COMBINED THREE-DIMENSIONAL QUANTITATIVE STRUCTURE-ACTIVITYRELATIONSHIP ANALYSIS OF CYTOCHROME P450 2B6 SUBSTRATES AND PROTEIN

HOMOLOGY MODELING

QINMI WANG AND JAMES R. HALPERT

Department of Pharmacology and Toxicology, University of Texas Medical Branch, Galveston, Texas

(Received July 24, 2001; accepted October 4, 2001)

This paper is available online at http://dmd.aspetjournals.org

ABSTRACT:

Understanding the basis of the substrate specificity of cyto-chrome P450 2B6 (CYP2B6) is important for determining the roleof this enzyme in drug metabolism and for predicting new sub-strates. Pharmacophores were generated for 16 structurally di-verse CYP2B6 substrates with Catalyst after overlapping thereaction sites. Two pharmacophores were determined for theCYP2B6 binding site. Both include two hydrophobes and onehydrogen bond acceptor. The three-dimensional structure ofCYP2B6 was then modeled based on the crystal structure ofCYP2C5. Benzyloxyresorufin and 7-ethoxy-4-trifluoromethyl-coumarin, the two lowest Km substrates in the training set, werethen docked in the active site of CYP2B6. The pharmacophoreswere combined with the CYP2B6 model by comparing the dock-ing results and the mapping of the two substrates with the

pharmacophores. The results indicated that the active site ofCYP2B6 complements the pharmacophores. The pharmaco-phores and the CYP2B6 model were used in conjunction topredict the Km values of substrates in a test set of five com-pounds and yielded satisfactory predictions for benzphetamine,cinnarizine, bupropion, and verapamil but not lidocaine. TheCYP2B6 model, the pharmacophores, and the combination ofthe model with these pharmacophores provide insight into theinteractions of CYP2B6 with substrates. The pharmacophoresmay be used as queries to search a database to predict newsubstrates for CYP2B6 when the reaction site is known (N- orO-dealkylation). For C-hydroxylation, the CYP2B6 model is help-ful in evaluating the possible reaction sites in order for thepharmacophores to predict corresponding Km values.

The cytochromes P450 (P450) are a superfamily of hemoproteinsthat are involved in oxidative, peroxidative, and reductive metabolicbiotransformation of a wide variety of endogenous compounds, suchas steroids, prostaglandins, and fatty acids, and exogenous com-pounds, such as drugs and carcinogens (Ortiz de Montellano, 1995;Nelson et al., 1996). Individual P450s exhibit unique substrate spec-ificity and regio- and stereoselectivity profiles that reflect differenttertiary structures. The CYP2B subfamily contains some of the firstP450s to be cloned and purified and has served as a model system formammalian P450 structure-function analysis through the use of allelicvariants, site-directed mutagenesis, and computer-aided molecularmodeling (Lewis et al., 1999; Domanski and Halpert, 2001). TheCYP2B1 substrate recognition site residues 103, 114, 115, 206, 209,290, 294, 297, 298, 302, 362, 363, 367, 477, 478, and 480 have beendemonstrated to be very important for substrate metabolism (Doman-ski and Halpert, 2001; Domanski et al., 2001). Human CYP2B6metabolizes about 3% of drugs in clinical use, and structure-function

analyses have only recently begun as new clinically relevant sub-strates have been identified (Rendic and Di Carlo, 1997) and heter-ologous expression has been accomplished (Hanna et al., 2000). Morework is necessary to understand the structure and specificity ofCYP2B6 and its role in metabolism.

Recently efforts have intensified to use the quantitative structure-activity relationship (QSAR) for understanding P450 active sites(Ekins et al., 2001). Such approaches are based on the idea ofcombining chemical knowledge about small compounds with exper-imental data obtained from in vitro systems. A considerable number ofcurrent QSAR models have been generated for P450s, especially forCYP2B6, 2C9, 2D6, and 3A4 (Koymans et al., 1992; Strobl et al.,1993; Jones et al., 1996; Rao et al., 2000; Ekins et al., 1999a,b,c,d,2000). The QSAR approaches used include comparative molecular-field analysis, Catalyst, and molecular surface-weighted holistic in-variant molecular analysis. Comparative molecular-field analysis isdependent on the alignment of small compounds, whereas Catalyst isalignment independent. In some cases, the pharmacophores werecombined with homology models of P450s and provide a morepowerful tool to investigate the active-site features and substrateinteractions (de Groot et al., 1999a,b; Afzelius et al., 2001). A noveland selective CYP2D6 substrate, which is suitable for high-through-put screening, has already been successfully designed with the help ofone of these models (Onderwater et al., 1999).

Two 3D-QSAR models were generated for CYP2B6 substrates(Ekins et al., 1999c). One is a pharmacophore model built by Catalystand consisting of three hydrophobes and one hydrogen bond acceptor

Supported by AstraZeneca and National Institutes of Health Grant ES03619(J.R.H.) and Center Grant ES06676.

1 Abbreviations used are: P450, cytochrome P450; 3D-QSAR, three-dimen-sional quantitative structure-activity relationship; MM, molecular mechanics;7-EFC, 7-ethoxy-4-trifluoromethylcoumarin; HBA, hydrogen bond acceptor.

Address correspondence to: Dr. James R. Halpert, Professor and Chairman,Department of Pharmacology and Toxicology, University of Texas Medical Branch,301 University Blvd., Galveston, TX 77555-1031. E-mail: jhalpert@utmb.edu

0090-9556/02/3001-86–95$3.00DRUG METABOLISM AND DISPOSITION Vol. 30, No. 1Copyright © 2002 by The American Society for Pharmacology and Experimental Therapeutics 531/953755DMD 30:86–95, 2002 Printed in U.S.A.

