Identification of Inhibitors of the Nicotine Metabolising CYP2A6 Enzyme—An in Silico Approach

Identification of inhibitors of the nicotine metabolising CYP2A6 enzyme—an in silico approach

M Rahnasto1, C Wittekindt2, The compulsive nature of tobacco use is attributable to nicotine addiction. 1 3,4 Nicotine is eliminated by metabolism through the cytochrome P450 2A6 RO Juvonen , M Turpeinen , (CYP2A6) enzyme in liver. Inhibition of CYP2A6 by chemical compounds 4 3 A Petsalo , O Pelkonen , may represent a potential supplement to anti-smoking therapy. The purpose A Poso2, G Stahl5, H-D Ho¨ltje5 of this study was to rationally design potent inhibitors of CYP2A6. 3D-QSAR and H Raunio1 models were constructed to find out which structural characteristics are important for inhibition potency. Specifically located hydrophobic and 1Department of Pharmacology and Toxicology, hydrogen donor features were found to affect inhibition potency. These University of Kuopio, Kuopio, Finland; features were used in virtual screening of over 60 000 compounds in the 2 Department of Pharmaceutical Chemistry, Maybridge chemical database. A total of 22 candidate molecules were University of Kuopio, Kuopio, Finland; 3Department of Pharmacology and Toxicology, selected and tested for inhibition potency. Four of these were potent and 4 University of Oulu, Oulu, Finland; Novamass selective CYP2A6 inhibitors with IC50 values lower than 1 mM. They represent Analytical Ltd, Kiviharjuntie, Oulu, Finland and novel structures of CYP2A6 inhibitors, especially N1-(4-fluorophenyl)cyclo- 5 Institute of Pharmaceutical and Medicinal propane-1-carboxamide. This compound can be used as a lead in the design Chemistry, Heinrich-Heine University of Du¨sseldorf, Universitatsstrasse 1, Du¨sseldorf, of CYP2A6 inhibitor drugs to combat nicotine addiction. Germany The Pharmacogenomics Journal (2008) 8, 328–338; doi:10.1038/sj.tpj.6500481; published online 9 October 2007 Correspondence: M Rahnasto, Department of Pharmacology and Keywords: nicotine; CYP; QSAR; cytochrome P450; molecular modelling; smoking reduction Toxicology, PO BOX 1627, 70211 Kuopio, Finland. E-mail: [email protected]

Introduction

According to the World Health Organization (WHO), one-third of the global population over 15 years of age are smokers.1 Nicotine is the primary compound present in tobacco, and it plays a crucial role in establishing and maintaining tobacco dependence. Various nicotine preparations have been developed as medications to assist in smoking cessation. Nicotine has also been evaluated in the treatment of a variety of medical disorders.2 Understanding the pattern of nicotine metabolism and the sources of variations of this metabolism is important because of the key role of nicotine in producing tobacco dependence. Nicotine is eliminated by metabolism, and in humans, the key metabolite of nicotine is cotinine. The transformation of nicotine to cotinine involves two steps. The first step is its 50-oxidation to produce 50-hydroxynicotine and the nicotine-D10(50)-iminium ion. This reaction is mediated by the cytochrome P450 2A6 (CYP2A6) enzyme.3–5 The second step, formation of cotinine, is catalysed by cytosolic aldehyde oxidase.6 Received 21 January 2007; revised 9 July 2007; accepted 30 July 2007; published online Several variant alleles of the CYP2A6 gene have been characterised, causing 9 October 2007 reduced or absent enzyme activity.7–10 Individuals with genetically inactivated Rational design of CYP2A6 inhibitors M Rahnasto et al 329

CYP2A6 are slower at converting nicotine to cotinine.11 It from a chemical database to search for novel types of has been shown that nicotine-dependent slow inactivators, inhibitors. Four novel potent and selective inhibitors were that is individuals with at least one inactive CYP2A6 allele, identified. These compounds can serve as lead structures in smoke fewer cigarettes.12 Slow nicotine inactivators also developing clinically useful CYP2A6 inhibitors. smoke for a shorter duration before giving up smoking, are found more often among former smokers, and have Results increased success in quitting smoking during clinical 13–15 trials. There is growing evidence that the protection Comparison between homology model and crystal structure offered by slow metabolism can be mimicked by inhibiting After we had constructed the CYP2A6 homology model 16 CYP2A6 activity in vivo with chemical inhibitors. Proof-of- combined with the initial 3D-QSAR model and structure- concept studies have been carried out with the CYP2A6 based screening, the 3D structure of CYP2A6 co-crystallised inhibitors methoxsalen (used in the treatment of psoriasis) with coumarin and with methoxsalen was published.24 and tranylcypromine (a monoamine oxidase inhibitor used Later three other CYP2A6 structures co-crystallised with 12,16,17 in the treatment of depression). N,N-dimethyl(5-(pyridin-3-yl)furan-2-yl)methamine, N-di- Based on this data, a drug that acts as an inhibitor of methyl(5-(pyridin-3-yl)furan-2-yl)methamine, dimethyl(5- CYP2A6 could be used for two purposes: (1) to make it (pyridin-3-yl)furan-2-yl)methamine were published.25 In possible to administer of nicotine in a tablet form by mouth our homology model, the side chain of Thr106 forms a by increasing its oral bioavailability, and (2) as a drug to be hydrogen bond with the ligand. The crystal structure taken during smoking, reducing the amount of cigarettes revealed the hydrogen bonds between the ligand and the smoked and thus diminishing the adverse health effects of side chain of Asn297. To obtain more accurate data on smoking. In both cases, with or without nicotine augmenta- ligand orientation and hydrogen bonding, we created also a tion, CYP2A6 inhibitors should decrease levels of new 3D-QSAR model by docking molecules into the crystal smoking, promote cessation and reduce procarcinogen structure. activation.11,16,17 Several compounds have been tested for their inhibitory Statistics of CoMFA and CoMSIA models effects on the CYP2A6 enzyme in vitro.18–20 We have To determine which structural features are important for previously carried out quantitative structure–activity rela- inhibition potency, two 3D-QSAR models (Comparative tionship (QSAR) analysis of inhibitors of the human CYP2A6 Molecular Field Analysis (CoMFA) and Comparative Mole- and mouse CYP2A5 enzymes.21–23 The purpose of this study cular Similarity Indices Analysis (CoMSIA)) were developed was to rationally design novel potent inhibitors of CYP2A6. based on the CYP2A6 homology model. The stability of the Protein homology modelling and three-dimensional quan- models was tested by cross-validation with two and five titative structure–activity relationship (3D-QSAR) analysis groups (Table 1). As described previously,26 the cross- were carried out to determine which structural features are validation procedure provides a reliable picture of the important for the inhibition potency. Data from these predictivity of QSAR models. All the statistical values models were used to carry out structure-based screening obtained from our current CoMFA and CoMSIA models

