Why Is Quinidine an Inhibitor of Cytochrome P450 2D6?

We have previously shown that Phe120, Glu216, and Asp301 in the active site of cytochrome P450 2D6 (CYP2D6) play a key role in substrate recognition by this important drug-metabolizing enzyme (Paine, M. J., McLaughlin, L. A., Flanagan, J. U., Kemp, C. A., Sutcliffe, M. J., Roberts, G. C., and Wolf, C. R. (2003) J. Biol. Chem. 278, 4021–4027 and Flanagan, J. U., Maréchal, J.-D., Ward, R., Kemp, C. A., McLaughlin, L. A., Sutcliffe, M. J., Roberts, G. C., Paine, M. J., and Wolf, C. R. (2004) Biochem. J. 380, 353–360). We have now examined the effect of mutations of these residues on interactions of the enzyme with the prototypical CYP2D6 inhibitor, quinidine. Abolition of the negative charge at either or both residues 216 and 301 decreased quinidine inhibition of bufuralol 1′-hydroxylation and dextromethorphan O-demethylation by at least 100-fold. The apparent dissociation constants (Kd) for quinidine binding to the wild-type enzyme and the E216D and D301E mutants were 0.25–0.50 μm. The amide substitution of Glu216 or Asp301 resulted in 30–64-fold increases in the Kd for quinidine. The double mutant E216Q/D301Q showed the largest decrease in quinidine affinity, with a Kd of 65 μm. Alanine substitution of Phe120, Phe481,or Phe483 had only a minor effect on the inhibition of bufuralol 1′-hydroxylation and dextromethorphan O-demethylation and on binding. In contrast to the wild-type enzyme, a number of the mutants studied were found to be able to metabolize quinidine. E216F produced O-demethylated quinidine, and F120A and E216Q/D301Q produced both O-demethylated quinidine and 3-hydroxyquinidine metabolites. Homology modeling and molecular docking were used to predict the modes of quinidine binding to the wild-type and mutant enzymes; these were able to rationalize the experimental observations.

Human cytochrome P450 2D6 (CYP2D6) 4 plays a central role in drug metabolism, metabolizing Ͼ30% of the most commonly prescribed drugs (1). The CYP2D6 gene is highly polymorphic, leading to wide interindividual and ethnic differences in CYP2D6-mediated drug metabolism (2)(3)(4). Cytochrome P450-drug and drug-drug interactions involving CYP2D6 ligands are thus a prime consideration in the development of new drugs, emphasizing the importance of a detailed understanding of the factors that govern the substrate specificity of this enzyme.
Quinidine is not metabolized by CYP2D6 and has long been established as a potent competitive inhibitor of the enzyme (5)(6)(7)(8)(9). The fact that quinidine is an inhibitor rather than a substrate is intriguing because it produces a classical type I binding spectrum with CYP2D6 (10) that is usually associated with the binding of substrate molecules (11). In addition, quinidine possesses a number of features normally associated with CYP2D6 substrates, including a basic nitrogen atom, a flat hydrophobic region, and a negative molecular electrostatic potential (12). Studies of the relationship between structure and inhibitory activity for quinidine and its (less potent) stereoisomer quinine have been reported (13), but the protein-ligand interactions that are responsible for the fact that quinidine can bind tightly, but not in an orientation favorable for catalysis, have not hitherto been established.
Recent models of the active site of CYP2D6 (e.g. Ref. 14) suggest that two carboxylate groups (at Glu 216 and Asp 301 ) may play key roles in the recognition of substrates containing a basic nitrogen atom, and support for this has come from mutagenesis experiments (15)(16)(17). It has also been suggested that the aromatic residues Phe 120 , Phe 481 , and Phe 483 may have roles in substrate binding through -interactions with the planar hydrophobic regions common to many CYP2D6 substrates (10,14,18,19). Here, we describe studies of a series of mutants of these five residues aimed at investigating their role in quinidine binding and in determining whether quinidine is a substrate or an inhibitor of this important drug-metabolizing enzyme.