86

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

region. The other partial least-squares model was generated usingmolecular surface-weighted holistic invariant molecular descriptors.Molecular size, positive electrostatic potential, hydrogen bond accep-tors, and hydrophobicity were found to be important for CYP2B6substrate binding. Although both 3D-QSAR models predicted satis-factory Km values for the majority of the test set molecules, thepharmacophore generated by Catalyst does not overlay the oxidationsite of the substrates in the training set. This may be a real limitationfor a pharmacophore made for substrates rather than inhibitors.

Only one mammalian P450 (rabbit CYP2C5) has been crystallizedto date (Williams et al., 2000). This is a revolution in the study of thestructures of these membrane-bound enzymes, but generation of fur-ther structures is impeded by the difficulty in obtaining diffractionquality crystals. In the absence of experimental data, homology mod-eling becomes an important tool to predict the three-dimensionalstructures of P450 enzymes. Several reviews have recently addressedthe homology modeling of P450s (de Groot and Vermeulen, 1997;Szklarz and Halpert, 1997a; Peterson and Graham, 1998). Earlymodels were based on one or more of four bacterial crystal structures(Szklarz and Halpert, 1997b). However, recently models have beenbuilt based on CYP2C5 structures, including CYP2B1, 2B4, and 2B5(Spatzenegger et al., 2001) and CYP2C9 (Afzelius et al., 2001). Thesehomology models provide a structural basis for understanding themechanism of P450-mediated drug metabolism and drug-drug inter-actions.

In this study, we used Catalyst to generate pharmacophores forCYP2B6 substrates with the reaction site of each substrate overlaid. Inaddition, a model of CYP2B6 was constructed by homology modelingbased on the crystal structure of CYP2C5. The pharmacophores werecombined with the CYP2B6 model by docking the substrates in theactive site of CYP2B6. Finally, the pharmacophores were used topredict the Km value for test set molecules in conjunction with theCYP2B6 model. The combined model holds increased potential tostudy and predict CYP2B6-mediated drug metabolism.

Materials and Methods

Pharmacophore Generation for CYP2B6 Substrates. Pharmacophoregeneration was done using Catalyst (Molecular Simulations, Inc., San Diego,CA). The structures of CYP2B6 substrates were built interactively using theView Compounds Workbench. The conformational search was performed oneach structure using the best conformer generation method in Catalyst. Thenumber of conformers for each substrate was limited to a maximum of 250with an energy range of 20 kcal/mol. The three-dimensional pharmacophoremodels were generated based on the Km values and the conformers of thesubstrates by selecting the default chemical functions in the Feature Dictionaryin Catalyst, including hydrogen bond donor or acceptor, hydrophobic, negativeor positive ionizable features, etc. The substrates were then fit to the generatedpharmacophore models. The results showed that the sites of oxidation of thesubstrates were not overlaid when they were fit to the pharmacophore modelsbecause there is no default function in the Feature Dictionary that can recog-nize the reaction sites of the substrates and then overlay them together when apharmacophore is generated.

To force overlay of the reaction sites, a novel function was created toinclude the structural features of the reaction site for each substrate. The newfunction was then saved in the Feature Dictionary of Catalyst. For example, forcompounds that undergo N-demethylation (Fig. 1), the common N-CH3 groupfor each substrate was put in the novel function in the Feature Dictionary.Catalyst thus can recognize the reaction by searching for the N-CH3 group inthe structure of a pertinent compound. For O-deethylation substrates, thecommon O-CH2CH3 group was included in the Feature Dictionary. For sub-strates that undergo C-hydroxylation, the specific substructure, including thereaction site was included in the Feature Dictionary in order for Catalyst torecognize the reaction site. The pharmacophore models were again generated

by selecting the hydrogen bond donor, hydrogen bond acceptor, hydrophobicgroups, and the novel functions in the Feature Dictionary.

The pharmacophore models generated were evaluated by cost analysis inCatalyst. The lowest cost hypothesis is considered to be the best. However,hypotheses with costs within 10 to 15 arbitrary units of the lowest costhypothesis should also be considered good candidates (Catalyst Tutorials;Molecular Simulations, Inc.).

Homology Modeling of CYP2B6. The three-dimensional structure of hu-man CYP2B6 was built based on the crystal structure of CYP2C5 (Williams etal., 2000) using Insight II/Homology and Insight II/Discover-3 modules (Mo-lecular Simulations, Inc.). The sequence of CYP2B6 was obtained fromSwissProt (accession number P20813). The sequence alignment of CYP2B6and 2C5 was done by GCG (Wisconsin Package Version 10.0; GeneticsComputer Group, Madison, WI). The crystal structure of CYP2C5 lacks theN-terminal residues 1 to 29 and F-G loop residues 212 to 222. Correspond-ingly, the CYP2B6 was modeled from residues 31 to 491. The segment inCYP2B6 corresponding to residues 212 to 222 in CYP2C5 was constructedbased on the coordinates of CYP2C5 containing one of two alternative modelsfor density corresponding to the F-G loop (E. F. Johnson, unpublished obser-vation). For residues 276 to 278, the only segment not considered to beconserved, the coordinates were generated by searching the PDB databank tofind the regions of proteins that meet the geometric criteria of the loop. Thecoordinates of the other residues were assigned based on CYP2C5. The hemegroup was copied from CYP2C5 into the CYP2B6 model.