Table 1 Statistics of CoMFA and CoMSIA PLS analyses

Model CoMFA CoMSIA No. a 2b c d 2e f a 2b c d 2e f cv q N SPRESS r S cv q N SPRESS r S

1. LSO 0.67 2 0.60 0.81 0.47 LSO 0.63 2 0.65 0.73 0.55 2. LSO 0.72 3 0.56 0.87 0.39 LSO 0.64 3 0.64 0.78 0.50 3.g LSO 0.74 4 0.55 0.91 0.32 LSO 0.68 4 0.61 0.83 0.44 4. LSO 0.73 5 0.56 0.95 0.28 LSO 0.63 5 0.63 0.87 0.40 5. LHO 0.64 2 0.63 0.81 0.47 LHO 0.44 2 6. LHO 0.68 3 0.60 0.87 0.39 LHO 0.53 3 7. LHO 0.68 4 0.60 0.91 0.32 LHO 0.61 4 8. LOO 0.74 4 0.55 0.91 0.31 LOO 0.66 4 0.62 0.83 0.44

Abbreviations: CoMFA, Comparative Molecular Field Analysis; CoMSIA, Comparative Molecular Similarity Indices Analysis; LHO, leave-half-out; LOO, leave-one-out; LSO, leave-some-out; PLS, partial least squares. aCross-validation method. bCross-validated correlation coefficient. cNumber of components. d SPRESS ¼ Standard deviation for the error of prediction. er2 ¼ Correlation coefficient. fStandard error of estimate. g Model used for prediction of IC50 values of new molecules.

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 330

Figure 1 CoMFA fields of CYP2A6 with nicotine as the template structure. CoMFA electrostatic fields: blue, negative-charge disfavoured area; red, negative-charge favoured area. CoMFA steric field: green, bulk favoured area; yellow, bulk disfavoured area.

Figure 3 Database query features. N1-(4-fluorophenyl)cyclopropane- 1-carboxamide is the reference structure. The tolerance for hydrophobic sites was 3.0 A˚ .

of metabolism in the five-member ring of nicotine, and the second one is close to the nitrogen atom of the pyridine ring. Both red and blue areas were associated with the presence of hydrogen bonds between inhibitor and protein. More detailed information about the position of the hydrogen bond acceptors is provided in the CoMSIA model (Figure 2). An acceptor group in the inhibitor, illustrated as a magenta region in the contour maps, increases inhibition Figure 2 CoMSIA acceptor fields for CYP2A6 model with nicotine as potency. Red colour denotes the area where hydrogen the template. Magenta, hydrogen acceptor favoured area of inhibitor; acceptors decrease inhibition potency. red, hydrogen acceptor disfavoured area of inhibitor. When assessing the CoMFA and CoMSIA fields together, it is evident that hydrogen bonding near the nitrogen atom of the pyrimidine ring in nicotine together with bulkiness near were significant (Table 1). The leave-some-out (LSO) and the five-member ring is crucial for potent inhibitors. A 3D leave-one-out (LOO) methods gave statistically the most database query was created to search for novel CYP2A6 significant CoMFA models. Since the LSO method was more inhibitor structures using these features. Specific hydropho- stringent than the LOO method, the LSO model with four bicity, acceptor/donor and planar characteristics that influ- components was chosen for the final model. Also the leave- ence inhibitor potency were set in this query (Figure 3). The half-out (LHO) method has been reported to give a good query features contained a plane and hydrophobic sites estimation of the model predictivity.27,28 The CoMSIA which are illustrated as the larger grey sphere in Figure 3. model yielded slightly lower statistical values than the The hydrophobic features in the query were set to select CoMFA model. compounds having moieties matching the hydrophobic nature of the CYP2A6 active site. The plane feature in the CoMFA and CoMSIA fields query defined its optimal position. The smaller grey sphere 3D colour contour maps of the CoMFA and CoMSIA models indicates the position of a required acceptor/donor atom. with nicotine as the reference structure are illustrated in Figures 1 and 2. Figure 1 shows the steric and electrostatic Database screening and analysis of selected compounds contour maps. Green and yellow areas denote steric Screening the Maybridge database returned 600 molecules. interactions. Bulkier groups in the green area in the vicinity After predicting their IC50 values using the 3D-QSAR models of the five-member ring enhance the potency of inhibitor. and visual inspection, several compounds were selected and The steric fields suggest that inhibition potency increases if their biological activity was tested (Table 2). All the selected the green area of the steric fields is occupied by a compounds had predicted IC50 values of less than 100 mM.In hydrophobic group. enzyme activity assay, 10 compounds showed IC50 values Red regions in Figure 1 indicate areas where an increase in lower than 10 mM. The most potent inhibitor was N1- electronegativity enhances inhibition potency and blue (4-fluorophenyl)cyclopropane-1-carboxamide (IC50 ¼ 90 nM). regions where electronegativity decreases inhibition po- Interestingly, this compound was over 1100 times more tency. There are two large red areas where negative charge is potent than several closely related structures, such as N1- favourable for inhibition. The first area is located at the site (4-fluorophenyl)acetamide (IC504100 mM). The presence of

