Abstract
Traditional cytochrome P450 (P450) based drug-drug interaction (DDI) predictions are based on the ratio of an inhibitor's physiological concentration [I] and its inhibition constant Ki. Determining [I] at the enzymatic site, although critical for predicting clinical DDIs, remains a technical challenge. In our previous study, a novel approach using cryopreserved human hepatocytes suspended in human plasma was investigated to mimic the in vivo concentration of ketoconazole at the enzymatic site (Lu et al., 2007), effectively eliminating the estimation of the elusive [I] value. P450 inhibition in this system appears to model that in vivo. Using the ketoconazole inhibition information in a human hepatocyte-plasma suspension together with quantitative P450 phenotypic information, we successfully predicted the pharmacokinetic DDIs for a small set of drugs, such as theophylline, tolbutamide, omeprazole, desipramine, midazolam, loratadine, cyclosporine, and alprazolam, as well as an investigational compound. For the applicability of this model on a wider scale the in vitro-in vivo correlation data set needed to be expanded. However, for most drugs in the literature there is not enough quantitative information on the involvement of individual P450s to predict DDIs retrospectively. To facilitate that, in this study we determined quantitative P450 phenotyping for seven marketed drugs: budesonide, buprenorphine, loratadine, sirolimus, tacrolimus, docetaxel, and methylprednisolone. Augmentation of the new data set with the one generated previously produced broader a database that provided further support for the wider applicability of this approach using ketoconazole as a potent CYP3A inhibitor. This application is predicted to be equally effective with other P450 inhibitors that are not substrates of efflux pumps.
The prediction of pharmacokinetic drug-drug interactions (DDIs) for humans in the recent past has relied heavily on the ratio [I]/Ki, use of which has shown some success. However, to date its broader applicability has not been demonstrated, mainly because of the elusive parameter [I], which signifies the free inhibitor concentration at the enzyme site. Because the value cannot be determined directly, scientists have resorted to finding the next best parameter to estimate that concentration, as described in the previous communication (Lu et al., 2007). But the applicability of one method over another generally leads to not-so-useful predictions. Thus, use of the [I]/Ki ratio to predict drug-drug interaction remains a challenge.
We recently proposed a new method for DDI prediction that enabled us to avoid the dependence on [I]/Ki ratio (Lu et al., 2007). In our method, various concentrations of the inhibitor ketoconazole were incubated in human hepatocytes that were suspended in human plasma to construct an inhibition titration curve. After equilibrium, the major P450 activities remaining were measured using prototypical substrates. It was assumed that if an extracellular (plasma) concentration of that inhibitor in vitro equaled that of the inhibitor in vivo, the intracellular concentration of that inhibitor in vitro would also be comparable with the intracellular concentration in vivo. Furthermore, if the intracellular concentrations are comparable, the enzyme activity remaining measured in vitro would represent that measured in vivo, therefore, circumventing the need to estimate [I] and determine the Ki of that inhibitor. The remaining enzyme activity information obtained from in vitro studies could then be used to aid the DDI prediction based on the knowledge of relative enzyme contributions to the total drug clearance (Rostami-Hodjegan and Tucker, 2004; Ito et al., 2005; Obach et al., 2006; Lu et al., 2007). Similar experimental conditions were reported by several researchers for predicting hepatic clearance (Shibata et al., 2002; Bachmann et al., 2003, Blanchard et al., 2006, Skaggs et al., 2006).
P450 reactive phenotyping, a quantitative measurement of the relative contributions of each P450 to the overall metabolism of a drug, is another factor that we used to predict DDIs. To prove wider applicability of the approach we have extended the correlation to several other drugs for which clinical interaction data were available. However, the corresponding P450 phenotype data were not found in the literature to help build the correlation. The group of drugs, which showed significant clinical DDIs with ketoconazole, was selected from the University of Washington DDI database (2004, http://www.druginteractioninfo.org). All of the selected drugs are reported to show at least 50% increases in area under the curve (AUC) when coadministered with ketoconazole. This report provides in vitro prediction of AUC change in humans in the presence of ketoconazole and how that correlates to the in vivo observations.