This preliminary structure was then subjected to molecular mechanics(MM) energy refinement by the Insight II/Discover-3 module with conjugategradients to a maximum of 1 kcal mol�1 �1. The parameters for the hemegroup were described previously (Paulsen and Ornstein, 1991, 1992). Thedefault cvff force field of Insight II was used for the rest of the model. First,the splices between residues 275 and 276, 278 and 279 were repaired to avoidsteric hindrance in these junction regions. The loop of residues 276 to 278 wasrelaxed using molecular dynamics followed by MM minimization. All thehydrogen atoms were put into minimization while the remaining heavy atomswere fixed. The side chains were minimized with the backbone atoms fixed. A25-Šsphere and a 3-Šsurface layer of water were soaked into and around theCYP2B6 structure to provide a solution environment. Finally, MM energyminimization was performed again on the whole soaked enzyme followed bya 150-ps molecular dynamics calculation with the soaked waters fixed.

The CYP2B6 model was analyzed by Prostat in the Insight II/Homologymodule. The cutoff used, which represents the significant difference for bondlength, bond angle, and torsion from the reference value, is 5 S.D. No bonddistances, seven bond angles, and 10 dihedral angles were identified to havemore than 5 S.D.

Docking. Benzyloxyresorufin and 7-ethoxy-4-trifluoromethylcoumarin (7-EFC) were docked automatically in the active site of the CYP2B6 model usingthe Insight II/Affinity module. The conformer of each docked substrate that fitsthe pharmacophore the best was exported from Catalyst to be the initialstructure for docking.

Pharmacophore Validation by a Test Set of Substrates. A series ofsubstrates in the test set were used to validate the pharmacophore models. Aconformational search was also performed on the test set substrates. Eachsubstrate was mapped to the pharmacophore by the Best Fit Compare methodin Catalyst. All conformers of the substrate were checked during the mappingprocess. Therefore, a number of mappings to the pharmacophore were obtainedfor each substrate. The predicted activity, which indicates how active themolecule is estimated to be and how well the molecule matches to thepharmacophore, was estimated for each mapping of the substrate. The mappingin which the substrate is located in the orientation for metabolism was selectedwith the help of the CYP2B6 model. In this manner, the final predicted Km foreach substrate was determined.

Results

Pharmacophores of CYP2B6 Substrates. Pharmacophore modelswere generated for a training set of 16 structurally diverse substratesof CYP2B6 (Fig. 1) with Km values shown in Table 1. The Km valueswere obtained from the literature and were generated with CYP2B6expressed in �-lymphoblastoid cells. Eight hypotheses (Table 2) were

87SUBSTRATE AND PROTEIN MODELS OF CYP2B6

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

FIG. 1. The structures of CYP2B6 substrates in the training set and test set.

�, denotes a chiral carbon.

88 WANG AND HALPERT

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

generated by Catalyst. All the hypotheses consist of two hydrophobesand one hydrogen bond acceptor, which are located in differentrelative positions in each instance.

Cluster analysis on the eight hypotheses showed that they can bedivided into two groups (Table 2). According to the criteria ofCatalyst, the lowest cost hypothesis is the best one, although hypoth-eses within 10 to 15 arbitrary units of the lowest cost should beconsidered good candidates. Therefore, the lowest cost hypothesiswas selected to be the representative pharmacophore in each group.They are hypothesis 1 representing the first group and hypothesis 2representing the second group (Table 2).

Catalyst tries to map all functions in a hypothesis to one of the twomost active training set molecules. Benzyloxyresorufin and 7-EFC arethe two lowest Km substrates in the training set (Table 1). Theirstructures differ significantly (Fig. 1). The molecular size of benzy-loxyresorufin is larger than that of 7-EFC. Therefore, Catalyst gener-ated two kinds of pharmacophores primarily based on the differentstructures of the two lowest Km substrates.

The estimated Km values of benzyloxyresorufin and 7-EFC werepredicted by hypotheses 1 and 2, respectively. The experimental Km

value (1.3 �M) of benzyloxyresorufin is consistent with its estimatedKm value by hypothesis 1 (2.0 �M) but not by hypothesis 2 (34 �M).The experimental Km value (1.7 �M) of 7-EFC agrees with theestimated Km value predicted by hypothesis 2 (1.3 �M) but not byhypothesis 1 (48 �M). These results suggested that the correlation

coefficient of hypothesis 1 would be increased if 7-EFC were ex-cluded from the training set substrates. Pharmacophores were regen-erated for the remaining 15 substrates in the training set. Two hy-potheses were generated (Table 2B), also including two hydrophobesand one hydrogen bond acceptor. The relative positions of thesefeatures are very similar to the two hypotheses in the first group above(Table 2A, hypotheses 1 and 3), but the correlation coefficient for thelowest cost hypothesis is 0.84 rather than 0.76. Similarly, benzy-loxyresorufin was excluded from the training substrates to increasethe correlation coefficient of hypothesis 2, and pharmacophores werere-generated for the remaining 15 substrates in the training set. Sixhypotheses were generated (Table 2B) in which the functions have thesame relative positions and are very similar to those of the secondgroup hypotheses (Table 2A). The correlation coefficient of the lowestcost hypothesis is 0.82 instead of 0.75 before excluding benzy-loxyresorufin. The two kinds of pharmacophore models with correla-tion coefficients 0.84 and 0.82 (Table 2B) are shown as pharmacoph-ore A and B in Fig. 2. Both of the pharmacophores include twohydrophobes and one hydrogen bond acceptor located in differentrelative positions.

Benzyloxyresorufin was mapped to pharmacophore A, and 7-EFCwas mapped to pharmacophore B (Fig. 3). In Fig. 3A, the two phenylrings of benzyloxyresorufin fit to the hydrophobic features H1 and H2.It can be inferred that the two phenyl rings may form hydrophobicinteractions with the enzyme. The oxygen atom on one of the rings of

TABLE 1

Training set substrates

The Km values were generated using CYP2B6 expressed in �-lymphoblastoid cells.

