The Structure of Human Microsomal Cytochrome P 450 3 A 4 Determined by X-ray Crystallography to 2 . 05-Å Resolution *

The structure of P450 3A4 was determined by x-ray crystallography to 2.05-Å resolution. P450 3A4 catalyzes the metabolic clearance of a large number of clinically used drugs, and a number of adverse drug-drug interactions reflect the inhibition or induction of the enzyme. P450 3A4 exhibits a relatively large substrate-binding cavity that is consistent with its capacity to oxidize bulky substrates such as cyclosporin, statins, taxanes, and macrolide antibiotics. Family 3A P450s also exhibit unusual kinetic characteristics that suggest simultaneous occupancy by smaller substrates. Although the active site volume is similar to that of P450 2C8 (PDB code: 1PQ2), the shape of the active site cavity differs considerably due to differences in the folding and packing of portions of the protein that form the cavity. Compared with P450 2C8, the active site cavity of 3A4 is much larger near the heme iron. The lower constraints on the motions of small substrates near the site of oxygen activation may diminish the efficiency of substrate oxidation, which may, in turn, be improved by space restrictions imposed by the presence of a second substrate molecule. The structure of P450 3A4 should facilitate a better understanding of the substrate selectivity of the enzyme.

The structure of P450 3A4 was determined by x-ray crystallography to 2.05-Å resolution. P450 3A4 catalyzes the metabolic clearance of a large number of clinically used drugs, and a number of adverse drug-drug interactions reflect the inhibition or induction of the enzyme. P450 3A4 exhibits a relatively large substrate-binding cavity that is consistent with its capacity to oxidize bulky substrates such as cyclosporin, statins, taxanes, and macrolide antibiotics. Family 3A P450s also exhibit unusual kinetic characteristics that suggest simultaneous occupancy by smaller substrates. Although the active site volume is similar to that of P450 2C8 (PDB code: 1PQ2), the shape of the active site cavity differs considerably due to differences in the folding and packing of portions of the protein that form the cavity. Compared with P450 2C8, the active site cavity of 3A4 is much larger near the heme iron. The lower constraints on the motions of small substrates near the site of oxygen activation may diminish the efficiency of substrate oxidation, which may, in turn, be improved by space restrictions imposed by the presence of a second substrate molecule. The structure of P450 3A4 should facilitate a better understanding of the substrate selectivity of the enzyme.
Determination of the structure of P450 1 3A4 is of particular interest because the enzyme contributes extensively to human drug metabolism due to its high level of expression in liver (1) and broad capacity to oxidize structurally diverse substrates (2,3). The enzyme also provides a significant barrier to the bioavailability of new drug candidates contributing to attrition from the developmental pipeline. Additionally, metabolic drug-drug interactions between substrates and inhibitors of the enzyme can profoundly affect the safety or efficacy of drug therapy (4,5).
Our laboratory was the first to demonstrate that microsomal P450s could be crystallized for structural determination by x-ray crystallography when the proteins were modified for expression as conditional membrane proteins (6,7). As a result, structures for P450s in family 2, subfamilies B and C are now available (8 -14). P450s of family 3, subfamily A exhibit less than 40% amino acid sequence identity with family 2 P450s. In addition, family 3 P450s often exhibit complex kinetic properties such as substrate and effector activation. Effectors or alternative substrates can modulate the apparent binding affinity for other inhibitors (15) and substrates (16). Moreover, there are a number of examples where alternative substrates fail to inhibit the oxidation of specific substrates leading to kinetic models based on the occupancy of the substrate-binding cavity by two substrates that each can be oxidized by the reactive, hypervalent oxy-perferryl heme intermediate without interference from the other (17,18). The observation that P450 3A4 oxidizes some of the largest substrates identified for P450s, such as cyclosporin, bromocryptine, and macrolide antibiotics (3), has generally suggested the likelihood that the active site cavity of the enzyme is relatively large compared with other P450s. A large active site cavity would also be consistent with models where two or more molecules of smaller substrates are postulated to simultaneously occupy the active site cavity potentially altering the dissociation constant for substrate binding and/or catalytic efficiency by constraining the substrate close to the reactive oxygen intermediate during catalysis. The structure of human microsomal P450 3A4 described in this report exhibits a relatively large substratebinding cavity that is consistent with these notions regarding Charitable Trust. the enzyme. Of the structures determined previously for family 2 P450s, 3A4 is most similar to that of P450 2C8 (PDB code: 1PQ2) that also oxidizes relatively large substrates and exhibits a large active site cavity. However, the two enzymes differ considerably in the architectures of their active sites in ways that are likely to correspond to the unusual kinetic properties of P450 3A4.