Abstract
Cytochrome P450 3A4 (CYP3A4) is the major enzyme responsible for phase I drug metabolism of many anticancer agents. It is also a major route for metabolism of many drugs used by patients to treat the symptoms caused by cancer and its treatment as well as their other illnesses, for example, cardiovascular disease. To assess the ability to inhibit CYP3A4 of drugs most commonly used by our patients during cancer therapy, we have made in silico predictions based on the crystal structures of CYP3A4. From this set of 33 common comedicated drugs, 10 were predicted to be inhibitors of CYP3A4, with the antidiarrheal drug loperamide predicted to be the most potent. There was significant correlation (r2 = 0.75–0.66) between predicted affinity and our measured IC50 values, and loperamide was confirmed as a potent inhibitor (IC50 of 0.050 ± 0.006 μM). Active site docking studies predicted an orientation of loperamide consistent with formation of the major (N-demethylated) metabolite, where it interacts with the phenylalanine cluster and Arg-212 and Glu-374; experimental evidence for the latter interaction comes from the ∼12-fold increase in KM for loperamide observed for the Glu-374-Gln mutant. The commonly prescribed drugs loperamide, amitriptyline, diltiazem, domperidone, lansoprazole, omeprazole, and simvastatin were identified by our in silico and in vitro screens as relatively potent inhibitors of CYP3A4 that have the potential to interact with cytotoxic agents to cause adverse effects, highlighting the likelihood of drug-drug interactions affecting chemotherapy treatment.
Cytochrome P450 3A4 (CYP3A4) is the major xenobiotic metabolizing enzyme in humans. A broad specificity coupled with high levels of expression in the liver means it is responsible for the metabolism of more than half of all prescribed drugs (Guengerich, 1997). When patients receive several medications concurrently, unwanted and life-threatening effects can result from the competition for the same drug-metabolizing enzyme affecting the blood levels of the competing drugs. Cancer patients would seem to be significantly at risk in this respect, because CYP3A4 metabolizes a large number of anticancer drugs and patients are generally prescribed other medications to relieve symptoms (e.g., analgesics) and side effects (e.g., antiemetics and antidiarrheals) and to treat comorbidities. The anticancer drugs metabolized by CYP3A4 include docetaxel (Marre et al., 1996), cyclophosphamide (Chang et al., 1993), ifosfamide (Walker et al., 1994), etoposide (Kawashiro et al., 1998), tamoxifen (Crewe et al., 1997), irinotecan (Santos et al., 2000), vinblastine (Zhou-Pan et al., 1993), and vinorelbine (Kajita et al., 2000). Although there is a marked interindividual variation of pharmacokinetic parameters between patients (Evans and Relling, 1999), and such variation in patient response is often attributed to polymorphism in P450 genes, CYP3A4 is an exception because only a small percentage of the variation in activity can be attributed to genotype (Lamba et al., 2002a,b). Interactions with comedicated compounds are therefore likely to be particularly important in explaining variations in anticancer drug pharmacokinetics and side effects. If a patient experiences significant toxicity during chemotherapy, the clinician will usually reduce the dose of the cytotoxic drug, reducing the anticancer effect. A more appropriate action might be to substitute a different comedication that will not interact with the therapy and so maintain dose intensity of the cytotoxic drug. Thus, identifying potential drug-drug interactions involving CYP3A4 is important for improving the treatment of cancer. The development of methods to predict such interactions could lead to the administration of more effective, less toxic drug regimes.
In recent years, in silico methods have proved to be a useful tool for predicting the binding properties of ligands to mammalian cytochromes P450. Studies (for a review, see Ekins et al., 2003) based on quantitative structure-activity relationships and pharmacophore models have generated useful information on ligand binding of CYP3A substrates (Ekins et al., 1999b) and CYP3A4 inhibitors (Ekins et al., 1999a). These methods are relatively fast and have been useful in determining important features such as autoactivation of CYP3A4 (Ekins et al., 2003). We have used an integrated hypothesis-driven structure-based approach to study CYP2D6 and have been able to identify residues that play a key role in metabolism (Smith et al., 1998; Kirton et al., 2002; Paine et al., 2003; Flanagan et al., 2004), reproduced the binding orientation and affinity of ligands in the active site (Kemp et al., 2004), and discriminated between tightly and weakly binding compounds (Kemp et al., 2004).