Materials and Methods
Reagents. Pooled human liver microsomes from 50 donors were purchased from XenoTech, LLC (Kansas City, KS). 4-Hydroxytolbutamide, 4-hydroxymephenytoin, 1′-hydroxymidazolam, (S)-(+)-(N)-(3)-benzylnirvanol (benzylnirvanol), and azamulin were purchased from BD Gentest (Woburn, MA). Phenacetin, acetaminophen, tolbutamide, dextromethorphan, dextrorphan, testosterone, 6β-hydroxytestosterone, furafylline, sulfaphenazole, quinidine, NADPH, MgCl2, and compounds in third column of Table 1 were purchased from Sigma-Aldrich (St. Louis, MO). S-Mephenytoin was purchased from BIOMOL Research Laboratories, Inc. (Plymouth, PA).
Reactive Phenotyping Using P450 Selective Chemical Inhibitors. In 96-well plates, microsomes (25 μl, final concentration of 0.5 mg/ml in 0.1 M potassium phosphate buffer, pH 7.4) were prewarmed, in duplicate, with 25 μl of test compounds (prepared in 2% acetonitrile-0.1 M potassium phosphate buffer (v/v, final concentration of 2 μM) and 25 μl of P450 selective inhibitors (prepared in 0.4% dimethylsulfoxide-2% acetonitrile-0.1 M potassium phosphate buffer (v/v/v, predetermined concentrations) for 5 min at 37°C. The reactions were initiated by the addition of 25 μl of NADPH-MgCl2 in 0.1 M potassium phosphate buffer (final concentration of 2 and 3 mM, respectively) and incubated for 15 min. The addition of 0.1% dimethylsulfoxide and 1% acetonitrile is to increase compound solubility with minimal effect on P450 activity.
The reactions were terminated by the addition of 100 μl of acetonitrile containing 1 μM of carbutamide as the internal standard. After storage for 30 min in a refrigerator, the sample plates were centrifuged at 3000g for 10 min, and the supernatants were analyzed using liquid chromatography-mass spectrometry. In a parallel study, P450-selective substrates were incubated with microsomes in the presence of P450 selective inhibitors to demonstrate optimal inhibition conditions, i.e., a high degree of inhibition with a high degree of selectivity. This information was then used to correct the partial inhibition and cross-reactivity of these inhibitors on the metabolism of the test compound. The final concentrations of respective P450-selective inhibitors and probe substrates were 100 μM furafylline and 30 μM phenacetin for CYP1A2, 5 μM sulfaphenazole and 150 μM tolbutamide for CYP2C9, 20 μM benzylnirvanol and 100 μM S-mephenytoin for CYP2C19, 5 μM quinidine and 8 μM dextromethorphan for CYP2D6, and 2 μM azamulin and 50 μM testosterone for CYP3A4. The percentage of metabolic activity remaining was calculated by comparing the parent compound remaining (test compounds) or metabolite formation (P450 substrates) in the presence of inhibitors to their vehicle controls.
The liquid chromatography-tandem mass spectrometry system used to determine the P450 substrate metabolites and test compound remaining consisted of an Agilent 1100 high-performance liquid chromatograph, a CTC PAL autosampler (LEAP Technologies, Carrboro, NC), and a SCIEX API 4000 detector (Applied Biosystems, Framingham, MA). Metabolite separation was achieved on a Phenomenex Synergi C18 column (75 × 4.6 mm) with a gradient consisting of 0.1% formic acid/water (mobile phase A) and 0.1% formic acid-acetonitrile (mobile phase B) at a flow rate of 1.0 ml/min. Specifically, 5% of mobile phase B was applied for 0.5 min after injection and increased linearly to 95% B from 0.5 to 3.5 min. Mobile phase B was held at 95% from 3.5 to 3.6 min, and the column was reequilibrated to 5% B from 3.6 to 5.0 min. A positive ion spray in the multiple-reaction monitoring mode was applied with predetermined parent/product mass transition ion pairs for P450 probe substrate metabolites and test compounds.