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 331

Table 2 Structures and IC50 values of CYP2A6 inhibitors

Inhibitor Structure IC50 value (mM) Predicted IC50 value (mM)

N1-(4-Fluorophenyl)cyclopropane-1-carboxamidea 0.09 0.8

N F O

N1-(4-Fluorophenyl)acetamidea 4100 3.8 N F O

1-Benzo[b]thiophen-3-ylethan-1-onea 5.6 8 O

1-(3,4-Dimethylphenyl)piperazinea 4100 5.31 CH3

N CH3 HN

4-(2-Thienyl)pyrimidine-2-thiola 41.0 5.47 S N S N

5-Phenyl-1,2,3-thiadiazolea 3.42 5.5 N N S

2-(2-Thienylmethyl)-4,5-dihydro-1,3-thiazol-4-onea 13.9 5.7 S S

O N

N,N-Dimethyl-4-(4-pyridinyl)anilinea 4100 100

N N

4-(4-Pyridylmethyl)anilinea 24.4 80

N NH2

4-(4-Chlorobenzyl)pyridineb 1.85 10

N Cl

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 332

Table 2 Continued

Inhibitor Structure IC50 value (mM) Predicted IC50 value (mM)

4-(4-Chlorobenzoyl)pyridineb 21.9 70 O

N Cl

4-Benzylpyridineb 1.83 40

4-Benzoylpyridineb 21.9 16.5 O

4-(4-Nitrobenzyl)pyridineb 12.9 28

N NO2 p-Dimethylaminobenzaldehydeb 0.45 30 O N

4-Bromobenzaldehydeb 19.6 28 O Br

4-Bromobenzylbromideb 20.7 49.8 Br Br

4-Chlorobenzylamineb 0.45 15 H2N Cl

4-Fluorobenzylamineb 18 25 H2N F

4-Bromobenzylamineb 0.54 4.1 H2N Br

2-Chlorobenzylamineb 0.91 3.2 Cl

H2N

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 333

Table 2 Continued

Inhibitor Structure IC50 value (mM) Predicted IC50 value (mM)

3-Fluorobenzylamineb 4.1 12 F

H2N

a12 compounds chosen based on virtual screening. b13 additional compounds.

Table 3 Inhibition of CYP1A2, CYP2A6, CYP2B6, CYP2C8, CYP2C9, CYP2C19, CYP2D6, CYP2E1 and CYP3A4 by selected inhibitors

CYP 4-bromo- p-Dimethyl-amino- 2-chloro- N1-(4-Fluorophenyl) benzylamine benzaldehyde benzylamine cyclopropane-1-carboxamide

1A2 69.0 4100 46.6 100 2A6 1.47 4.75 0.45 0.41 2B6 10.0 4100 7.63 5.0 2C8 4100 4100 4100 4100 2C9 4100 4100 13.9 10.0 2C19 5-OH-OMEa 4100 4100 46.6 69.6 dem-OMEb 4100 4100 55.6 98.8

2D6 55.6 4100 56.5 4100 2E1 86.9 62.6 17.7 10.0 3A4 a-OH-MDZc 4100 4100 4100 4100 6b-TESd 4100 4100 4100 4100 SO-OMEe 4100 4100 4100 4100 3-OH-OMEf 4100 4100 4100 4100

Abbreviations: MDZ, midazolam; OME, omeprazole; TES, testosterone.

The numbers are IC50 values in the N-in-one multisubstrate assay. aOmeprazole 5-hydroxylation. bOmeprazole demethylation. cMidazolam a-hydroxylation. dTestosterone 6b-hydroxylation. eOmeprazole sulphoxidation. fOmeprazole 3-hydroxylation. the hydrophobic cyclopropyl group in N1-(4-fluorophenyl)- The CYP selectivity of four most potent inhibitors was cyclopropane-1-carboxamide increased its inhibition po- characterised with human liver microsomes as the enzyme tency compared with N1-(4-fluorophenyl)acetamide. The source. As shown in Table 3, these compounds (N1- bulky cyclopropyl and fluorobenzyl groups are located near (4-fluorophenyl)cyclopropane-1-carboxamide, 4-bromobenzyla- the green areas in the CoMFA map (Figure 1), indicating mine, 2-chlorobenzylamine, p-dimethylaminobenzaldehyde) increased inhibition potency. inhibited preferentially CYP2A6. These compounds, except In addition, 13 other chemicals including pyridine, p-dimethylaminobenzaldehyde, inhibited also CYP2B6 ac- benzaldehyde and benzylamine derivatives were measured tivity at higher concentrations. The CYP selecti- for their inhibition potency (Table 2). Benzylamine deriva- vity ratios are listed in Table 4. Relative CYP selectivity tives were also quite potent inhibitors, especially those with ratios are calculated as IC50(CYPX)/IC50(CYP2A6). Judging chloro or bromo substitutions at the meta position as well as from the selectivity ratios, the most selective inhibitor was chloro substitutions at the ortho-position, since their IC50 2-chlorobenzylamine, which was 17 times more potent in values were below 1 mM. Interestingly, 4-benzylpyridine inhibiting CYP2A6 than CYP2B6. This compound inhibited derivatives were 10 times more potent inhibitors than the other CYP forms much less, as the selectivity ratios for other corresponding 4-benzoylpyridines, indicating that the CYPs varied between 30.9 and 4222 (Table 4). Substrate 2 presence of a carbonyl group at the sp carbon reduces concentrations were close or clearly below the apparent Km inhibition potency. values for indicated CYP activities. At these low concentrations