Substrates Km Metabolic Pathway References

�M

Amitriptyline 144 N-Demethylation Ekins et al., 1999c�-Arteether 28 O-Deethylation Grace et al., 1998Benzyloxyresorufin 1.3 O-Debenzylation Ekins et al., 19984-Chloromethyl-7-ethoxycoumarin 34 O-Deethylation Ekins et al., 19983-Cyano-7-ethoxycoumarin 71 O-Deethylation Ekins et al., 1998Dextromethorphan 350 N-Demethylation Ekins et al., 1999cDiazepam 113 N-Demethylation Ono et al., 19961,2-Dibromoethane 9700 2-Bromoacetaldehyde formation Wormhoudt et al., 19967-Ethoxycoumarin 115 O-Deethylation Yamazaki et al., 19967-Ethoxy-4-trifluoromethylcoumarin 1.7 O-Deethylation Ekins et al., 1997Imipramine 383 N-Demethylation Koyama et al., 1997Midazolam 46 1�-Hydroxylation Ekins et al., 1998Propofol 10 4-Hydroxylation Court et al., 2001RP73401 23 Ring hydroxylation Stevens et al., 1997S-Mephenytoin 564 N-Demethylation Heyn et al., 1996Testosterone 51 16�-Hydroxylation Ekins et al., 1998

TABLE 2

Hypotheses generated for CYP2B6 substrates in the training set

AHypotheses Generated from the Original Training Set of 16 Substrates

1 2 3 4 5 6 7 8

r 0.76 0.75 0.71 0.71 0.48 0.49 0.47 0.44Total cost 72.22 72.49 74.45 75.41 86.16 86.58 87.72 88.54Group First Second First Second Second Second Second Second

BHypothesesa Hypothesesb

1c 2 1d 2 3 4 5 6

r 0.84 0.73 0.82 0.81 0.74 0.74 0.67 0.62Total cost 60.30 64.42 63.59 63.94 66.63 67.54 71.11 72.31

a Hypotheses generated from 15 substrates excluding 7-EFC from the original training set.b Hypotheses generated from 15 substrates excluding benzyloxyresorufin from the original training set.c Pharmacophore A in Fig. 2.d Pharmacophore B in Fig. 2.

89SUBSTRATE AND PROTEIN MODELS OF CYP2B6

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

benzyloxyresorufin matches the hydrogen bond acceptor HBA1 (Fig.3A). This oxygen atom may accept a hydrogen atom and form ahydrogen bond with the enzyme or a water molecule in the active site.All functions in the pharmacophore B were fit by 7-EFC (Fig. 3B).One of the rings fits the hydrophobic function H3, and the trifluoro-methyl group fits another hydrophobic function H4. Both of them mayalso form hydrophobic interactions with the enzyme. The oxygenatom on the ring fits the hydrogen bond acceptor HBA2 and may forma hydrogen bond with CYP2B6 or a water molecule in the active site.

Homology Modeling. To determine how well the pharmacophoresfit the active site of CYP2B6, a three-dimensional structure of humanCYP2B6 was built based on the crystal structure of CYP2C5. Thesequence alignment is shown in Fig. 4. The amino acid residues thathave been determined previously to be important for substrate metab-olism in the CYP2B subfamily (Domanski and Halpert, 2001; Doman-ski et al., 2001) are shown in Fig. 5.

Combination of the Pharmacophores and the CYP2B6 Modelby Docking. The pharmacophores of the substrates should help one to

FIG. 2. Pharmacophores A and B for CYP2B6 substrates.

The R regions (orange) represent the overlaid reaction sites of the substrates. The cyan regions H1, H2, H3, H4 are hydrophobes. The green regions HBA1 and HBA2

stand for hydrogen bond acceptors with the vectors in the directions of the putative hydrogen bond. �H1RH2 � 88.6°; �H1RHBA1 � 46.1°; �H2RHBA1 � 56.8°;�H3RH4 � 19.9°; �H3RHBA2 � 43.4°; �H4RHBA2 � 36.3°.

FIG. 3. A, mapping of benzyloxyresorufin with pharmacophore A; B, mapping of 7-EFC with pharmacophore B.

90 WANG AND HALPERT

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

infer complementary structural features of the CYP2B6 active site. Togain confidence in the pharmacophores, we assessed how well thepharmacophores fit the active site by automatic docking of benzy-loxyresorufin and 7-EFC into CYP2B6. The docking results areshown in Figs. 6 and 7. The substrates are oriented with the metabolicsite pointing to the heme.

The mapping results of benzyloxyresorufin with pharmacophore A(Fig. 3A) and 7-EFC with pharmacophore B (Fig. 3B) were com-pared with the docking results of the two compounds in the activesite of CYP2B6 (Figs. 6 and 7, A and B). Comparison of thepositions of benzyloxyresorufin in both Figs. 3A and 6 allowedidentification of the reaction site region R, the two hydrophobes H1

and H2, and the hydrogen bond acceptor HBA1 of pharmacophore A inthe active site of the CYP2B6 model (Fig. 6). No hydrogen bond wasidentified between benzyloxyresorufin and CYP2B6 from the dockingresults, suggesting that a water molecule located near the hydrogen bondacceptor may form a hydrogen bond with benzyloxyresorufin. The dis-tances and angles among H1, H2, and R located in the active site ofCYP2B6 are very similar to those of pharmacophore A. Thus, pharma-cophore A complements the active site of the CYP2B6 model very well.

Docking of 7-EFC (Fig. 7, A and B) indicated that there are twopossible orientations for 7-EFC in the active site of CYP2B6. As withbenzyloxyresorufin, no hydrogen bond was found between 7-EFC andCYP2B6. From the location of 7-EFC in the active site and themapping with pharmacophore B (Fig. 3B), the reaction site R and thehydrophobes H3 and H4 of pharmacophore B can be located inCYP2B6, as shown in Fig. 7A and B. The distances and angles amongH3, H4, and R in pharmacophore B are very similar to those in theactive site. This indicates that pharmacophore B also complements theactive site of CYP2B6 very well.