In this article, we now describe the use of our structure-based in silico approach, which we have used effectively with CYP2D6, to identify likely drug interactions that would inhibit CYP3A4. We have screened the set of 33 drugs commonly used by our patients with cancer using two independently determined crystal structures of the enzyme (Williams et al., 2004; Yano et al., 2004). This approach correctly predicted that loperamide had the highest affinity for CYP3A4 of all the compounds we tested. This prompted us to investigate its mode of binding. Site-directed mutagenesis supported a predicted key interaction between loperamide and CYP3A4. The relative affinities predicted in silico correlated well with the experimental IC50 values for the 15 comedication compounds in the set that significantly inhibited the enzyme. Such inhibition may have unwanted effects on the efficacy and toxicity of anticancer agents using the same pathway, and this merits further study.
Materials and Methods
Molecular Docking. We chose a set of 33 drugs based on an analysis of the medications taken by 100 newly diagnosed patients undergoing chemotherapy for their lung cancer (Table 1). Docking studies for these drugs were carried out as described previously (Kemp et al., 2004). In brief, the program GOLDv2.2 (Jones et al., 1997) was used with the ChemScore (Eldridge et al., 1997; Verdonk et al., 2003) fitness function to generate 10 solutions for each ligand, and the dockings were ranked according to the value of the ChemScore fitness function; only the best ranked solution for each ligand was included in further analysis. Dockings were performed into the two available crystal structures of ligand-free CYP3A4 [Protein Data Bank (Berman et al., 2000) accession codes 1w0e (Williams et al., 2004) and 1tqn (Yano et al., 2004)]. The two structures are closely similar with the exception of the region of SRS2 (Gotoh, 1992), where the side chain of Arg-212 is oriented either toward (1tqn) or away from (1w0e) the heme iron. To assess the utility of our approach over the use of relatively fast and simple ligand-based descriptors, we determined the Slog P value, number of hydrogen bond donors, number of hydrogen bond acceptors, and molecular weight using MOE (Chemical Computing Group, Montreal, QC, Canada).
Chemicals. The test compounds 5,5-diphenylhydantoin, allopurinol, amitriptyline, aspirin, atenolol, caffeine, cefuroxime, citalopram, dexamethasone, diclofenac, diltiazem, domperidone, fluoxetine, frusemide, gabapentin, glucosamine, ibuprofen, lansoprazole, loperamide, lorazepam, metformin, metoclopramide, metronidazole, omeprazole, ondansetron, oxazepam, paracetamol, prednisolone, ranitidine, simvastatin, theophylline, and R-warfarin were purchased from Sigma-Aldrich (Dorset, UK). Tramadol hydrochloride was purchased from Fluka BioChemika (Poole, UK). BFC was obtained from Ultrafine Chemicals (Manchester, UK). All other reagents used were of the highest available quality.
Mutagenesis and Coexpression of the P450s and P450 Reductase inEscherichia coli. The isolation of the cDNAs and construction of expression plasmids ompA CYP3A4(His6) (pB84) and pJR7 [human NADPH cytochrome P450 oxidoreductase (CPR)] have been described elsewhere (Pritchard et al., 1997). Site-directed mutagenesis was performed using the single-stranded DNA template method (Kunkel et al., 1987), using pB84 as a template, the dut–ung–E. coli strain CJ236, and the oligonucleotide 5′-TTTGCAGACCCTCCTAAGTCTCATAGC-3′ for E374Q. The presence of the desired mutations was confirmed by DNA sequencing. Coexpression of wild-type CYP3A4 or the E374Q mutant with CPR was carried out as previously described (Pritchard et al., 1997; Smith et al., 1998; Paine et al., 2003; Flanagan et al., 2004), where wild-type CYP3A4 and E374Q gave average yields of 250 and 150 nmol P450/l bacterial culture, respectively. CPR activity was estimated by NADPH-dependent cytochrome c reduction (Strobel and Dignam, 1978).