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 334

Table 4 Relative CYP selectivity ratios

CYP 4-bromo- p-Dimethyl-amino- 2-chloro- N1-(4-Fluorophenyl) benzylamine benzaldehyde benzylamine cyclopropane-1-carboxamide

1A2/2A6 47 421.1 103.6 244 2B6/2A6 6.80 421.1 17.0 12.2 2C8/2A6 468.0 421.1 4222.2 4244 2C9/2A6 468.0 421.1 30.9 24.4 2C19/2A6 5-OH-OMEa 468.0 421.1 103.6 169.8 dem-OMEb 468.0 421.1 123.6 241 2D6/2A6 37.8 421.1 125.6 4244 2E1/2A6 59.1 13.2 39.3 24.4 3A4/2A6 a-OH-MDZc 468.0 421.1 4222 4244 6b-TESd 468.0 421.1 4222 4244 SO-OMEe 468.0 421.1 4222 4244 3-OH-OMEf 468.0 421.1 4222 4244

Abbreviations: MDZ, midazolam; OME, omeprazole; TES, testosterone.

The ratio is calculated as IC50 (CYPX)/ IC50 (CYP2A6). The IC50 values are from Table 3. aOmeprazole 5-hydroxylation. bOmeprazole demethylation. cMidazolam a-hydroxylation. dTestosterone 6b-hydroxylation. eOmeprazole sulphoxidation. fOmeprazole 3-hydroxylation.

Figure 4 CoMFA fields of the model combined with crystal structure with nicotine as the template structure. CoMFA electrostatic fields: blue, negative-charge disfavoured area; red, negative-charge favoured area. CoMFA steric field: green, bulk favoured area; yellow, bulk disfavoured area.

the IC50 values obtained are not too far from the experi- 29,30 mentally obtained Ki values. The CYP2A6 crystal structure with coumarin, metho- Figure 5 CoMFA fields of the model combined with crystal structure 24 trexate, N,N-dimethyl(5-(pyridin-3-yl)furan-2-yl)methamine, with coumarin as the template structure. CoMFA electrostatic fields: N-dimethyl(5-(pyridin-3-yl)furan-2-yl)methamine, and di- blue, negative-charge disfavoured area: red, negative-charge favoured methyl(5-(pyridin-3-yl)furan-2-yl)methamine25 and our homo- area. CoMFA steric field: green, bulk favoured area; yellow, bulk logy model differed with respect to the amino acid forming a disfavoured area. hydrogen bond to the substrate. We therefore generated a new CoMFA model using the coumarin-bound CYP2A6 crystal structure.24 Before creating the model, the structure was relaxed of the hydrogen bond. The main difference between these by molecular dynamics simulation (GROMACS). The new two models was the position of the second negative charge CoMFA model was created in the same way as the initial one favourable (red) region, which in the new model was located and all the statistical values obtained were significant (data not above the five-member ring of nicotine. The interaction shown). with the substrate (coumarin) and CYP2A6 protein is shown 3D colour contour maps of the new CoMFA model with in Figure 5. The map includes an area where negative charge nicotine as the reference structure are illustrated in Figures 4 was linked with increasing inhibition potency near Asn297. and 5. The new model showed more accurately the location The CYP2A6 active site is surrounded by mostly lipophilic