Calculation of AUC Changes from in Vitro Data. The P450 contents in the gut are a small fraction of that in the liver (Obach et al., 2006), and the contribution of metabolism by gut to the overall metabolism in humans is compound- and dose regimen-dependent. For oral drugs that are subject to gut metabolism the effect becomes significant if those drugs are highly subject to CYP3A clearance and have low permeability or are substrates of efflux pumps. However the information on Fg′/Fg factor is not available for the compounds of interest (Rostami-Hodjegan and Tucker, 2004; Ito et al., 2005; Galetin et al., 2006). Thus, this factor was not included in our calculation. The predicted AUC changes were calculated using the method reported previously (Lu et al., 2007) for a clinical ketoconazole dosing regimen of 200 mg b.i.d. For other doses, the corresponding ketoconazole concentration was used to calculate the enzyme activity remaining in the hepatocyte incubation, assuming linear pharmacokinetics.
Results
Table 1 shows a list of compounds that have been reported to have clinical DDIs with ketoconazole, causing at least 50% increases in AUC (University of Washington DDI database, 2004, http://www.druginteractioninfo.org). Among the 28 compounds listed in the first column of Table 1, 15 of them (third column) are available from commercial sources, such as Sigma-Aldrich. These 15 compounds were tested in microsomal incubation (15 min with 0.5 mg/ml microsomal protein) for parent compound remaining in a manner similar to that described earlier (Uttamsingh et al., 2005). Only those compounds that showed significant metabolism, e.g., <60% parent remaining, were further studied to determine the quantitative reactive phenotyping (fourth column in Table 1).
The reactive phenotyping results (fm) for five major P450s (1A2, 2C9, 2C19, 2D6, and 3A4) are presented in Table 2 for budesonide, buprenorphine, docetaxel, loratadine, methylprednisolone, sirolimus, and tacrolimus. The activities remaining (fA) for these five P450s in the presence of various concentrations of ketoconazole were determined previously (Table 3) (Lu et al., 2007). For example, ketoconazole was added into hepatocytes suspended in plasma and allowed a short period of time to equilibrate. P450 probe substrates were then added to measure the activity remaining. fA was calculated by comparing the peak area ratio over internal standard from the samples with ketoconazole to the vehicle control samples. By using the drug-drug interaction model described in our previous work (eq. 1 in Lu et al., 2007) and the fm and ketoconazole Cmax listed in Table 4, the predicted change of AUCs for these seven compounds are presented in Table 5 along with other compounds studied previously (Lu et al., 2007). where fm, hep is the fraction of clearance by hepatic metabolism, fm, P450 is the relative contribution of an individual P450 to the total metabolism of the compound (without inhibitor), and fA is the fraction of enzyme activity remaining in the presence of the inhibitor. fother is the fraction of clearance by other routes such as renal or biliary excretion (fm + fother = 1).
Figure 1 is a correlation plot of the predicted AUC changes for the entire set of compounds studied versus their clinically observed AUC changes. A correlation coefficient (r2) of 0.966 and a slope close to unity (0.931) suggested a good prediction of clinical DDIs from the in vitro data.
Discussion
There are numerous literature reports on using in vitro data to predict clinical drug-drug interactions. Almost all of them are based on the ratio of inhibitor's physiological concentration [I] over the inhibitor's inhibition constant Ki ([I]/Ki) (Blanchard et al., 2004; Cook et al., 2004; Ito et al., 2004; Bachmann, 2006; Obach et al., 2006). Controversies exist regarding the accuracy of both of these parameters. Theoretically, Ki is an absolute value dependent only on the inhibitor-enzyme affinity, but it has to be determined in an experimental setting. Thus, the Ki value determined or so-called apparent Ki value could be influenced by the following factors: 1) the amount of protein used in the incubation (i.e., the more protein used, the higher the Ki value appeared to be); 2) the substrates used to determine the enzyme activity (different substrates often generate different Ki values); and 3) the enzyme system used (e.g., recombinant enzymes could generate different values compared with human liver microsomes). Thus, the Ki values for ketoconazole in the literature vary from 0.015 to 8 μM (Thummel and Wilkinson, 1998), a range of 500-fold. The Ki values for ketoconazole listed in the recently published Food and Drug Administration Drug-Drug Interaction Draft Guidance (2006, http://www.fda.gov/cder/guidance/6695dft.htm) also show a 50-fold difference range, from 0.0037 to 0.18 μM. In addition to the difficulty in determining a reliable value of Ki, the determination of the physiological inhibitor concentration [I] is not practical; therefore, various methods have been used to estimate it (Blanchard et al., 2004; Cook et al., 2004; Ito et al., 2004; Bachmann, 2006; Obach et al., 2006). Thus, using the ratio of these two parameters, [I]/Ki, to predict drug-drug interaction may be expected, more often than not, to yield results that are not quantitatively representative of the in vivo situation.