Mutagenesis and Expression in E. coli-The Glu 216 and Asp 301 mutants of CYP2D6 used in this study were constructed and expressed in E. coli along with human cytochrome P450 reductase as described previously (17). To obtain the remaining mutants, site-directed mutagenesis was performed following the single-stranded DNA template method (20) using pB81 as a template and the dut Ϫ ung Ϫ E. coli strain CJ236 along with an appropriate mutagenic oligonucleotide: F120A, 3Ј-ata gcg cgc cag agc cac ccc ttg gga-5Ј; F481A, 3Ј-cac cag gaa agc agc gac acc atg gtg-5Ј; and F483A, 3Ј-cac cag gaa agc agc gac acc atg gtg-5Ј. Note that the oligonucleotide sequences are reverse-complemented. Once the presence of the desired mutation was confirmed by automated DNA sequencing, the mutants were coexpressed with human cytochrome P450 reductase as described above. Varying quantities of cytochrome P420, reflecting different degrees of stability in the presence of dithionite, were observed for the different mutants. E216Q, D301E, D301Q, F481A, and F483A had P450:P420 peak area ratios of ϳ9:1; E216N, E216Q/D301Q, and F120A had ratios of ϳ1:1; and E216F, E216A, and D301N had ratios of ϳ1:5.
Quinidine Inhibition of Bufuralol 1Ј-Hydroxylation and Dextromethorphan O-Demethylation-Incubations were carried out in triplicate at 37°C with shaking in 300 l of 50 mM potassium phosphate (pH 7.4) containing E. coli membranes equivalent to 10 pmol of CYP2D6 (wild-type or mutant), quinidine (0, 1, 10, or 100 M), an NADPHgenerating system (comprising 5 mM glucose 6-phosphate, 1 unit of glucose-6-phosphate dehydrogenase, and 1 mM NADP ϩ ), and bufuralol or dextromethorphan at concentrations equivalent to the K m of each sample. The specific substrate concentrations used were as follows: 1 M bufuralol and 9.5 M dextromethorphan. After a 3-min preincubation at 37°C, reactions were initiated by the addition of the NADPH-generating system and were allowed to proceed for 6 min before being stopped by the addition of 15 l of 60% perchloric acid. 100-l aliquots of the reaction supernatant were used for HPLC, separating the bufuralol and dextromethorphan metabolites as described previously (17), using a Hewlett-Packard 1100 HPLC system and ChemStation software. As reported previously (17,21), the only significant decrease in activity for the mutants relative to the wild-type enzyme in the absence of inhibitor was for dextromethorphan O-demethylation by E216Q (ϳ4-fold) and E216K (ϳ10-fold).
Quinidine Metabolism-To investigate quinidine metabolism, reaction mixtures consisted of 50 mM potassium phosphate (pH 7.4) containing 100 M quinidine, E. coli membranes equivalent to 10 pmol of CYP2D6 (wild-type or mutant) and an NADPH-generating system (as described above) in a total volume of 200 l. After a 3-min preincubation at 37°C, reactions were initiated by the addition of the NADPHgenerating system and incubated for an additional 15 min before being stopped with 100 l of ice-cold methanol. Samples were left on ice for 10 min prior to centrifugation at 16,100 ϫ g for 10 min. Metabolites were separated by HPLC using a Hypersil BDS-C18 column (5 m, 250 ϫ 4.6 mm) at a flow rate of 1 ml/min. Mobile phases of acetonitrile (solvent A) and sodium perchlorate/perchloric acid (14.05 g of sodium perchlorate and 1.6 ml of 60% perchloric acid dissolved in 5 liters of distilled H 2 O; solvent B) were mixed at a constant ratio of 15% solvent A to 85% solvent B (v/v) for the first 5 min, and then a linear gradient was applied over 4 min, ending at 31% solvent A to 69% solvent B (v/v), which was maintained for an additional 7 min. The retention times of (3S)-3Јhydroxyquinidine and quinidine N-oxide were established using authentic metabolite standards, with fluorescence detection at ex ϭ 252 nm and em ϭ 302 nm.