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The pSE3A4His expression plasmid (19) was obtained from James Halpert (University of Texas Medical Branch, Galveston, TX) for the expression of P450 3A4 with a C-terminal His tag in Escherichia coli. The plasmid was modified to express P450 3A4 without the trans-membrane leader sequence of the microsomal protein, amino acid residues 3-23, to improve its solubility and facilitate crystallization. No other mutations were employed. The resulting construct, pSE3A4dH, was transformed into E. coli strain, DH5␣. The expression and purification of P450 3A4dH closely follows the protocols described previously for P450 2C5dH that employed metal ion affinity chromatography followed by CM-Sepharose chromatography (20). In contrast to P450 2C5dH, which is largely dissociated from membranes by high salt buffers, about 80% of 3A4dH is found in the membrane fraction of high salt lysates. For this reason, P450 3A4dH was solubilized using 0.4% (w/v) of CHAPS (Calbiochem). CYMAL6 (Anatrace, Maumee, OH) was exchanged for CHAPS during subsequent chromatography procedures, and the buffers used during purification procedures were supplemented with 0.4 mM erythromycin (Sigma). The purified protein was concentrated for crystallization to 0.3 mM in 50 mM potassium phosphate buffer, pH 7.4, containing 1 mM EDTA, 0.2 mM dithiothreitol, 0.5 M NaCl, 0.4 mM erythromycin, and 20% glycerol using a centrifugal membrane-concentrating device with a 50-kDa cutoff (Millipore, Billerica, MA). The protein exhibited a molecular weight of 55,778 when analyzed by matrix-assisted laser desorption/ionization time-of-flight spectrometry at the Scripps Center for Mass Spectroscopy, which is close to the predicted value of 55,526.
Spectral Binding Studies-Substrate binding to the purified protein was assayed by monitoring the conversion of the heme prosthetic group from a low spin to a high spin state spectrophotometrically as described (21). Purified protein was diluted to 0.8 -1.0 M in 50 mM potassium phosphate buffer, pH 7.4, containing 500 mM NaCl, 20% glycerol, 1 mM EDTA, and 0.2 mM dithiothreitol in a 1-cm path length microcuvette. Aliquots of either 7,8-benzoflavone or testosterone dissolved in methanol were added to the solution, and the differences in the absorption between 394 and 418 nm were recorded. The apparent binding constant was estimated as described previously (22). The results confirmed that the purified protein retained its capacity to bind testosterone and 7,8benzoflavone with the same dissociation constant as P450 3A4 expressed from the pSE3A4HIS plasmid in E. coli, 16 M.
Crystallization and Structure Determination-The protein was crystallized by vapor diffusion in sitting drops against a well solution of 6% PEG 5000 MME, 0.1 M HEPES, pH 7.0, and 0.4 mM erythromycin. The drop consisted of 0.4 l of 3A4dH protein solution, 0.1 l of 40 mM CHAPS, and 1.25 l of well solution and was covered with a layer of paraffin oil after combining the components. Single crystals were prepared for data collection by sequential 60-s soaks in 5.6% PEG 5000 MME, 0.04 M HEPES, pH 7.0, containing 20 and 30% ethylene glycol, respectively. The crystal was rapidly frozen in liquid nitrogen before placing it in the cryo-stream for data collection. The initial structure of P450 3A4dH was solved by the MAD of the native iron of the heme prosthetic group of the enzyme using data collected at the Stanford Synchrotron Radiation Laboratory Beamline 9-2. CNS (23) was used for MAD phasing, solvent flattening, and density modification to produce the initial electron density maps. The initial maps from the MAD phasing provided a trace for almost all of the main chain as well as significant information regarding the larger side chains. Subsequent phase combination and extension allowed the initial model to be built and refined using CNS (23). The final model was built and refined using a high resolution data set obtained for a crystal diffracting to 2.0 Å at the Advanced Light Source Beamline 5.0.2. The model encompasses residues His-28 to Thr-499, where the numbering corresponds to that of the wild-type enzyme, and exhibits R and R free values of 0.24 and 0.29, respectively. An external loop between helices H and I, residues 282-285, could not be modeled. This is likely to reflect the flexibility of this region, which is not constrained by crystal contacts. Similarly, 6 residues at the N terminus and 8 residues at the C terminus of the truncated and His-tagged protein used for crystallization were not visible. A statistical analysis of the data, the final refinement statistics, and stereochemical analysis of the model are given in supplemental Table I.