We have proposed a new hepatocyte suspension in plasma model that, combined with reactive phenotyping information (such as fm), gave predictions for a set of compounds with the potent P4503A4 inhibitor ketoconazole with results that were consistent with the in vivo findings. In our method, once the enzyme activity remaining (fA) in the presence of the inhibitor is determined, that information could technically be used for most substrates. The second set of parameters needed to make the DDI prediction is the reactive phenotyping of the compound (e.g., fm). There is limited information in the literature on quantitative reactive phenotyping of drugs (Soars et al., 2003).
Determining the inhibition of a substrate's metabolism could be done by either monitoring metabolite formation or parent compound disappearance. In the drug discovery stage, most metabolite standards are generally not available; thus, the parent compound disappearance method is usually the main choice, unless radiolabeled materials are available early on. In addition, the parent compound disappearance method uses a sub-Km drug concentration, which is relevant to the physiological concentration. At such a concentration, some allosteric effects, such as substrate inhibition, are not often seen. Most importantly, the parent disappearance method allows one to capture the total metabolism rather than focus on a major pathway in the metabolite formation method. One disadvantage of the parent compound disappearance method, however, is that it has to have enough metabolism, so the effect of an inhibitor on the metabolism could be distinguished from any variations in the experiment. For the above reasons, only compounds (total of seven) with greater that 40% metabolism in a 15-min microsomal incubation were selected and were phenotyped for P450 reactivity. The study was limited to five major P450s—CYP1A2, CYP2C9, CYP2C19, CYP2D6, and CYP3A4 — because they are responsible for the clearance of more than 90% of marketed drugs. Among the seven compounds studied, loratadine, sirolimus, docetaxel, and methylprednisolone had total percent contributions from five P450s exceed 100%, probably due to experimental variations. The contributions were then normalized to 100%, assuming metabolism was only via the five major P450s. Two other compounds, budesonide and buprenorphine, had total contributions from five P450s close to 100%; thus, no adjustments were made. A few percentages short from 100% on a compound may be attributed to data variation or metabolism by enzymes other than the major five. However, the seventh compound, tacrolimus, showed 47, 1, and 8% contributions from CYP3A4, CYP2D6, and CYP2C9, respectively, and no contribution from the other two P450s, with the total value being well below the 100% target. Based on these data, the predicted AUC change for tacrolimus would be 1.95-fold, which is actually close to the observed clinical change of 2.39-fold. Tacrolimus is reported to be a substrate of CYP3A with a preference toward CYP3A5 (Thervet et al., 2003; Op den Buijsch et al., 2007). The CYP3A inhibitor, azamulin, used in this study was reported to have a much higher inhibition potency toward CYP3A4 compared with CYP3A5 (∼15-fold) (Stresser et al., 2004). Thus, we might not have the optimal inhibitor concentration to inhibit the majority of CYP3A5 activity. In other words, the prototypical substrate midazolam used to correct the partial and cross inhibition (Lu et al., 2007), in this case, is not a representative substrate for tacrolimus because tacrolimus is a preferred CYP3A5 substrate. Azamulin inhibits the metabolism of tacrolimus only partially, because azamulin has low potency toward CYP3A5 and tacrolimus is a CYP3A5 substrate. Similarly, ketoconazole is also shown to have higher potency toward CYP3A4 compared with CYP3A5 (∼5-fold) (Stresser et al., 2004). The fact that the 400 mg q.d. dose of ketoconazole only causes a 2.39-fold increase in the AUC of tacrolimus suggests an incomplete inhibition of the metabolizing enzymes, such as CYP3A5. It is also known that at this dose CYP3A4 is inhibited completely (Chien et al., 2006; Obach et al., 2006). Because the concentration of ketoconazole in our in vitro study was chosen to mimic that in the clinic, the incomplete inhibition of CYP3A5 should happen both in vitro and in vivo. This is evident by the fact that the predicted AUC change in vitro is comparable with that in vivo.