Identification of the Novel Quinidine Metabolite-Further analysis of the novel quinidine metabolite was undertaken by HPLC with mass spectrometric detection. 25 l of the stopped incubation mixture was separated on a Luna C18 column (3 m, 150 ϫ 2 mm; Phenomenex, Cheshire, UK) with a linear gradient of 5 mM ammonium formate (pH 3.5) (solvent A) and acetonitrile (solvent B) delivered by a Waters 2795 separations module. The gradient ran from 5 to 30% solvent A over 20 min at a flow rate of 200 l/min before returning to the starting conditions. The eluent was introduced into the source of a Quattro micro mass spectrometer (Micromass, Manchester, UK) and was ionized by electrospray ionization in the positive ion mode. The main parameters were as follows: capillary voltage, 3.3 kV; cone voltage, 30 V; source and desolvation temperatures, 100 and 300°C, respectively; and cone and desolvation nitrogen gas flows, 90 and 300 liters/h Ϫ1 , respectively. In collision-induced dissociation experiments, argon was used as the collision gas with a collision energy of 30 eV. Data were acquired and analyzed by MassLynx software.
Quinidine Binding-Quinidine binding was measured by optical difference spectroscopy of E. coli membranes containing CYP2D6 and NADPH-cytochrome P450 reductase (EC 1.6.2.4) using a Cary 4000 UV-visible spectrophotometer. E. coli membranes containing wild-type or mutant CYP2D6 enzymes were diluted in 100 mM potassium phosphate buffer (pH 7.4) to a final concentration of 0.5 M cytochrome P450 and split into two matched black-walled quartz cuvettes. After running a base line, 1-l aliquots of quinidine dissolved in deionized water were added to the sample cuvette, and equal volumes of water were added to the reference cuvette. The samples were left for 2 min between additions to equilibrate, and the difference spectrum was then run between 360 and 460 nm. The final volume of additions was kept to Ͻ2.5% of the total volume. Changes in absorbance as a function of quinidine concentration, at wavelengths selected on the basis of the spectral characteristics of the individual sample, were used to calculate binding constants using nonlinear regression analysis (Prism). Spectral determinations were performed at least twice for each mutant and found to be reproducible with respect to the spectral profile and the position of max and min .
Modeling and Molecular Docking-Homology modeling and molecular docking were used to predict the modes of quinidine binding to wild-type and mutant enzymes. The homology model of CYP2D6 was produced as described previously (14). In brief, the model was produced using the comparative modeling program Modeler (22) with five structural templates: cytochromes P450cam (23), P450terp (24), P450eryF (25), P450BM3 (26), and P4502C5 (27). Model structures for mutants E216F, E216Q/D301Q, and F120A were generated by replacing residues within SYBYL (28) and optimizing the positions of the new side chains with the rest of the protein held fixed. (Minimizations of atoms in a 10-Å sphere around the mutated residue produced no noticeable changes; these results were therefore discarded.) Docking studies were performed using the program GOLD (version 2.2) (29) with the ChemScore fitness function (30,31) to generate 50 possible binding orientations for quinidine into both the deoxygenated (i.e. type I-like) and oxygenated (i.e. compound I-like) forms of each of the wild-type and mutant CYP2D6 models. The orientations were ranked according to the value of the ChemScore fitness function. In addition "tethered" dockings were performed by applying constraints to the distances between the O-methyl group and 3-carbon atom of quinidine and the heme iron and between the basic nitrogen atom of quinidine and the carboxyl group of Glu 216 (see TABLE TWO for details of the tethered dockings used). The docked energy of a solution that positioned the tethered groups farther than 4.5 Å apart was penalized, the size of the penalty being determined using a harmonic force constant of 5.0 kJ⅐mol Ϫ1 ⅐Å Ϫ2 .