Automated Docking of Erythromycin in the Structure-Computerbased automated docking of erythromycin A into the active site of P450 3A4 was performed using a grid based docking program, AUTODOCK 3.05 (24), with a modified genetic search algorithm employing a local minimum refinement. A 58 ϫ 72 ϫ 70-point grid with a spacing of 0.375 Å centered at Ϫ21.6, Ϫ22.7, Ϫ11.9 Å was used for a protein model. Coordinates for the initial model of erythromycin A were obtained from the Cambridge Crystallographic Data Base.

RESULTS AND DISCUSSION
The most significant differences between the overall structure of P450 3A4 and other P450s are seen for the helix D to H and the helix B to C regions (Fig. 1A). The structure is most similar to the structures of mammalian family 2 P450s, which exhibit less than 40% sequence identity with P450 3A4. Eukaryotic P450s, including 3A4, exhibit longer sequences between helices F and G that generally exhibit two additional helices, FЈ and GЈ. This region is thought to form a membrane interaction domain (7), and as seen in family 2 P450s, the outer surfaces of helices FЈ and GЈ are hydrophobic. 2C8 was chosen for comparison in Fig. 1B because it shares a capacity with 3A4 to oxidize relatively large substrates such as taxanes and statins. The solvent-accessible surfaces of the active site cavities are rendered as a mesh in Fig. 1, A-C. The cavities of 2C8 and 3A4 exhibit similar volumes of 1438 Å 3 (8) and 1386 Å 3 , respectively, when truncated at the narrowest constrictions of the solvent channels. The shapes of the cavities differ in ways that are likely to affect substrate selectivity and enzyme catalysis, and this, in turn, reflects differences in the secondary and tertiary structure of the proteins. The structure of 3A4 is much more open in the vicinity of the heme iron, and the cavity volume is more uniformly distributed than the sinuous cavity of 2C8. The larger volume of the active site in the vicinity of the heme iron when compared with 2C8 reflects changes in the conformation of the protein in the SRS5 region as it passes from helix K to strand ␤1-4 as well as structural differences in the SRS2 and SRS3 regions that pass above the heme (Fig. 1,  C and D). A large cavity extends outward along the surface of sheet ␤-1 under helix FЈ where a relatively large solvent channel to the surface exits between sheet ␤-1, the end of helix FЈ and SRS1 (Fig. 1, A and C). Arg-106 extends across one edge of the channel and participates in a hydrogen-bonding network formed by Arg-106, Glu-374, Asp-76, Arg-372, Asp-61 and Tyr-53 that is stabilized by charge interactions (Fig. 1E). A smaller solvent channel exits the active site cavity on the other side of residues 106 -108.
In contrast to the structures of family 2 P450s, helices F and G do not pass over the active site cavity in 3A4 (Fig. 1, A and B) because these helices are shorter. Residues 209 -217 and 237-242 that connect the F and G helices to helices FЈ and GЈ form the roof of the active site cavity, are part of SRS2 and SRS3, respectively, and do not exhibit a regular secondary structure (Fig. 1D). A third solvent channel exits the active site cavity under the SRS2 region between SRS4 and SRS6. Several phenylalanine residues that include Phe-108, Phe-213, Phe-215, Phe-241, and Phe-304 contribute to the packing that closes the roof of the active site above the heme between the SRS1, SRS2, SRS3, and SRS4 regions. Interestingly, the reported mutations L211F and D214E located on the outer edge of the SRS2 region above the third solvent channel alter the homotropic cooperativity exhibited by 3A4 catalyzed steroid hydroxylations and confer a hyperbolic dependence on substrate concentration (16). This effect was thought to arise from a reduction of the active site volume due to the increased volume occupied by the side chains of the mutated residues. These residues are not directly in the active site of the structure determined here, and it is difficult to predict how these mutations would affect protein conformation. Mutations to additional residues in this region, such as Leu-210, Phe-213, and Phe-215, that form the roof of the active site have also been shown to alter the cooperativity exhibited by the enzyme (19,25). We suspect that the atypical structure of the region forming the roof of the active site cavity could accommodate the simultaneous binding of multiple substrate molecules because of its flexibility.