The residues within 5Å of the center of the phenyl ring matching H1

in Fig. 6 are 114, 115, 206, 298, 363, 367, and 477. The residueswithin 5Å of the center of the ring matching H2 are 206, 297, 298,301, 302, 363, 477, and 478. With the exception of residue 301, whichhas not been tested, all of these residues are consistent with experi-mental results of site-directed mutagenesis of CYP2B1 in Fig. 5(Domanski and Halpert, 2001).

For the first orientation of 7-EFC (Fig. 7A), residues 206, 298, 302,363, 367, 477, and 478 are within 5Å of the center of the phenyl ringmatching H3, and residues 301, 302, 305, 306, 362, 363, 477, and 479are within 5Å of the trifluoromethyl group matching H4. For thesecond orientation of 7-EFC (Fig. 7B), the residues within 5Å of thecenter of the ring matching H3 are 114, 206, 297, 298, 302, 367, and

FIG. 5. Active site of a CYP2B6 homology model constructed based on thecrystal structure of CYP2C5.

The heme group is shown in red. Part of the backbone in the active site is shownas gray ribbons. The substrate recognition site residues determined to be importantfor substrate metabolism in CYP2B1 are shown in purple.

FIG. 4. Sequence alignment between CYP2B6 and 2C5 from residue 31 to 491.

The GAP method in GCG (Genetics Computer Group) was used for sequence alignment. The sequence identity between CYP2B6 and CYP2C5 is 49%, and thesimilarity is 61%.

91SUBSTRATE AND PROTEIN MODELS OF CYP2B6

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

477. The residues within 5Å of the trifluoromethyl group matching H4

are 100, 103, 114, 115, 297, and 367. Most of these residues also showexcellent agreement with previous CYP2B1 mutagenesis experimentsin Fig. 5 (Domanski and Halpert, 2001).

Predictions of the Pharmacophores in Conjunction with theCYP2B6 Model. The significance of a pharmacophore is determinedby its ability to predict new compounds. Pharmacophores A and Bwere used to predict the Km values for substrates in the test set (Table3) in conjunction with the CYP2B6 model. The conformers werecalculated for each substrate in the test set. A number of mappingswere obtained when each substrate was fit to the pharmacophore byCatalyst because each substrate has a number of conformers. Knowingthe orientations and positions of pharmacophores in the active site ofCYP2B6 helps to select which mapping is the one most suitable formetabolism by CYP2B6. Table 3 gives the experimental and predictedKm values for the test set molecules. Of the total substrates in the testset, the Km values for most were successfully predicted with at leastone of the two pharmacophores. It should be noted that the Km valuesreported for bupropion and verapamil refer to the racemic mixtures.Since the R- and S-enantiomers may bind differently to the enzyme,Km values were predicted for both forms. For both enantiomers ofbupropion, pharmacophores A and B give satisfactory predicted Km

values, indicating that bupropion may have more than one bindingposition in the active site of CYP2B6. For verapamil, the predicted Km

values are close to the experimental data except for R-verapamil andpharmacophore B. An exception to the predictive power of the phar-macophores is lidocaine. There are two ethyl groups attached to the Natom (Fig. 1). Although lidocaine fits pharmacophores A and B verywell (the predicted Km is low; Table 3), the two ethyl groups mayhinder the access of the N atom to the heme iron. This may explainwhy the predicted Km is lower than the experimental Km (Table 3).

Discussion

Two pharmacophores were determined for CYP2B6 substrates inthis study, both of which include two hydrophobic regions and onehydrogen bond acceptor. The correlation coefficients of the twopharmacophores are 0.84 and 0.82. Both pharmacophores comple-ment the active site of a CYP2B6 homology model based on CYP2C5

and gave satisfactory estimated Km values for the majority of the testset molecules in conjunction with the protein model. PharmacophoreA is located in one part of the active site, and pharmacophore B islocated in another position, with some overlap between the twopharmacophores, indicating that different possible binding locationsexist within the active site of the enzyme. The results are consistentwith previous QSAR studies on CYP2B proteins that suggested theimportance of hydrophobic and electronic properties in the binding ofsubstrates (Lewis et al., 1999; Ekins et al., 1999c).

Initially, we sought to investigate the active-site features ofCYP2B6 by combining our new three-dimensional homology modelwith the existing substrate pharmacophore of CYP2B6 generated byCatalyst (Ekins et al., 1999c). Following the procedures in that studywe recreated the pharmacophore, which gave satisfactory predictedKm values for the majority of the test set molecules (data not shown).However, we realized that the oxidation sites of the substrates in thetraining set had not been overlaid when they were mapped to thepharmacophore. Since the site of oxidation of the substrates shouldpoint toward the heme iron in the enzyme, a new function was madeand saved in the Feature Dictionary in Catalyst to force the reactionsite of each substrate to be overlaid when the pharmacophores weregenerated.

In constructing the new pharmacophores, we made some modifi-cations to the initial training set of 16 compounds. In particular,antipyrine was deleted due to questions about the validity of the veryhigh Km value (17.7 mM). In addition, benzphetamine and cinnarizinewere moved to the test set (Table 1). These three compounds werereplaced in the training set by arteether, amitriptyline, and propofol.Five compounds with different metabolic pathways are thus includedin the test set (Table 3) in this study to assess pharmacophore A andB. In addition, the R- and S-enantiomers of bupropion and verapamilin the test set were considered. The reported Km values of the twocompounds correspond to the racemic mixtures. However, the R- andS-forms differ in their binding with the CYP2B6 model. Therefore,the Km values were predicted for both R- and S-enantiomers of the twocompounds. Consideration of the stereochemistry had relatively littleimpact with bupropion or with pharmacophore A and verapamil.However, pharmacophore B predicted a Km in the millimolar range

FIG. 6. Benzyloxyresorufin docked in the active site of CYP2B6.