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 335

amino acids. A steric favourable field was located near these hydrophobic amino acids.24,25 The physicochemical proper- hydrophobic amino acids indicating the importance of ties of the most potent inhibitor (N1-(4-fluorophenyl)cyclo- hydrophobic interactions with inhibitor and the protein. propane-1-carboxamide) was optimal and effectively filled the CYP2A6 active site. Docking of this molecule in the crystal structure showed that it interacts effectively with Discussion the hydrophobic amino acids in the active site (data not shown). The inhibitor also includes a donor NH group This study shows that 3D-QSAR analysis can be successfully which can form hydrogen bonding with Asn297. This kind applied in the design of potent and selective inhibitors of of donor feature has not been considered to be important for the nicotine metabolising CYP2A6 enzyme. We have shown CYP2A6 inhibition previously. The benzylamine derivatives previously that 3D-QSAR methods are useful approaches if (4-chlorobenzylamine, 4-bromobenzylamine and 2-chloro- one wishes to evaluate the structural features that are benzylamine) were also potent inhibitors. All these com- important for inhibition potency against CYP enzymes pounds contain an amine group and an aromatic ring both and other targets.23,27,28,31,32 In this study, we showed that of which are believed to be important for inhibition novel CYP2A6 inhibitor structures can be designed using potency. The amine group in the benzylamine derivatives CoMFA based on homology modelling. As CYP2A6 crystal is the donor in hydrogen bonding with Asn 297, thereby structures became available during this work, we were able increasing inhibition potency. Optimal halogen substitution to ascertain the accuracy of the CoMFA model using the and hydrophobicity features of inhibitors also increase crystal structure data. inhibition potency. These compounds are also selective The 3D-QSAR models created in this study add to the data inhibitors of CYP2A6 with some effect on CYP2B6, another we have published previously.21–23 Use of the crystal nicotine-metabolising enzyme.6 structure 24 in the alignment of ligands improved There is considerable interest in the development of the current QSAR model since the protein–ligand interac- clinically useful inhibitors of the CYP2A6 enzyme.12,16,17 tions are taken into account more precisely. Especially in None of the currently known CYP2A6 inhibitors are suitable the crystal structure-based CoMFA model, electrostatic for routine clinical use and thus there is a need to develop and acceptor fields revealed where the hydrogen bond novel inhibitors with suitable potency, selectivity and acceptors or donors should be located in the inhibitor toxicity profiles. In conclusion, protein homology model- molecule. The influence of hydrophobicity on inhibition ling, 3D-QSAR and chemical database screening proved to potency was emphasised in both the CoMFA and CoMSIA be valuable approaches in identifying the structural features steric and hydrophobic fields. These fields indicated that that are important for CYP2A6 inhibition and searching for the presence of hydrophobic bulky groups near the novel inhibitor candidates. One of the compounds found in metabolising site of nicotine led to an increase in inhibition this study, N1-(4-fluorophenyl)cyclopropane-1-carboxa- potency. These fields were useful when the search query mide, proved to be a potent and reasonably selective was planned. The predicted IC50 values for most of the inhibitor and may represent a lead compound in the design new inhibitors agreed quite well with the actual values of inhibitors to be used as adjuvant drugs in smoking although their structures did not resemble the ones in the cessation therapy. training set. In the crystal structure, Asn297 acts as a hydrogen bond Materials and methods donor to the carbonyl oxygen of coumarin. Asn279 is surrounded by mostly lipophilic amino acids which limit Homology model the size of potent inhibitors.24 The present crystal structure- Initially, there was sufficient information available about based 3D-QSAR model also showed that a hydrogen bond the mouse CYP2A5 protein to construct a CYP2A5 compara- with Asn297 and van der Waals interactions between the tive model based on the CYP2C5 crystal structure.33 Details hydrophobic amino acids and inhibitor is an important of the CYP2A5 homology model were published recently.34 determinant of inhibition potency. The CYP2A6 crystal This model was used as a template for the current CYP2A6 structure shows that in suicide inhibitors electron-rich model because of the relatively high sequence similarity regions coordinate to haem iron.25 In the current CoMFA (82%) between the CYP2A6 and CYP2A5 proteins. The final models, the negative charge favourable (red) regions above CYP2A6 model was validated by checking its stereochemical the haem iron indicate that inhibition potency correlates accuracy and docking procedure as well as performing a with electron density. molecular dynamics simulation. The novel inhibitor molecules identified in this study differ in structure from the recently reported nicotine derivatives.25 Some of the novel compounds were very Docking potent, especially N1-(4-fluorophenyl)cyclopropane-1- The crucial step in constructing a 3D-QSAR model is the carboxamide. The cyclopropyl group in this compound is alignment of the training set molecules. The training set of very important for the interaction since it increases hydro- chemicals used in these models is an extension of our phobicity. The CYP2A6 crystal structure revealed that the previously published dataset.23,35 The training set of enzyme has a very compact active site containing several chemicals consisted of 80 compounds covering IC50 values

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 336