Inhibition of CYP2D6 Mutants by Quinidine-
We investigated the effects of mutations of the active-site residues Phe 120 , Glu 216 , Asp 301 , Phe 481 , and Phe 483 on the inhibition of CYP2D6 activity by quinidine. The inhibition profile of a panel of 12 mutants was examined by measuring bufuralol 1Ј-hydroxylation and dextromethorphan O-demethylation in the presence of 1, 10, or 100 M quinidine (Fig. 1). Many of these mutants affect substrate binding; so, to isolate the effects of the mutations on quinidine inhibition, the substrate concentrations used were chosen to be equal to the measured K m for each specific mutant (17,21). (The concentrations used are given under "Experimental Procedures.") For the wild-type enzyme under the conditions of this assay, the lowest concentration of quinidine used (1 M) led to Ͼ95% inhibition of both bufuralol 1Ј-hydroxylation and dextromethorphan O-demethylation.
It is clear that the negative charges at Glu 216 and Asp 301 are important for the inhibitory effect of quinidine. The conservative substitutions E216D and D301E showed behavior similar to that of the wild-type enzyme, with Ͼ90% inhibition by 1 M quinidine, whereas enzymes with non-conservative replacements were at least 50% active at 10 M quinidine. The double mutant E216Q/D301Q, with complete removal of the charge but not the polarity, was found to be strikingly insensitive to inhibition by quinidine, retaining 80% of its bufuralol 1Ј-hydroxylase activity and 85% of its dextromethorphan O-demethylase activity in the presence of 100 M quinidine. By contrast, alanine substitution of the aromatic side chain of Phe 120 , Phe 481 , or Phe 483 had only a minor effect on the inhibition of catalytic activity by quinidine. The effects of the mutations were generally similar for quinidine inhibition of both bufuralol 1Ј-hydroxylase and dextromethorphan O-demethylase activities, although for most of the mutants, quinidine was found to be a somewhat better inhibitor with respect to dextromethorphan compared with bufuralol. These observations suggest that the negative charges at Glu 216 and Asp 301 , but not the aromatic rings of the three phenylalanine residues, are important for the binding of quinidine; this is broadly consistent with the effects of mutation of these residues on the K m values of substrates containing a basic nitrogen atom (10, 16 -18, 21, 32).
Quinidine Binding to the Glu 216 and Asp 301 Mutants-The effects of the mutations on quinidine binding were determined directly by measuring optical difference spectra upon addition of quinidine to bacterial membranes expressing cytochrome P450. Wild-type CYP2D6 showed a type I binding spectrum upon quinidine addition, with max and min of ϳ420 and ϳ390 nm, respectively ( Fig. 2A), characteristic of the change from a low to high spin state of the ferric iron that is usually associated with the binding of substrate molecules (11). None of the mutants showed evidence of a type II spectrum, characteristic of direct coordination to the heme iron. The majority showed type I difference spectra ( Fig. 2A) or variations thereof (Fig. 2C), but three showed a different form of spectrum with an increase in absorbance at shorter wavelengths (Fig. 2B), suggesting either a change in the heme environment or light scattering from membrane or protein aggregation. 5 The K d(app) values for quinidine binding, derived from the dependence of the amplitude of the difference spectrum on quinidine concentration, are shown in TABLE ONE. The alanine substitutions of Phe 120 , Phe 481 , and Phe 483 led to no more than a factor of 2 decrease in binding affinity. However, removal of the negative charge from either Glu 216 or Asp 301 produced clear increases in the K d . Wild-type CYP2D6 and the  NOVEMBER 18, 2005 • VOLUME 280 • NUMBER 46 two conservative mutants E216D and D301E had K d values in the 0.4 -0.5 M range, whereas the K d values for the non-conservative substitutions were at least 30-fold higher, in the 15-65 M range. Consistent with the inhibitory effects on bufuralol and dextromethorphan metabolism (Fig. 1), the largest increase in the K d(app) was observed with the double mutant E216Q/D301Q. These results are consistent with similar decreases in affinity indicated by K m estimates for bufuralol 1Ј-hydroxylation (17).