The heme group is shown in red. Benzyloxyresorufin is shown in orange, and residues within 5 Å are shown in green. The reaction site R and hydrophobes H1 and H2

of pharmacophore A are located according to the orientation and position of benzyloxyresorufin in the active site when compared with the mapping in Fig. 3A. The distancebetween H1 (the center of the phenyl ring matching H1) and R is 3.9 Å, and the angle �H1RH2 equals 90.3°. The values are very similar to the corresponding distance(4.0 Å) and angle (88.6°) in pharmacophore A (Fig. 2A). The distance between R and H2 does not vary.

92 WANG AND HALPERT

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

for R-verapamil. In the article by Ekins et al. (1999c), the R- andS-forms were not distinguished when the Km values were predicted forbupropion and verapamil.

The pharmacophore model made by Ekins consists of three hydro-phobes and one hydrogen bond acceptor region. The pharmacophoresgenerated in this study include two instead of three hydrophobes, onehydrogen bond acceptor, and an additional reaction site. If the map-ping of 7-EFC with the pharmacophore in the article by Ekins et al.(1999c) is compared with that (Fig. 3B) in this study, we can see that

the phenyl ring to which the ethoxy group attaches matches a hydro-phobe in both studies. The trifluoromethyl group attached to position4 of the other ring also fits a second hydrophobe in both studies.Therefore, the hydrophobicity of the trifluoromethyl group and thephenyl ring may be very important for 7-EFC binding. If the structuresof 7-ethoxy-4-trifluoromethylcoumarin, 4-chloromethyl-7-ethoxycou-marin, and 7-ethoxycoumarin (Fig. 1) are compared, only the first twocompounds have a group in position 4 of the phenyl ring, and their Km

values (1.7 and 33.7 �M; Table 1) are lower than that of the third

TABLE 3

Test set substrates

Substrates Experimental Km A Predicted Kma B Predicted Km

b Metabolic Pathway References

�M �M �MBenzphetamine 93.4 32 (�0.47)c 3700 N-Demethylation Ekins et al., 1998Bupropion 107.5d R: 140 (0.11) R: 82 (�0.12) Hydroxylation Ekins et al., 1999c

S: 190 (0.25) S: 95 (�0.05)Cinnarizine 17.2 3300 81 (0.67) p-Hydroxylation Kariya et al., 1996Lidocaine 537.6 32 (�1.23) 41 (�1.12) N-Deethylation Ekins et al., 1999cVerapamil 137.4d R: 120 (�0.06) R: 3200 O-Demethylation Ekins et al., 1999c

S: 87 (�0.20) S: 66 (�0.32)

R, the R-enantiomer; S, the S-enantiomer.a Km values predicted by pharmacophore A.b Km values predicted by pharmacophore B.c Values in parentheses represent the log units of predicted-experimental Km values.d The experimental Km value corresponds to the racemic mixture.

FIG. 7. 7-EFC docked in the active site of CYP2B6.

Two possible orientations were found for 7-EFC (orange) due to its small molecular size. The enzyme is oriented in the same direction in the two pictures. The hemegroup is shown in red, and the residues within 5 Å of 7-EFC are shown in green. The reaction site R and hydrophobes H3 and H4 of pharmacophore B are located in theactive site according to the orientations and positions of 7-EFC when compared with the mapping in Fig. 3B. The structure of 7-EFC is rigid, requiring that R, H3, andH4 are located in almost the same relative position as in pharmacophore B.

93SUBSTRATE AND PROTEIN MODELS OF CYP2B6

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

compound (115 �M; Table 1). This suggests that 4-substituent con-tributes to substrate binding in the active site of CYP2B6.

Lewis et al. (1999) constructed a three-dimensional homologymodel of CYP2B6 based on bacterial CYP102. A series of substrateswere docked into the active site of the constructed model. Thesubstrate template was then obtained by superimposing the dockedsubstrates. Although the template yielded some important features forthe substrates, the low sequence similarity between CYP2B6 andCYP102 is a major drawback. In our study, we had the advantage ofhaving a closely related mammalian CYP2C5 crystal structure toconstruct the homology model CYP2B6. The higher sequence identityand similarity between the CYP2C5 and CYP2B6 make the con-structed model more reliable.

Several recent studies have focused on the combination of pharma-cophores of small compounds and homology models of P450 en-zymes. Afzelius et al. (2001) generated a 3D-QSAR model for 29structurally diverse competitive CYP2C9 inhibitors. The 3D-QSARmodel was constructed with the help of the homology model ofCYP2C9. The CYP2C9 inhibitors were docked into the active site ofthe homology model, and then the active conformers of the inhibitorswere selected and used in the 3D-QSAR analysis. The 3D-QSARmodel was able to predict the entire test set molecules within 0.5 logunits of the experimental Ki values. The active site residues of thehomology model of CYP2C9 complement the features of the 3D-QSAR model. Furthermore, de Groot et al. (1999a,b) have alsopublished a combined protein and pharmacophore model forCYP2D6. The pharmacophore model was constructed by overlaying40 substrates using some distance criteria. Then the derived pharma-cophore model was oriented into the active site of the CYP2D6homology model. The pharmacophore model complemented the pro-tein model perfectly, although the two models were derived indepen-dently, thereby validating each other. Normally, pharmacophore or3D-QSAR models are used to study substrate/inhibitor binding sitefeatures and design new compounds without knowledge of the targetprotein structure. However, if pharmacophore or 3D-QSAR modelsare combined with three-dimensional structures of proteins, the reli-ability and the predictive power of the model should be enhanced. Inthis study, the two pharmacophores were also combined with theCYP2B6 model to deduce the orientations and positions of the phar-macophores in the active site of the enzyme. In the future, thepharmacophores, in conjunction with the protein model, may be usedas queries to search a database to predict substrates for CYP2B6. Fora compound in which the reaction site is known (N- or O-dealkyla-tion), the two pharmacophores, along with knowledge of their orien-tations and positions in the active site, can be used directly to predictKm values and evaluate the possibility of metabolism by CYP2B6. Forcompounds that undergo C-hydroxylation, the reaction sites could bemultiple. The homology model can also be used to predict the reactionsite by docking the compound into the active site. Then the Km valuescan be predicted based on the knowledge of the orientations andpositions of the pharmacophores in the active site.