between 0.55 mM (2-bromonaphthalene) and 35.4 mM (coti- be a reliable indicator of model complexity, and thus, it nine). Since the coordinates of the active site of the CYP2A6 helps to avoid over-fitting due to too many PLS compo- protein were known from the homology model, we applied nents. Specifically, a number of PLS components yielding a 36,37 2 2 the virtual docking tool GOLD to fit the ligand (dq /dryy’) slope near unity should be optimal. In the case of molecules into the active site. Virtual docking tools generate the final CoMFA models, both cross-validation and pro- different possible conformations and orientations of the gressive scrambling data suggested that four components ligands at the active site and evaluate the free energy of were optimal. binding by a scoring function. In the present study, the empirical scoring function ChemScore was applied38 since it has been shown to reproduce correctly the complex Database screening geometry in the crystal structures of other CYP proteins.39 A search query was created for chemical database mining The best docked conformations according to the ChemScore based on the properties of both the 3D-QSAR models and scoring function were used to construct the 3D-QSAR the CYP2A6 active site. Naphthalene was used as template to models. define the query features containing a spatial plane and hydrophobic sites (tolerance 3.0 A˚ ). The search query contained also one donor atom (either in the protein or CoMFA and CoMSIA analysis inhibitor) near Thr106 (tolerance 0.5 A˚ ). The Lipinski rule Both CoMFA36,40 and CoMSIA41–43 were created for the of five45 was applied with the following cut-offs for the CYP2A6 enzyme. These methods correlate the variation in compounds: maximum weight 400 g molÀ1, maximum of the property fields in a set of molecules, such as steric, four hydrogen bond donors and maximum of five hydrogen electrostatic or hydrogen bond acceptor and donor fields, bond acceptors. Flexible 3D search was performed using the with the variation in their biological activity. The outcome database searching program Unity 4.3.1 implemented in of these methods is a statistical model that can be used Sybyl. The Maybridge database (www.maybridge.com), con- to predict the inhibition potency of unknown compounds taining approximately 60 000 compounds, was screened for CYP2A6. Furthermore, the contour fields that contribute with a return of 600 hit compounds. The hit compounds significantly to the models yield insights into the CYP2A6– were docked into the active site of CYP2A6 using the GOLD 36,37 inhibitor interactions. The CoMFA and CoMSIA methods program as described before. The IC50 values for these differ in the way that the property fields are calculated. molecules were predicted with the CoMFA model. Our own In CoMFA, electrostatic and steric interaction fields were database, consisting of potent and weak inhibitors, was used calculated, and in CoMSIA, steric, electrostatic, hydrophobic, as positive control to test this virtual screen before the actual hydrogen bond acceptor and hydrogen bond donor similar- screening process. ity indices were calculated using the SYBYL default settings. The hit compounds having predicted IC50 values in the The resulting partial least squares (PLS) models were range of 0.8–100 mM were chosen for further visual analysis. validated using the LOO method, 2 random-group cross- The location of the docked conformations at the active site, validation method LHO or 5 random-group validation as well as their ability to match the requirements of the method LSO. Both LHO and LSO cross-validation proce- search queries and CoMFA and CoMSIA fields, was visually dures were repeated 20 times, and the average statistical evaluated. Nine compounds were selected for in vitro values were calculated. Also, a progressive scrambling inhibition testing based on the predicted protein–ligand method was applied to validate the models, because this conformation and new structural features. In addition, 13 provides valuable information when one analyses a large other compounds were obtained and their IC50 values were dataset containing redundant information. This novel determined (Table 2). validation technique was developed to address the overly optimistic cross-validation or response randomisation results for redundant data sets. In this approach, small Chemicals random perturbations are introduced into a data set. This N1-(4-fluorophenyl)cyclopropane-1-carboxamide, N1-(4-fluoro- causes the nominal predictivity of unstable models to fall off phenyl)acetamide, 1-(3,4-dimethylphenyl)piperazine, 4-(2- rapidly, whereas robust models are relatively stable.44 thienyl)pyrimidine-2-thiol, 5-phenyl-1,2,3-thiadiazole, 2-(2- Progressive scrambling of the biological data produces three thienylmethyl)-4,5-dihydro-1,3-thiazol-4-one, 4-(4-pyridyl- statistics. Q2 and cSDEP values characterise the predictivity methyl)aniline, 1-benzo[b]thiophen-3-ylethan-1-one and N, of the model and the calculated cross-validated standard N-dimethyl-4-(4-pyridinyl)aniline were ordered from May- error, respectively. The current CoMFA models were shown bridge. In addition, 13 other compounds were ordered and to be stable, and the adjusted statistical values Q2 and cSDEP measured for inhibition potency. 4-(4-Chlorobenzoyl)py- were comparable with the corresponding q2 and SPRESS of ridine, 4-benzylpyridine, 4-benzoylpyridine, 4-(4-nitrobenzyl) cross-validation. The third statistic from progressive scram- pyridine, 4-chlorobenzylamine, 4-bromobenzaldehyde, 4-bro- bling is the instantaneous slope of the predictivity with mobenzylbromide, 4-(4-chlorobenzyl)pyridine, p-dimethy- 2 2 respect to the degree of perturbation (dq /dryy’). It depicts laminobenzaldehyde, 2-(4-chlorophenyl)ethylamine, 4-bromo- the model sensitivity to perturbation at a critical threshold benzylamine, 3-fluorobenzylamine and 4-fluorobenzylamine level of perturbation (here 0.85).44 This value is reported to were purchased from Sigma (St. Louis, MO, USA) (Table 2).