Single-Point Screen and IC50 Determinations. Assays were performed in 96-well microtiter plates using the fluorogenic substrate BFC in a final volume of 200 μl. Compounds dissolved in methanol or H2O (Table 1) were added to a final volume of 100 μl of 2× enzyme/substrate stock solution (0.1 pmol/μl P450 in 50 μM Hepes and 30 mM MgCl2, pH 7.4, with 50 μM BFC). A solvent control was included to correct for any solvent effects across the dilution range. Plates were then preincubated for 3 min at 37°C, and the enzyme reaction was initiated by the addition of a 100-μl aliquot of prewarmed 2× NADPH-generating system (1.3 mM NADP+, 6.6 mM glucose 6-phosphate, 6.6 mM MgCl2, and 0.8 U/ml glucose-6-phosphate dehydrogenase) in 50 mM Hepes buffer (pH 7.4). The reaction was maintained at 37°C, and production of the fluorescent metabolite 7-hydroxytrifluoromethyl coumarin measured using a Fluoroskan Ascent FL microtiter plate reader (λex = 405, λem = 530 nm; Labsystems, Cambridge, UK). For the single-point screen, the inhibition of BFC hydroxylation was measured using 50 μM substrate (close to the KM of 51 μM measured for BFC) in the presence of 100 μM of each comedication compound as described below. Results of the screening were expressed as the percentage activity in the presence of 100 μM comedication compound compared with activity of the appropriate solvent control. Those comedication compounds that inhibited enzyme activity to <70% of control activity in duplicate experiments were evaluated further by measuring the percentage of inhibition across a concentration range covering 0 to 100% inhibition of CYP3A4 activity, and IC50 values were calculated using GraFit 5.0.4 (Erithacus Software Ltd., Surrey, UK).
Loperamide Kinetics. The KMs of loperamide as a substrate for wild-type CYP3A4 and the E374Q mutant were determined by high-performance liquid chromatography-tandem mass spectrometry. Loperamide (0–100 μM) was incubated with 20 pmol CYP3A4 or the E374Q mutant in 10 mM Hepes buffer (pH 7.4) containing 6 mM MgCl2, and reactions were initiated by the addition of a 2× NADPH-regenerating system in a total volume of 200 μl. After incubation for 8 min at 37°C, the reaction was stopped by the addition of 200 μl of ice-cold methanol. After centrifugation at 16,000g and 5 min to remove particulate material, 10 μl of the reaction supernatant was injected into the high-performance liquid chromatography-tandem mass spectrometry system (Micromass Quattro Micro mass spectrometer; Waters, Hertfordshire, UK). The analytes were separated on a Hyperclone BDS C18 column (3 μm; 50 × 2.0 mm) at a flow rate of 0.2 ml/min. A linear gradient was applied using 10 mM ammonium formate and 0.1% (v/v) formic acid (A) and acetonitrile (B). The gradient of the mobile phase ran from 25 to 90% B in 5 min followed by 90% B for 3 min before returning to the initial conditions. The total run time was 13 min. The major MS parameters were capillary voltage = 3.6 kV, sample cone voltage = 25 V, collision energy = 23 eV, desolvation temperature = 350°C, source temperature = 120°C, and the transition for the N-demethylated product was 463 → 252 m/z.