Acknowledgments. We thank Drs. James M. Briggs and Gillian C.Lynch at the University of Houston (Houston, TX) for helpful sug-gestions. We also acknowledge Drs. Emily E. Scott and Tammy L.Domanski for help with the manuscript.

References

Afzelius L, Zamora I, Ridderstrom M, Andersson TB, Karlen A, and Masimirembwa CM (2001)Competitive CYP2C9 inhibitors: enzyme inhibition studies, protein homology modeling, andthree-dimensional quantitative structure-activity relationship analysis. Mol Pharmacol 59:909–919.

Court MH, Duan SX, Hesse LM, Venkatakrishnan K, and Greenblatt D (2001) Cytochrome

P-450 2B6 is responsible for interindividual variability of propofol hydroxylation by humanliver microsomes. Anesthesiology 94:110–119.

de Groot MJ, Ackland MJ, Horne VA, Alex AA, and Jones BC (1999a) Novel approach topredicting P450-mediated drug metabolism: development of a combined protein and pharma-cophore model for CYP2D6. J Med Chem 42:1515–1524.

de Groot MJ, Ackland MJ, Horne VA, Alex AA, and Jones BC (1999b) A novel approach topredicting P450-mediated drug metabolism. CYP2D6 catalyzed N-dealkylation reactions andqualitative metabolite predictions using a combined protein and pharmacophore model forCYP2D6. J Med Chem 42:4062–4070.

de Groot MJ and Vermeulen NP (1997) Modeling the active sites of cytochrome P450s andglutathione S-transferases, two of the most important biotransformation enzymes. Drug MetabRev 29:747–799.

Domanski TL and Halpert JR (2001) Analysis of mammalian cytochrome P450 structure andfunction by site-directed mutagenesis. Curr Drug Metab 2:117–137.

Domanski TL, He Y-Q, Scott EE, Wang Q, and Halpert JR (2001) The role of cytochrome 2B1substrate recognition site residues 115, 294, 297, 298 and 362 in the oxidation of steroids and7-alkoxycoumarins. Arch Biochem Biophys 394:21–28.

Ekins S, Bravi G, Binkley S, Gillespie JS, Ring BJ, Wikel JH, and Wrighton SA (1999a) Three-and four-dimensional-quantitative structure activity relationship (3D/4D-QSAR) analyses ofCYP2D6 inhibitors. Pharmacogenetics 9:477–489.

Ekins S, Bravi G, Binkley S, Gillespie JS, Ring BJ, Wikel JH, and Wrighton SA (1999b)Three-dimensional-quantitative structure activity relationship (3D-QSAR) analyses of inhib-itors for CYP3A4. J Pharmacol Exp Ther 290:429–438.

Ekins S, Bravi G, Binkley S, Gillespie JS, Ring B, Wikel J, and Wrighton SA (2000) Three- andfour-dimensional-quantitative structure activity relationship (3D/4D-QSAR) analyses ofCYP2C9 inhibitors. Drug Metab Dispos 28:994–1002.

Ekins S, Bravi G, Ring BJ, Gillespie TA, Gillespie JS, VandenBranden M, Wrighton SA andWikel JH (1999c) Three-dimensional-quantitative structure activity relationship (3D-QSAR)analyses of substrates for CYP2B6. J Pharmacol Exp Ther 288:21–29.

Ekins S, Bravi G, Wikel JH, and Wrighton SA (1999d) Three-dimensional quantitative structureactivity relationship (3D-QSAR) analysis of CYP3A4 substrates. J Pharmacol Exp Ther291:424–433.

Ekins S, de Groot MJ, and Jones JP (2001) Pharmacophore and three-dimensional quantitativestructure activity relationship methods for modeling cytochrome P450 active sites. DrugMetab Dispos 29:936–944.

Ekins S, Vandenbranden M, Ring BJ, Gillespie JS, Yang TJ, Gelboin HV, and Wrighton SA(1998) Further characterization of the expression in liver and catalytic activity of CYP2B6.J Pharmacol Exp Ther 286:1253–1259.

Ekins S, VandenBranden M, Ring BJ, and Wrighton SA (1997) Examination of purported probesof human CYP2B6. Pharmacogenetics 7:165–179.

Grace JM, Aguilar AJ, Trotman KM, and Brewer TG (1998) Metabolism of B-arteether todihydroqinghaosu by human liver microsomes and recombinant cytochrome P450. DrugMetab Dispos 26:313–317.

Hanna IH, Reed JR, Guengerich FP, and Hollenberg PF (2000) Expression of human cytochromeP450 2B6 in Escherichia coli: characterization of catalytic activity and expression levels inhuman liver. Arch Biochem Biophys 376:206–216.

Heyn H, White RB, and Stevens JC (1996) Catalytic role of cytochrome P250 2B6 in theN-demethylation of S-mephenytoin. Drug Metab Dispos 24:948–954.