Role of Glu 216 /Asp 301 /Phe 120 in 2D6-Quinidine Interactions
Quinidine Metabolism-Both the inhibition assays and the direct measurements of binding showed that removing the negative charges from Glu 216 and/or Asp 301 produced a major quantitative effect on quinidine binding to CYP2D6. Because quinidine produces a type I binding difference spectrum, typical of CYP2D6 substrates, but is not metabolized by wild-type CYP2D6 (5, 7-9), we carried out experiments to examine the possibility that some of the mutants might be able to metabolize quinidine. Quinidine was incubated for 15 min with 50 nM wild-type or mutant CYP2D6, and the incubation mixture was analyzed by HPLC. As shown by the chromatograms in Fig. 3, quinidine was clearly metabolized by the three mutants E216F, E216Q/D301Q, and F120A. Two clear metabolite peaks with retention times of 8.2 min (M1) and 8.8 min (M2), respectively, were observed; both were produced by E216Q/D301Q and F120A, whereas E216F produced only the metabolite eluting at 8.8 min. The peak with a retention time of 8.2 min comigrated with a standard of (3S)-3-hydroxyquinidine, demonstrating that, in contrast to wildtype CYP2D6, the E216Q/D301Q and F120A mutants, alone among the mutants studied, were able to catalyze the 3-hydroxylation of quinidine. The second metabolite, M2 (Fig. 3), formed by E216F, E216Q/D301Q, and F120A, did not comigrate with either of the quinidine metabolite standards available to us ((3S)-3-hydroxyquinidine and quinidine N-oxide). Quinidine 3-hydroxylation turnover rates for E216Q/D301Q and F120A were estimated from these experiments as 0.14 and 0.07 min Ϫ1 , respectively, somewhat slower than the typical rates of 1-5 min Ϫ1 obtained for the wild-type enzyme for standard substrates such as bufuralol and dextromethorphan (17). Rates for O-demethylquinidine (metabolite M2) could not be measured due to lack of a metabolite standard.
Analysis by mass spectrometry showed the molecular ion of metabolite M2 at m/z 310.7, a decrease of m/z 14 compared with quinidine,

Role of Glu 216 /Asp 301 /Phe 120 in 2D6-Quinidine Interactions
suggesting the occurrence of a demethylation reaction (Fig. 4). Metabolite M2 was not present in the control reaction (Fig. 4A, upper chromatogram). Collision-induced dissociation generated the spectrum shown in Fig. 4B. The daughter ion at m/z 174.8 was assigned to the O-demethylated fragment 4-(hydroxymethyl)quinolin-6-ol, allowing us to identify metabolite M2 as O-demethylquinidine. The observation that quinidine was metabolized by three of the mutants studied here, but not by the wild-type enzyme, clearly implies that the mode of quinidine binding, and not just its affinity, is affected by these mutations. To help us understand the structural basis of this, we carried out computational docking of quinidine into structural models of the active site of the wild-type and mutant enzymes.
Modeling of Quinidine Binding to CYP2D6-To help rationalize the experimental data, computational docking studies were performed using our previously described model of wild-type CYP2D6 (14). 50 solutions for quinidine binding to oxygenated (compound I-like) and deoxygenated (type I-like) CYP2D6 were obtained, leading to solutions with good ChemScore values (Ϫ39.0 kJ⅐mol Ϫ1 ) (TABLE TWO), in keeping with the experimentally observed affinity (K d ϭ 0.4 M). In all solutions, the quinidine was positioned away from the heme; the orientation of quinidine within the active site in the best ranked solution is shown in Fig. 5A. The results from these docking studies on the wildtype enzyme are consistent with the experimental data insofar as they produced no solutions for quinidine binding close to the heme in a position appropriate for catalytic turnover.