Results
CYP3A4 Affinities of Comedication Drugs. The 33 comedication drugs used in this study are shown in Table 1. All these compounds could be successfully docked into the active site of CYP3A4, using either of the two available crystal structures of ligand-free CYP3A4 (Williams et al., 2004; Yano et al., 2004). These dockings yielded ChemScore values ranging from –41.0 to –10.1 kJ/mol (Table 1; note that the more negative the value, the tighter the predicted binding). The ChemScore value for loperamide (<–40 kJ/mol) suggests that it binds tightly to CYP3A4, and this was confirmed by its experimentally determined IC50 value of 0.050 ± 0.006 μM. Encouraged by this success with loperamide, we investigated the ability of our other 32 comedication compounds to inhibit CYP3A4. First, a single-point inhibition assay revealed that 18 of the compounds in the set were at best very weak inhibitors (<30% inhibition at 100 μM), consistent with the relatively poor scores for these compounds in our docking studies (ChemScore values > –29 kJ/mol; Table 1). The remaining 15 compounds were selected for IC50 determination (Table 1). The ChemScore values for these 15 “binders” correlate significantly with the experimental log(IC50) values, with r2 = 0.75 (q2 = 0.73) for docking into the CYP3A4 crystal structure 1tqn and r2 = 0.66 (q2 = 0.64) for docking into the CYP3A4 structure 1w0e. The difference between the results for the two structures should be treated with some caution because only three outliers (citalopram, lansoprazole, and omeprazole) led to the slightly poorer regression with 1w0e.
In our previous work on CYP2D6, we defined “tight binders” as those compounds with a ChemScore of <–30 kJ/mol and an IC50 value of <10 μM (Kemp et al., 2004). Although these values are somewhat arbitrary, applying the same criteria here reveals that five of the seven compounds with an IC50 value of <10 μM—loperamide, domperidone, simvastatin, diltiazem, and amitriptyline—are correctly predicted, by docking into both crystal structures, as tight binders. The other two compounds with IC50 value <10 μM—omeprazole and lansoprazole—are predicted correctly using the 1tqn, but not the 1w0e, structure. Conversely, all compounds with an IC50 value >100 μM—metoclopramide, dexamethasone, lorazepam, R-warfarin, and prednisolone—are correctly predicted to be weaker binders. The three remaining compounds—fluoxetine, ondansetron, and citalopram—with IC50 values in the “gray” area between 10 and 100 μM are all predicted by docking to be tight binders.
Loperamide Binding. Our docking studies correctly predicted loperamide to be the tightest binder with a ChemScore value < –40 kJ/mol and an IC50 value of 0.05 μM. This was of interest because at the time this drug was not known to be a CYP3A4 ligand. Two studies reported since this part of our work was carried out have demonstrated that loperamide is a substrate of CYP3A4 (Kalgutkar and Nguyen, 2004; Kim et al., 2004), although the nature of its interactions within the active site has not been investigated. Our docking studies predict that in the favored (lowest energy) loperamide-CYP3A4 complex, the loperamide is positioned for formation of the major, N-demethylated, product (Kalgutkar and Nguyen, 2004; Kim et al., 2004) (Fig. 1). In these models, loperamide interacts with CYP3A4 via 1) nonpolar interactions with the phenylalanine cluster (Fig. 1B) and 2) polar interactions with Arg-212 (the guanidinium moiety of Arg-212 is predicted to hydrogen bond to the amide carbonyl) and Glu-374 (a carboxyl oxygen of Glu-374 is predicted to hydrogen bond to the piperidine hydroxyl). Interestingly, the hydrogen bond acceptor and hydrophobic interactions of loperamide with CYP3A4 (Fig. 1, A and B) appears similar to mibefradil fitted to a CYP3A4 pharmacophore (Figure 2a in Ekins et al., 2003). The hydrogen bond with Arg-212 is only observed in the 1tqn structure, reflecting the rotation of the Arg-212 side chain toward the heme in this structure. The absence of such an interaction in the dockings into the 1w0e structure does not change the overall orientation of the ligand in the binding site, but in these models, the tertiary amide moiety is positioned closer to the iron atom where it might be more readily oxidized.
To confirm the predicted role of Glu-374 in loperamide binding, mutagenesis experiments were performed. When this negatively charged amino acid was replaced by the polar (uncharged) glutamine, the mutant produced an approximate 50% reduction in the yield of P450. We also noted a 3-fold increase in P420 content in the membranes relative to wild type, suggestive of some protein misfolding. We determined a KM for wild-type CYP3A4 of 2.6 ± 0.3 μM for the production of the N-demethylated product, slightly lower than the value of 6.3 μM reported by Kim et al. (2004). The E374Q mutation has a significant effect on loperamide binding for N-demethylation, with a ∼12-fold increase in KM to 31.4 ± 1.9 μM (i.e., weaker binding). These results indicate a functional role for Glu-374 in the binding of loperamide to CYP3A4 and that our in silico study has correctly identified a significant ligand binding interaction.