Although erythromycin was present during crystallization of the protein, 2͉F o ͉ Ϫ ͉F c ͉ (where F o indicates the observed structure factor, and F c indicates the calculated structure factor) electron density maps did not confirm its presence in the crystallized protein. Erythromycin occupancy of the crystallized protein may have been reduced by its limited solubility in the Structure of Human Microsomal Cytochrome P450 3A4 38093 crystallization buffer, nonspecific binding to micelles of the detergent CHAPS, extraction into the solutions used to protect the crystal during freezing, or extraction of erythromycin into the paraffin oil that was used to reduce the rate of vapor phase equilibration during crystallization. Automated docking studies using the model indicated that the active site cavity was sufficiently large to accommodate the substrate in an orientation that positioned the dimethylamino group of the amino sugar appropriately for oxidation (Fig. 1E) without altering the positions of any amino acids. The figure also displays several amino acid side chains that line the active site cavity and that contact the docked erythromycin molecule. The residues shown in Fig. 1E include residues 119, 301, 304, 305, 369, 370, and 374 that have been implicated as active site residues by site directed mutagenesis studies (26 -30).
Another unusual feature of P450 3A4 is the presence of Arg-212 in the active site cavity. The side chain is positioned to donate a hydrogen bond to the backbone carbonyl of Phe-304, a residue that has also been implicated in the cooperativity of the enzyme (25), as well as to a hydrogen-bonding network that includes a cluster of waters above the heme (Fig. 1F). In this position, the backbone of residues 211 and 212 would potentially disrupt the putative proton transfer pathway between the conserved residues, Glu-308 and Thr-309, on helix I. Glu-308 is hydrogen-bonded to the backbone amide of Arg-212. Repositioning of the peptide bond between the conserved acidic residue, Glu-308, and the conserved threonine, Thr-309, is thought to position a water molecule for proton transfer to the reduced oxygen intermediates bound to the heme iron leading to the cleavage of the dioxygen bond to form a water molecule and the highly reactive perferryloxo heme intermediate that oxidizes the substrate (31,32). A reduction in proton transfer can lead to increased lifetimes for the peroxy and hydroperoxy anionic intermediates that can also oxidize substrates (33). This could contribute to the P450 3A4-catalyzed conversion of the cyano moiety of pinacidil to an amide group, which had been attributed to nucleophilic oxidation by the hydroperoxy anion intermediate (34). The structure of the enzyme suggests that the side chain of Arg-212 could easily reorient to reside outside the active site cavity, and characterization of the R212A mutant indicated that the substitution had little affect on testosterone hydroxylation or its stimulation by 7,8-benzoflavone (19). Substrate binding could influence the orientation of the Arg-212 side chain. In the docked model for erythromycin binding, Arg-212 hydrogen bonds to heteroatoms of the sugar moieties of the substrate (Fig. 1E). Docking studies also indicate that there is sufficient space in the active site cavity for simultaneous occupancy by 7,8-benzoflavone and testosterone in several possible combinations of locations (data not shown).
Basic residues are conserved at alignment position 212 of family 3 P450s, and it is interesting to speculate that these basic residues could influence peroxide-supported substrate oxidation by providing general base catalysis of peroxide cleavage as occurs for a bacterial P450 fatty acid alpha hydroxylase (35). In this regard, it is interesting to note that family 3A enzymes exhibit relatively high ratios of heterolysis to homolysis of peroxyquinols (36) and that the basic residue 212 could contribute to heterolysis by general base-assisted catalysis.
In summary, the structure of 3A4 exhibits a relatively large substrate-binding cavity that is consistent with the sizes of substrates that are oxidized by the enzyme. In addition, the relatively large size of the active site cavity near the catalytic center of the enzyme may contribute to heterotropic cooperativity by facilitating alternative binding modes for multiple substrate molecules. The large size of the cavity reflects signif-icant differences in the conformations of SRS1, SRS2, and SRS5 regions in 3A4 relative to family 2 P450s. Mutations in the SRS2 region, which is not helical in 3A4, generally alter homo-and heterotropic kinetics linking the unusual conformation of the enzyme in this region to characteristic aspects of 3A4 function. The presence of a basic residue, Arg-212, in the active site was unanticipated, and its potential role in peroxidesupported oxidations catalyzed by 3A4 deserves further experimental investigation.