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 337

Biochemical assays excision, cut into pieces, snap-frozen in liquid nitrogen The coumarin 7-hydroxylation activity assay is based on the and stored at À80 1C until the microsomes were prepared detection of fluorescence emitted by 7-hydroxycoumarin as by standard differential ultracentrifugation. A weight- described.46 We have adapted this method to a 96-well plate balanced microsomal pool of seven liver microsomal format to facilitate higher throughput inhibition analysis. preparations which have been extensively characterised to In each well, the 100 ml incubation volume contained 50 mM be used for the primary screening (sufficient model Tris-HCl buffer (pH 7.4), 5.0 mM MgCl2,10mM coumarin, activities, no known polymorphisms, expected effects of 20 mg microsomal proteins and 0.3 mM nicotinamide ade- model inhibitors, quantification of CYPs by western blot- nine dinucleotide phosphate-oxidase (NADPH). The reac- ting) was employed. The final microsomal pellet was tion was initiated by addition of NADPH, incubated at 37 1C suspended in 100 mM phosphate buffer pH 7.4. The use of for 10 min and terminated by adding 60 ml 10% trichloro- human liver microsomes as an enzyme source is justified by acetic acid. Immediately before the measurement, 140 ml the fact that the coumarin 7-hydroxylation reaction is 1.6 M glycine–NaOH buffer (pH 10.4) was added. The mediated exclusively by the CYP2A6 enzyme in human formed fluorescence was measured with a Victor2 plate liver. Therefore, in vitro inhibition of coumarin 7-hydroxyla- counter (Perkin Elmer Life Sciences Wallac, Turku, Finland) tion in human liver microsomes reflects solely inhibition at 355 nm excitation and 460 nm emission. The linearity of of CYP2A6 enzymes with no participation by other CYP the reaction with respect to incubation time and micro- forms. somal protein concentration was determined. Several control incubations were carried out to determine the effect of quenching by the inhibitors and other interfering factors. Abbreviations Each inhibitor was pre-screened using inhibitor concentra- CYP cytochrome P450 tions ranging from 0.1 to 1000 mM. The actual IC50 values CoMFA Comparative Molecular Field Analysis were determined using narrower inhibitor concentration CoMSIA Comparative Molecular Similarity Indices Analysis LHO leave-half-out ranges with five to seven concentrations. All IC50 values were determined in duplicate both from human liver LOO leave-one-out LSO leave-some-out microsomal preparations and recombinant CYP2A6 enzyme. PLS partial least squares Baculovirus-insect cell expressed human CYP2A6 was pur- 3D-QSAR three-dimensional quantitative structure–activity relationship chased from BD Biosciences Discovery Labware (Bedford, MA, USA) and used according to the manufacturer’s instructions. The N-in-one multisubstrate CYP selectivity assay has Acknowledgments been previously described in detail.29,30 Briefly, each incubation mixture contained 0.5 mg of microsomal We thank Ms. Hannele Jaatinen for her excellent technical help and Dr Ewen MacDonald for his help in preparing the manuscript. protein per ml, 0.1 M phosphate buffer (pH 7.4), 1 mM NADPH and all the 10 probe substrates. Substrates, their concentrations and target enzymes for the incubations were Duality of interest melatonin (4 mM, CYP1A2), coumarin (2 mM, CYP2A6), bupropion (1 mM, CYP2B6), amodiaquine (2 mM, CYP2C8), None declared. tolbutamide (4 mM, CYP2C9), omeprazole (2 mM, CYP2C19 and CYP3A4), dextromethorphan (0.2 mM, CYP2D6), References chlorzoxazone (6 mM, CYP2E1), midazolam (0.4 mM, CYP3A4) and testosterone (1 mM, CYP3A4). The reaction mixture, 1 Ezzati M, Hoorn SV, Rodgers A, Lopez AD, Mathers CD, Murray CJ et al. in a final volume of 200 ml, was pre-incubated for 2 min at Estimates of global and regional potential health gains from reducing þ 37 1C in a shaking incubator block (Eppendorf multiple major risk factors. Lancet 2003; 362: 271–280. 2 Benowitz NL. Pharmacology of nicotine: addiction and therapeutics. Thermomixer 5436, Hamburg, Germany) before the reac- Annu Rev Pharmacol Toxicol 1996; 36: 597–613. tion was initiated by addition of NADPH. Each reaction was 3 Berkman CE, Park SB, Wrighton SA, Cashman JR. In vitro–in vivo terminated after 20 min by adding 100 ml of ice-cold correlations of human (S)-nicotine metabolism. Biochem Pharmacol acetonitrile. Analysis was carried out using a LC/MS/MS 1995; 50: 565–570. 4 Nakajima M, Yamamoto T, Nunoya K, Yokoi T, Nagashima K, Inoue K method optimised for Micromass Quattro Micro triple et al. Role of human cytochrome P4502A6 in C-oxidation of nicotine. quadrupole mass spectrometer (Micromass Corp., Drug Metab Dispos 1996; 24: 1212–1217. Altrincham, UK).29 5 Messina ES, Tyndale RF, Sellers EM. A major role for CYP2A6 in nicotine Human liver samples used in this study were obtained C-oxidation by human liver microsomes. J Pharmacol Exp Ther 1997; 282: 1608–1614. from the University Hospital of Oulu as a surplus from 6 Hukkanen J, Jacob III P, Benowitz NL. Metabolism and disposition kidney transplantation donors. The patients have been kinetics of nicotine. Pharmacol Rev 2005; 57: 79–115. described previously.47 The collection of surplus tissue 7 Oscarson M. Genetic polymorphisms in the cytochrome P450 2A6 was approved by the Ethics Committee of the Medical (CYP2A6) gene: implications for interindividual differences in nicotine metabolism. Drug Metab Dispos 2001; 29: 91–95. Faculty of the University of Oulu, Finland. The livers 8 Raunio H, Rautio A, Gullsten H, Pelkonen O. Polymorphisms of CYP2A6 were transferred to ice immediately after the surgical and its practical consequences. Br J Clin Pharmacol 2001; 52: 357–363.

