentation use. The efficiency of data acquisition is several orders of magnitude greater than for HPLC-based systems. Moreover, the instrumentation requirements are also considerably less expensive. These two features are of considerable advantage when screening a large chemical series for the potential for cytochrome P450 inhibition. This advantage is compounded by the clear need to use multiple CYP3A4 substrates.

This report demonstrates that quantitative and qualitative inhibition parameters are substrate-dependent for CYP3A4. This fact must be taken into account when interpreting the results for cytochrome P450 inhibition testing. The lack of CYP3A4 inhibition with a single probe substrate may be applicable to other probe substrates, but this is not always the case. Clearly, a testing strategy using several probe substrates is prudent. If a single fluorometric substrate must be used, BFC appears to be the most conservative choice.