Glucuronidation is a major pathway besides CYP3A4-mediated oxidation for buprenorphine clearance. Other minor pathways involving, for example, CYP2C8 and CYP2D6, have also been reported (Zhang et al., 2003; Picard et al., 2005). In this study, metabolism via glucuronidation is not captured because the study was focused only on five major P450s, and UDP-glucuronosyltransferase cofactor was not used. Therefore, our phenotyping data are specific for P450-mediated pathways, and the method overpredicted the AUC increase. For this reason, the predicted AUC increase value was not included in the regression analysis for Fig. 1.
CYP3A4 was reported to contribute 60% to the metabolism of loratadine (Galetin et al., 2006). In another study, CYP3A4 and CYP2D6 were reported to contribute 53 and 20% of the metabolism of loratadine, respectively (Yumibe et al., 1996). The rest of the clearance routes have not been documented. In our previous exercise (Lu et al., 2007), it was assumed that ketoconazole did not affect the undocumented clearance routes and the CYP3A4 contribution was 60%. The predicted AUC change was 2.48-fold (Table 5). To avoid interlaboratory differences, in this study we determined the phenotyping for loratadine to be 41, 24, 18, and 16% for CYP3A4, CYP2D6, CYP2C9, and CYP1A2, respectively. The predicted AUC change with the 200-mg b.i.d. dose of ketoconazole is 1.92-fold. Because this latter value is based on the more complete set of P450s affected, it was the one used for correlation purposes. Nonetheless, use of either the average of the two predicted values (2.20) or the value (1.92) considering all major P450s did not change the slope (0.931 and 0.928, respectively) or r2 (0.966 versus 0.967, respectively) values obtained from the correlations.
Listed in Table 5 are the predicted AUC changes and the observed clinical values, which were adapted from Obach et al. (2006) and the DDI data from the University of Washington (2004, http://www.druginteractioninfo.org). Notably, these predictions compare very closely to the observed clinical values (r2 = 0.966 and slope = 0.931) (Fig. 1). For all of the compounds from this study, renal clearance plays a minor role (≤5%) (Table 5). Thus, the contribution of renal clearance was considered to be insignificant.
In summary, our new DDI prediction approach involving use of two in vitro determinations under the linear kinetics, reactive phenotyping and the P450 cross-reactivity of an inhibitor at close to the physiological condition (hepatocyte incubation in human plasma), provided projections that are close to reality. Ketoconazole is used in this study as an example. However, this approach is considered to be applicable to most inhibitors of interest, except those that are efflux pump substrates, which can be pumped out of hepatocytes, leading to reduced hepatocyte concentration and hence reduced potency. Thus, this approach provides a simple way of predicting the DDI early in the preclinical development stage using parameters that can be readily determined, as opposed to the seldom working [I]/Ki approach, in which [I] value determination remains elusive and ambiguous.
Acknowledgments
The authors thank Drs. Gerald T Miwa and Liang-Shang Gan for their valuable discussions.
Footnotes
-
Article, publication date, and citation information can be found at http://dmd.aspetjournals.org.
-
doi:10.1124/dmd.107.018796.
-
ABBREVIATIONS: DDI, drug-drug interactions; P450, cytochrome P450; AUC, area under the curve; CYP3A4, CYP3A4/5; fm, relative contribution of metabolizing enzymes toward the total clearance of a compound; fA, fraction of activity of a given enzyme remaining in the presence of inhibitor.
- Received September 7, 2007.
- Accepted March 31, 2008.
- The American Society for Pharmacology and Experimental Therapeutics