The predicted binding mode of quinidine in wild-type CYP2D6 appears to be influenced by interactions between the aromatic rings of quinidine and Phe 120 and Phe 483 and by a hydrogen bond between the hydroxyl group of quinidine and the carboxyl group of Glu 216 . To investigate the impact of side chain flexibility on quinidine binding, docking studies were also performed on versions of the CYP2D6 model in which alternative conformations had been generated for Phe 120 , Leu 213 , Glu 216 , and Asp 301 , based on the probability of a given rotamer occurring in known crystal structures, and in which no severe steric clashes were observed between the new side chain position and the rest of the protein. In all the docking solutions obtained, the quinidine was located in a position incompatible with metabolism. Given that Glu 216 is thought to play a key role in the recognition of substrates containing a basic nitrogen atom, tethered dockings were performed in which the basic nitrogen atom of quinidine was constrained to lie near the carboxyl group of Glu 216 . The resulting solutions exhibited substantially poorer ChemScore values (30.0 kJ⅐mol Ϫ1 ) compared with the unconstrained dockings (39.0 kJ⅐mol Ϫ1 ) (TABLE TWO). These results suggest that such a binding mode is energetically unfavorable and that, even if such a binding mode were adopted, it would be incompatible with metabolism. To enable comparison with metabolites produced by the mutants studied (see below), tethered dockings were also performed to produce orientations consistent with formation of 3-hydroxyquinidine (metabolite M1) and O-demethylquinidine (metabolite M2). The resulting solutions had significantly poorer ChemScore values (31.7 and 36.3 kJ⅐mol Ϫ1 , respectively) compared with the unconstrained dockings (39.0 kJ⅐mol Ϫ1 ). These results suggest that quinidine does not adopt a binding mode compatible with metabolism, consistent with its role as a competitive inhibitor of CYP2D6. In these tethered dockings, unfavorable contacts occurred with Phe 120 and Ala 305 in the orientation consistent with formation of 3-hydroxyquinidine and with Phe 120 , Leu 121 , and Glu 216 in the orientation consistent with formation of O-demethylquinidine. This suggests that these residues are likely important in preventing metabolism of quinidine in wild-type CYP2D6.
We also carried out docking studies with a model of the F120A mutant. The 50 solutions obtained could be divided into two clusters; the best ranked solutions from each cluster are shown in Fig. 5 (B and C). The solution with the best ChemScore value (37.1 kJ⅐mol Ϫ1 ) (TABLE    observation that this mutant was able to metabolize quinidine to its O-demethyl and 3-hydroxy derivatives. In our model of the E216F mutant, the most energetically favorable position of this substituted side chain forms a face-on-face interaction with Phe 219 and an edge-on-face interaction with Phe 483 . The 50 solutions obtained could be divided into two clusters. The solution with the best ChemScore value (37.8 kJ⅐mol Ϫ1 ) (TABLE TWO) is consistent with 1Ј-hydroxylation, i.e. formation of neither metabolite M1 nor M2. The solution in the second cluster with the best ChemScore value (37.2 kJ⅐mol Ϫ1 , i.e. only 0.6 kJ⅐mol Ϫ1 poorer than the best overall ChemScore value) is consistent with O-demethylation (Fig. 5D). This orientation suggests an interaction between the aromatic rings of quinidine and Phe 216 and Phe 120 and a hydrogen bond between the basic nitrogen atom of quinidine and the side chain of Ser 304 . To investigate the possibility of forming 3-hydroxyquinidine, tethered dockings were performed. This distance constraint is difficult to satisfy, producing unfavorable contacts with Phe 120 , Val 308 , and Phe 483 . These results are consistent with the experimental observation that the E216F mutant was able to metabolize quinidine to its O-demethyl (but not 3-hydroxy) derivative.