Discussion
In this study, we have shown that in silico screening, using GOLD with the ChemScore scoring function, allows rapid identification of compounds which bind tightly to CYP3A4. The use of two independently solved crystal structures of CYP3A4 leads to similar predictions (Table 1). The in silico screening correlates well with a single point in vitro experimental screen in identifying which compounds bind to CYP3A4 (Table 1). In addition, it identifies correctly most (seven of seven using structure 1tqn and five of seven using structure 1w0e) of the tightly binding (IC50 < 10 μM) compounds in our comedication set. All the compounds predicted to bind weakly (ChemScore >–30 kJ/mol) are found experimentally to have IC50 > 100 mM. The three compounds that are “false positives” (“best” ChemScore < –30 kJ/mol and IC50 > 10 μM) all have IC50 values in the “gray” area ranging from 10 to 100 μM. Significant correlation between theoretical and experimental binding affinities is observed, slightly better when using structure 1tqn (r2 = 0.75; q2 = 0.73) than structure 1w0e (r2 = 0.66; q2 = 0.64). This difference might seem surprising because in structure 1tqn, the side chain of Arg-212 occupies a position in the binding site above the heme, whereas in structure 1w0e, it is reoriented away from the binding site. The impact of Arg-212 on substrate binding and catalysis by CYP3A4 remains unclear (Harlow and Halpert, 1997; Yano et al., 2004), and we are currently investigating this further. The results we report here suggest that in many cases, the choice of CYP3A4 crystal structure is not a crucial factor; however, in most cases where there is a difference in the predicted ChemScore value between the two structures, the compounds are predicted to bind more tightly to structure 1tqn (Table 1).
The usefulness of relatively fast and simple ligand-based descriptors was investigated. The active site of CYP3A4 is generally regarded as large and hydrophobic, as borne out by a predominantly hydrophobic active site in the crystal structures (Williams et al., 2004; Yano et al., 2004). Therefore, a “null hypothesis” for ligands binding to CYP3A4 is that ligand binding in the active site of CYP3A4 is due to hydrophobic interactions (Smith et al., 1997). Setting a threshold Slog P value of 2.5 to define tight binders, we were able to correctly identify six of the seven tight bingers (IC50 < 10 μM using BFC as a probe substrate). Although this simple filter worked well qualitatively, it performed badly quantitatively (r2 = 0.37, q2 = 0.32 versus log IC50), suggesting (as expected) that additional factors over and above hydrophobicity come into play in the selectivity exhibited by 3A4. Using molecular weight (r2 = 0.22, q2 = 0.0.16 versus log IC50), number of hydrogen bond acceptors (r2 = 0.01, q2 = 0.07 versus log IC50), and number of hydrogen bond donors (r2 = 0.06, q2 = 0.02 versus log IC50) as descriptors also resulted in poor quantitative results.
In this work, we have used a structure-based in silico screening approach [with a program consistently found to perform well (see Kellenberger et al., 2004; Kontoyianni et al., 2004)] in which the structure of the protein remains rigid and a single compound (ligand) is docked at a time. Such an approach might be considered inappropriate for CYP3A4 because of its relatively large active site (Yano et al., 2004), the high degree of flexibility within the active site (Anzenbacherova et al., 2000; Anzenbacher and Hudecek, 2001), and the ability of the enzyme to bind multiple ligands simultaneously (Kenworthy et al., 2001; Khan et al., 2002). Despite these concerns, the approach worked well with this set of comedication compounds, showing that screening for likely drug-drug interactions is possible using a simple approach without the need for the additional computational overhead and complexity incurred when protein flexibility is invoked.