Jones JP, He M, Trager WF, and Rettie AE (1996) Three-dimensional quantitative structure-activity relationship for inhibitors of cytochrome P450 2C9. Drug Metab Dispos 24:1–6.

Kariya S, Isozaki S, Uchino K, Suzuki T, and Narimatsu S (1996) Oxidative metabolism offlunarizine and cinnarizine by microsomes from B-lymphoblastoid cell lines expressing humancytochrome P450 enzymes. Biol Pharm Bull 19:1511–1514.

Koyama E, Chiba K, Tani M, and Ishizaki T (1997) Reappraisal of human CYP isoformsinvolved in imipramine N-demethylation and 2-hydroxylation: a study using microsomesobtained from putative extensive and poor metabolizers of S-mephenytoin and eleven recom-binant human CYPs. J Pharmacol Exp Ther 281:1199–1210.

Koymans L, Vermeulen NP, van Acker SA, te Koppele JM, Heykats JJ, Lavrijsen K, Meulder-mans W, and Donne-Op den Kelder GM (1992) A predictive model for substrates ofcytochrome P450-debrisoquine (2D6). Chem Res Toxicol 5:211–219.

Lewis DF, Lake BD, Dickins M, Eddershaw PJ, Tarbit MH, and Goldfarb PS (1999) Molecularmodelling of CYP2B6, the human CYP2B isoform, by homology with the substrate-boundCYP102 crystal structure: evaluation of CYP2B6 substrate characteristics, the cytochrome b5binding site and comparisons with CYP2B1 and CYP2B4. Xenobiotica 29:361–393.

Nelson DR, Koymans L, Kamataki T, Stegeman JJ, Feyereisen R, Waxman DJ, Waterman MR,Gotoh O, Coon MJ, Estabrook RW, et al. (1996) P450 superfamily: update on new sequences,gene mapping, accession numbers and nomenclature. Pharmacogenetics 6:1–42.

Onderwater RC, Venhorst J, Commandeur JNM, and Vermeulen NPE (1999) Design, synthesis,and characterization of 7-methoxy-4-(aminomethyl)coumarin as a novel and selective cyto-chrome P450 2D6 substrate suitable for high-throughput screening. Chem Res Toxicol 12:555–559.

Ono S, Hatanaka T, Miyazawa S, Tsutsui M, Aoyama T, Gonzalez FJ, and Satoh T (1996)Human liver microsomal diazepam metabolism using cDNA-expressed cytochrome P450s:role of CYP2B6, 2C19, and the 3A subfamily. Xenobiotica 26:1155–1166.

Ortiz de Montellano P (1995) Cytochrome P450 Structure, Mechanism, and Biochemistry,Plenum Press, New York.

Paulsen MD and Ornstein RL (1991) A 175-psec molecular dynamics simulation of camphor-bound cytochrome P-450cam. Proteins 11:184–204.

Paulsen MD and Ornstein RL (1992) Predicting the product specificity and coupling of cyto-chrome P450cam. J Comp-Aided Mol Des 6:235–252.

Peterson JA and Graham SE (1998) A close family resemblance: the importance of structure inunderstanding cytochromes P450. Structure 6:1079–1085.

Rao S, Aoyama R, Schrag M, Trager WF, Rettie A, and Jones PJ (2000) A refined 3-dimensional QSARof cytochrome P450 2C9: computational predictions of drug interactions. J Med Chem 43:2789–2796.

Rendic S and Di Carlo FJ (1997) Human cytochrome P450 enzymes: a status report summarizingtheir reactions, substrates, inducers, and inhibitors. Drug Metab Rev 29:413–580.

Spatzenegger M, Wang Q, He Y-Q, Wester MR, Johnson EF, and Halpert JR (2001) Amino acidresidues critical for differential inhibition of CYP2B4, CYP2B5, and CYP2B1 by phenylimi-dazoles. Mol Pharmacol 59:475–485.

94 WANG AND HALPERT

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from

Stevens JC, White RB, Hsu SH, and Martinet M (1997) Human liver CYP2B6-catalyzedhydroxylation of RP73401. J Pharmacol Exp Ther 282:1389–1395.

Strobl GR, von Kruedener S, Stockigt J, Guengerich FP, and Wolff T (1993) Development of apharmacophore for inhibition of human liver cytochrome P-450 2D6: Molecular modeling andinhibition studies. J Med Chem 36:1136–1145.

Szklarz GD and Halpert JR (1997a) Use of homology modeling in conjunction with site-directedmutagenesis for analysis of structure-function relationships of mammalian cytochromes P450.Life Sci 61:2507–2520.

Szklarz GD and Halpert JR (1997b) Molecular modeling of cytochrome P450 3A4. J Comp-Aided Mol Des 11:265–272.

Williams PA, Cosme J, Sridhar V, Johnson EF, and McRee DE (2000) Mammalian microsomalcytochrome P450 monooxygenase: structural adaptations for membrane binding and func-tional diversity. Mol Cell 5:121–131.

Wormhoudt LW, Ploemen JHTM, de Waziers I, Commandeur JNM, Beaune PH, van BladerenPJ, and Vermeulen NPE (1996) Inter-individual variability in the oxidation of 1,2-dibromoethane: use of heterologously expressed human cytochrome P450 and human livermicrosomes. Chem Biol Interact 101:175–192.

Yamazaki H, Inoue K, Mimura M, Oda Y, Guengerich FP, and Shimada T (1996) 7-Ethoxy-coumarin O-deethylation catalyzed by cytochromes P450 1A2 and 2E1 in human livermicrosomes. Biochem Pharmacol 51:313–319.

95SUBSTRATE AND PROTEIN MODELS OF CYP2B6

at ASPE

T Journals on Septem

ber 1, 2018dm

d.aspetjournals.orgD

ownloaded from