By contrast, similar docking studies on the E216Q/D301Q mutant produced only solutions in which the quinidine molecule was positioned away from the heme, whereas experimentally, this mutant was able to metabolize quinidine. To address this apparent discrepancy, we carried out tethered docking of quinidine to investigate the possibility of solutions consistent with O-demethylation and 3-hydroxylation of quinidine using the structural model of the E216Q/D301Q mutant. The ChemScore values of these tethered solutions were less favorable than those of the solutions from the corresponding dockings without distance constraints (TABLE TWO). In the 3-carbon-tethered solutions, this was due to some unfavorable contacts, particularly with Phe 120 , Gln 216 , Thr 309 , and Phe 483 . However, Asp 301 is thought to contribute to the stabilization of the BЈ-C loop through a hydrogen bond with the backbone of Val 119 (14); and therefore, mutation of this residue to Gln could lead to a conformational change in the BЈ-C loop, thus impacting more globally on the nature of the active site.

DISCUSSION
Quinidine is a well established and potent competitive inhibitor of CYP2D6 (5,(7)(8)(9); indeed, inhibition by quinidine is often used as a diagnostic for involvement of CYP2D6 in drug metabolism. However, it has not been clear why quinidine is not a substrate because it has many of the features that are regarded as characteristic of CYP2D6 substrates, including a basic nitrogen atom and a flat hydrophobic region (12). Docking quinidine into our homology-based model of CYP2D6 (Fig. 5) led to a predicted mode of binding in which it fits into the active site, but is too far from the heme for catalytic turnover. Quinidine binding to CYP2D6 produces a classical type I optical difference spectrum (cf. Fig.  2), indicative of the change from a low to high spin state of the ferric iron that usually accompanies the binding of substrate molecules (11) and that is associated with the displacement of the water molecule bound to the iron in the "resting" enzyme, converting the iron from six-to fivecoordinate. In the proposed mode of binding of quinidine, the inhibitor is too far from the heme iron to displace the bound water molecule directly. However, it is clear (for example, from NMR and crystallographic studies of CYP102A (33,34)) that the binding of ligands relatively distant (ϳ9 Å) from the heme iron of cytochromes P450 can lead to the displacement of the bound water molecule.
Previous modeling and mutagenesis work had suggested that two carboxylate groups (at Glu 216 and Asp 301 ) and three phenylalanine res-idues (Phe 120 , Phe 481 , and Phe 483 ) play important roles in determining the binding of substrates in the active site of CYP2D6 (e.g. Refs. 10 and 14 -19). This work has shown that several of these residues also play significant roles in binding the inhibitor quinidine. In terms of the binding constants, the two carboxylate groups are clearly the most important; abolition of one or both of these charges increased the K d for quinidine by 30 -100-fold. In the best scoring docked orientation of quinidine in the active site of the wild-type enzyme, the basic quinuclidine nitrogen was closer to Glu 216 (5.4 Å) than to Asp 301 (8.8 Å), although the effects of substituting either residue with the corresponding amide were very similar (TABLE ONE). The fact that the double mutant E216Q/D301Q showed significantly weaker binding than either single mutant supports the idea that the electrostatic field of both residues is significant for quinidine binding. A recent study of a number of analogs of quinidine demonstrated that alkylation of the quinuclidine nitrogen with groups as bulky as naphthyl had no effect on the measured IC 50 , and it was concluded that the proposed charge-charge interaction with Asp 301 does not make a major contribution to binding (13). The ϳ30-fold increase in the K d for E216F, corresponding to a decrease in binding energy of ϳ9 kJ⅐mol Ϫ1 , is consistent with the loss of a hydrogen bond and/or a weak charge-charge interaction. This is consistent with our prediction that the hydroxyl group (rather than the basic nitrogen atom) of quinidine forms the dominant interaction with Glu 216 . More generally, it must be recognized that, particularly for the cytochromes P450, interpretations of the effects of altering either the protein or the ligand are complicated by the possibility (indeed, the likelihood) that these changes will lead to an altered mode of binding.