We predicted (ChemScore –45.0 kJ/mol), and subsequently verified experimentally (IC50 0.05 μM), that the antidiarrheal drug loperamide binds tightly to CYP3A4, consistent with the observation that loperamide probably interacts with CYP3A4 (Tayrouz et al., 2001). In addition, the binding orientation predicted by our computational docking is consistent with the production of the major, N-demethylated, metabolite. Soon after we completed this aspect of the work, and consistent with our findings, the major enzymes responsible for the metabolism of loperamide in humans were identified as CYP3A4, CYP2C8, and CYP2B6, the major product of its metabolism by CYP3A4 being the N-demethylated compound (Kalgutkar and Nguyen, 2004; Kim et al., 2004).
A number of comedication drugs have been identified in this work as inhibitors of CYP3A4. Clearly, the most important factor in determining the likelihood of a cytochrome P450-mediated drug-drug interaction is the concentration of a compound to which the P450 enzyme is exposed relative to its inhibitory potency (Ito et al., 2004). It should be noted that the concentration in vivo of a compound is confounded both by intestinal presystemic metabolism by CYP3A4 and by the activity of efflux transporters in the intestine. In the case of loperamide, there appears to be potential to cause clinical problems because the therapeutic levels (Cmax) of loperamide [∼0.04 μM (Kim et al., 2004)] are close to the IC50 value of 0.05 μM. Indeed, coadministration of loperamide has been shown to reduce exposure to the human immunodeficiency virus protease inhibitor saquinavir (Mikus et al., 2004), which is a CYP3A4 substrate (Eagling et al., 2002). Irinotecan (CPT-11), used as a treatment for colorectal cancer, is converted by carboxylesterases to a potent inhibitor of topoisomerase I. Dose-limiting diarrhea during irinotecan treatment is commonly treated with loperamide—but irinotecan is a known substrate of CYP3A4 (Haaz et al, 1998a,b), and its metabolism has been shown to be inhibited by loperamide (Haaz et al., 1998a,b)—raising the possibility of significant drug-drug interactions, although the quantitative importance of this remains to be established. In a similar way, the evidence for potent CYP3A4 inhibition by loperamide raises the possibility of interactions with docetaxel, cyclophosphamide and a number of other cytotoxics that can cause troublesome diarrhea. Several other drugs are highlighted by this study with low IC50 values (<10 μM; i.e., tight binders), including simvastatin, an HMG-CoA reductase inhibitor which is the agent most commonly prescribed in the UK for hypercholesterolemia, and omeprazole, a gastroprotectant often given to patients on steroids.
To summarize, we have shown that a relatively “simple” in silico method can be used to predict whether drugs commonly taken by patients with cancer as part of a comedication regime interact with CYP3A4. As validated by our subsequent experimental binding studies, which determined a self-consistent set of IC50 values, six of seven tight binding compounds were identified correctly, and all the compounds predicted to be weak binders or nonbinders to CYP3A4 were correctly identified. The antidiarrheal drug loperamide was identified as a particularly tight binder to CYP3A4 and thus warrants attention for possible involvement in drug-drug interactions. Although detailed pharmacokinetic studies will always be required to assess the quantitative importance of such interactions, the work described here demonstrates that our in silico approach is a valuable screen for identification of comedication compounds that may present problems. There are alternatives to many of the agents we identified as having the potential to cause significant, troublesome interactions with anticancer agents. Knowledge of these interactions may lead to more personalized and more appropriate prescribing by oncologists.
Footnotes
-
This work was supported by Cancer Research UK, the University of Manchester, and the Higher Education Reach-Out to Business and the Community Fund (HEROBC).
-
Article, publication date, and citation information can be found at http://dmd.aspetjournals.org.
-
doi:10.1124/dmd.105.007625.
-
ABBREVIATIONS: P450, cytochrome P450; CPR, cytochrome P450 oxidoreductase; BFC, 7-benzyloxy-4-trifluoromethylcoumarin; SRS, substrate recognition site.
- Received September 29, 2005.
- Accepted January 9, 2006.
- The American Society for Pharmacology and Experimental Therapeutics