The Pharmacogenomics Journal Rational design of CYP2A6 inhibitors M Rahnasto et al 338

9 Xu C, Goodz S, Sellers EM, Tyndale RF. CYP2A6 genetic variation and 29 Turpeinen M, Uusitalo J, Jalonen J, Pelkonen O. Multiple P450 substrates potential consequences. Adv Drug Deliv Rev 2002; 54: 1245–1256. in a single run: rapid and comprehensive in vitro interaction assay. Eur J 10 Nakajima M, Kuroiwa Y, Yokoi T. Interindividual differences in nicotine Pharm Sci 2005; 24: 123–132. metabolism and genetic polymorphisms of human CYP2A6. Drug 30 Turpeinen M, Korhonen LE, Tolonen A, Uusitalo J, Juvonen R, Raunio H Metab Rev 2002; 34: 865–877. et al. Cytochrome P450 (CYP) inhibition screening: comparison of three 11 Malaiyandi V, Sellers EM, Tyndale RF. Implications of CYP2A6 genetic tests. Eur J Pharm Sci 2006; 29: 130–138. variation for smoking behaviors and nicotine dependence. Clin 31 Korhonen LE, Rahnasto M, Mahonen NJ, Wittekindt C, Poso A, Juvonen Pharmacol Ther 2005; 77: 145–158. RO et al. Predictive three-dimensional quantitative structure-activity 12 Tyndale RF, Sellers EM. Variable CYP2A6-mediated nicotine metabolism relationship of cytochrome P450 1A2 inhibitors. J Med Chem 2005; 48: alters smoking behavior and risk. Drug Metab Dispos 2001; 29: 3808–3815. 548–552. 32 Korhonen LE, Turpeinen M, Rahnasto M, Wittekindt C, Poso A, 13 Gu DF, Hinks LJ, Morton NE, Day IN. The use of long PCR to confirm Pelkonen O et al. New potent and selective cytochrome P450 2B6 three common alleles at the CYP2A6 locus and the relationship between (CYP2B6) inhibitors based on three-dimensional quantitative structure- genotype and smoking habit. Ann Hum Genet 2000; 64: 383–390. activity relationship (3D-QSAR) analysis. Br J Pharmacol 2007; 150: 932–942. 14 Schoedel KA, Hoffmann EB, Rao Y, Sellers EM, Tyndale RF. Ethnic 33 Williams PA, Cosme J, Sridhar V, Johnson EF, McRee DE. Microsomal variation in CYP2A6 and association of genetically slow nicotine cytochrome P450 2C5: comparison to microbial P450 s and unique metabolism and smoking in adult Caucasians. Pharmacogenetics 2004; features. J Inorg Biochem 2000; 81: 183–190. 14: 615–626. 34 Stahl GR, Holtje HD. Development of models for cytochrome P450 2A5 15 Malaiyandi V, Lerman C, Benowitz NL, Jepson C, Patterson F, Tyndale as well as two of its mutants. Pharmazie 2005; 60: 247–253. RF. Impact of CYP2A6 genotype on pretreatment smoking behaviour 35 Juvonen RO, Gynther J, Pasanen M, Alhava E, Poso A. Pronounced and nicotine levels from and usage of nicotine replacement therapy. differences in inhibition potency of lactone and non-lactone com- Mol Psychiatry 2006; 11: 400–409. pounds for mouse and human coumarin 7-hydroxylases (CYP2A5 and 16 Sellers EM, Kaplan HL, Tyndale RF. Inhibition of cytochrome P450 2A6 CYP2A6). Xenobiotica 2000; 30: 81–92. increases nicotine’s oral bioavailability and decreases smoking. Clin 36 Jones G, Willett P, Glen RC, Leach AR, Taylor R. Development and Pharmacol Ther 2000; 68: 35–43. validation of a genetic algorithm for flexible docking. J Mol Biol 1997; 17 Zhang W, Kilicarslan T, Tyndale RF, Sellers EM. Evaluation of 267: 727–748. methoxsalen, tranylcypromine, and tryptamine as specific and selective 37 Verdonk ML, Cole JC, Hartshorn MJ, Murray CW, Taylor RD. Improved CYP2A6 inhibitors in vitro. Drug Metab Dispos 2001; 29: 897–902. protein-ligand docking using GOLD. Proteins 2003; 52: 609–623. 18 Draper AJ, Madan A, Parkinson A. Inhibition of coumarin 7-hydroxylase 38 Eldridge MD, Murray CW, Auton TR, Paolini GV, Mee RP. Empirical activity in human liver microsomes. Arch Biochem Biophys 1997; 341: 47–61. scoring functions: I. the development of a fast empirical scoring 19 Pelkonen O, Rautio A, Raunio H, Pasanen M. CYP2A6: a human function to estimate the binding affinity of ligands in receptor coumarin 7-hydroxylase. Toxicology 2000; 144: 139–147. complexes. J Comput Aided Mol Des 1997; 11: 425–445. 20 Denton TT, Zhang X, Cashman JR. 5-substituted, 6-substituted, and 39 de Graaf C, Vermeulen NP, Feenstra KA. Cytochrome p450 in unsubstituted 3-heteroaromatic pyridine analogues of nicotine as selec- silico: an integrative modeling approach. J Med Chem 2005; 48: tive inhibitors of cytochrome P-450 2A6. JMedChem2005; 48:224–239. 2725–2755. 21 Poso A, Gynther J, Juvonen R. A comparative molecular field analysis of 40 Cramer RI, Patterson D, Bunce J. Comparative molecular field analysis cytochrome P450 2A5 and 2A6 inhibitors. J Comput Aided Mol Des (CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. 2001; 15: 195–202. J Am Chem Soc 1988; 110: 5959–5960. 22 Rahnasto M, Raunio H, Poso A, Juvonen RO. More potent inhibition 41 Klebe G, Abraham U, Mietzner T. Molecular similarity indices in a of human CYP2A6 than mouse CYP2A5 enzyme activities by derivatives comparative analysis (CoMSIA) of drug molecules to correlate and of phenylethylamine and benzaldehyde. Xenobiotica 2003; 33: predict their biological activity. J Med Chem 1994; 37: 4130–4146. 529–539. 42 Bohm M, St rzebecher J, Klebe G. Three-dimensional quantitative 23 Rahnasto M, Raunio H, Poso A, Wittekindt C, Juvonen RO. Quantitative structure-activity relationship analyses using comparative molecular structure-activity relationship analysis of inhibitors of the nicotine field analysis and comparative molecular similarity indices analysis to metabolizing CYP2A6 enzyme. J Med Chem 2005; 48: 440–449. elucidate selectivity differences of inhibitors binding to trypsin, 24 Yano JK, Hsu MH, Griffin KJ, Stout CD, Johnson EF. Structures of human thrombin, and factor Xa. J Med Chem 1999; 42: 458–477. microsomal cytochrome P450 2A6 complexed with coumarin and 43 Klebe G, Abraham U. Comparative molecular similarity index analysis methoxsalen. Nat Struct Mol Biol 2005; 12: 822–823. (CoMSIA) to study hydrogen-bonding properties and to score 25 Yano JK, Denton TT, Cerny MA, Zhang X, Johnson EF, Cashman JR. combinatorial libraries. J Comput Aided Mol Des 1999; 13: 1–10. Synthetic inhibitors of cytochrome P-450 2A6: inhibitory activity, 44 Clark RD, Fox PC. Statistical variation in progressive scrambling. difference spectra, mechanism of inhibition, and protein cocrystalliza- J Comput Aided Mol Des 2004; 18: 563–576. tion. J Med Chem 2006; 49: 6987–7001. 45 Lipinski CA, Lombardo F, Dominy BW, Feeney PJ. Experimental and 26 Hawkins DM, Basak SC, Mills D. Assessing model fit by cross-validation. computational approaches to estimate solubility and permeability in drug J Chem Inf Comput Sci 2003; 43: 579–586. discovery and development settings. Adv Drug Deliv Rev 2001; 46:3–26. 27 Tervo AJ, Nyronen TH, Ronkko T, Poso A. A structure-activity relation- 46 Aitio A. A simple and sensitive assay of 7-ethoxycoumarin deethylation. ship study of catechol-O-methyltransferase inhibitors combining mole- Anal Biochem 1978; 85: 488–491. cular docking and 3D QSAR methods. J Comput Aided Mol Des 2003; 17: 47 Taavitsainen P, Juvonen R, Pelkonen O. In vitro inhibition of cytochrome 797–810. P450 enzymes in human liver microsomes by a potent CYP2A6 28 Salo OM, Savinainen JR, Parkkari T, Nevalainen T, Lahtela-Kakkonen M, inhibitor, trans-2-phenylcyclopropylamine (tranylcypromine), and its Gynther J et al. 3D-QSAR studies on cannabinoid CB1 receptor agonists: nonamine analog, cyclopropylbenzene. Drug Metab Dispos 2001; 29: G-protein activation as biological data. J Med Chem 2006; 49: 554–566. 217–222.

The Pharmacogenomics Journal