For several of the mutants studied here, it is clear that there was indeed a change in the mode of binding of quinidine because it became a substrate rather than an inhibitor. The F120A mutant and the double mutant E216Q/D301Q each formed both 3-hydroxyquinidine and O-demethylquinidine in significant quantities. 3-Hydroxyquinidine is a major product of quinidine metabolism by CYP3A4 (7) and, indeed, this reaction has been suggested as a specific marker for CYP3A4 in human liver microsomes (35). We have shown previously that this double mutant of CYP2D6 is able to catalyze another characteristic CYP3A4 reaction, the N-oxidation of nifedipine (17), and concluded that Glu 216 and Asp 301 have central roles in defining the specificity of CYP2D6. We also noted the possibility that the effects of mutation of Asp 301 are indirect because, in our model (14), the side chain of this residue interacts with the backbone of the BЈ-C loop, thus helping to position this loop, including Phe 120 , in the active site and, indeed, the F120A mutant metabolizes quinidine to the same two products as the E216Q/D301Q mutant. However, substitution of Asp 301 alone is not sufficient to enable CYP2D6 to metabolize quinidine, and Glu 216 clearly plays an important role in determining the mode of binding. This is emphasized by the fact that substitution of this residue with a bulky side chain in the E216F mutant confers on CYP2D6 the ability to catalyze the O-demethylation of quinidine and also, to some extent, the 6␤-hydroxylation of testosterone, another characteristic CYP3A4 reaction (17).
To obtain a structural picture of the possible changes in the binding mode of quinidine in the mutants, we docked quinidine into models of the mutants, obtained by side chain substitution in our earlier model of the wild-type enzyme (14). The model of the F120A mutant gave results consistent with the experimental observations in that quinidine was predicted to bind closer to the heme than in the wild-type model, with the most favorable modes of binding predicted to be those corresponding to 3-hydroxylation and O-demethylation, the observed routes of metabolism (Fig. 5, B and C). Within the limitations of the model, this indicates a direct role of the Phe 120 side chain in determining the unpro-ductive mode of binding of quinidine to CYP2D6. The model of the E216F mutant also gave results consistent with the experimental observations.
On the other hand, with the E216Q/D301Q mutant, simple docking calculations did not predict a productive mode of binding for quinidine. However, when appropriate distance constraints were introduced into the docking calculations, solutions were obtained that were consistent with the formation of the experimentally observed metabolites. The calculated ChemScore values of these solutions were somewhat poorer than those of the unconstrained solutions due largely to unfavorable interactions with the side chains of Phe 120 and Phe 483 . It is possible that, in these mutants, quinidine binds most of the time in a nonproductive mode, but that productive mode(s) of binding are accessible and lead to the observed turnover. However, the limitations of the models should be recognized. Phe 120 and Phe 483 , which appear to interact unfavorably with quinidine in the models of these two mutants, are both located in loops, in SRS1 and SRS6, respectively. Loop regions are often highly flexible and therefore difficult to represent with either a single model or a single crystal structure. The docking program we have used (GOLD Version 2.2) (29, 36) allows the ligand full translation and rotational freedom, but, like most available docking programs, it treats the protein as rigid. Thus, it may be that the flexibility in these loops is such that the two phenylalanine rings can move away from the bound quinidine, making the productive mode of binding the most favorable. Particularly in cytochromes P450, the problem of receptor flexibility is a significant challenge in predicting ligand binding.
The results of these mutagenesis experiments clearly show that Phe 120 , Glu 216 , and Asp 301 are important in determining the mode of binding of quinidine to CYP2D6, particularly in determining whether it binds in a nonproductive mode, as in the wild-type enzyme, and is thus an inhibitor, or whether it binds productively. They also demonstrate that changes in affinity and binding mode do not necessarily go handin-hand: in the F120A mutant, a change in the mode of binding (indicated by the ability to metabolize quinidine) was not accompanied by any change in K d , whereas in the E216K mutant, a 70-fold increase in K d was not accompanied by a change in the mode of binding sufficient to allow quinidine to be metabolized. A particularly important role is indicated for Phe 120 because substitution of this single residue allowed CYP2D6 to metabolize its "classical" inhibitor quinidine with no decrease in binding affinity, and the docking calculations strongly suggest that this residue has a direct effect in forcing quinidine to bind in an unproductive mode to CYP2D6.