Crystal Structures of Epothilone D-bound, Epothilone B-bound, and Substrate-free Forms of Cytochrome P450epoK*

Epothilones are potential anticancer drugs that stabilize microtubules by binding to tubulin in a manner similar to paclitaxel. Cytochrome P450epoK (P450epoK), a heme containing monooxygenase involved in epothilone biosynthesis in the myxobacterium Sorangium cellulosum, catalyzes the epoxidation of epothilones C and D into epothilones A and B, respectively. The 2.10-, 1.93-, and 2.65-Å crystal structures reported here for the epothilone D-bound, epothilone B-bound, and substrate-free forms, respectively, are the first crystal structures of an epothilone-binding protein. Although the substrate for P450epoK is the largest of a P450 whose x-ray structure is known, the structural changes along with substrate binding or product release are very minor and the overall fold is similar to other P450s. The epothilones are positioned with the macrolide ring roughly perpendicular to the heme plane and I helix, and the thiazole moiety provides key interactions that very likely are critical in determining substrate specificity. Interestingly, there are strong parallels between the epothilone/P450epoK and paclitaxel/tubulin interactions. Based on structural similarities, a plausible epothilone tubulin-binding mode is proposed.

Since the discovery of paclitaxel (an active ingredient of Taxol®) from Taxus brevifolia (1) and its clinical success as an anti-cancer drug, there have been extensive efforts to find compounds with similar action. Those efforts resulted in the identification of three other classes of compounds from natural sources: epothilones, produced by the cellulose-degrading myxobacterium Sorangium cellulosum (2), the marine spongederived discodermolide (3,4), and the coral-derived eleutherobins (5)/sarcodictyins (6) (Fig. 1). All three classes, like paclitaxel, bind to and stabilize microtubules leading to mitotic arrest of the cycle at G 2 -M phase and subsequently induction of cell death in several cell lines (7,8). Epothilones, however, offer some advantages, because they are effective against Pglycoprotein-expressing multidrug-resistant cell lines, are active in a cell line with paclitaxel resistance (7), and the water solubility of epothilones is significantly greater than that of paclitaxel. Another advantage is that epothilones can be produced in large quantities using a heterologous expression system (9).
One step in the biosynthesis of epothilones involves a C12-C13 epoxidation (Fig. 1a) by a cytochrome P450, P450epoK (9). Cytochrome P450s (P450s) 1 are heme-containing monooxygenases best known for their role in drug detoxification (10). However, P450s also are involved in steroid hormone biosynthesis (11) as well as the biosynthesis of important macrolide antibiotics like erythromycin (12) and rapamycin (13). To date, there is no known protein structure complexed with an epothilone. In addition, epothilone represents the largest substrate of a P450 where the crystal structure of the enzyme-substrate complex is known and thus provides important insights in understanding how P450 architecture adapts to the requirements of substrate binding. Here we present the crystal structures of the oxidized P450epoK in substrate-free, epothilone D-bound, and epothilone B-bound forms at 2.65-, 2.10-, and 1.93-Å resolution, respectively.

EXPERIMENTAL PROCEDURES
Protein Purification and Crystallization-P450epoK was overexpressed in Escherichia coli and purified as will be described elsewhere. 2 Single crystals of all forms of P450epoK reported here were obtained by the vapor-diffusion method using the sitting-drop technique. P450epoK crystals were obtained within a few days after microseeding. Crystals of the imidazole-bound form for multiple wavelength anomalous diffraction (MAD) phasing were grown at 23°C in 100 mM imidazole, pH 6.5/50 mM NaCl/16% polyethylene glycol 5000 monomethyl ether. The initial droplets contained 1 l of protein solution at a concentration of 40 mg/ml and 1 l of precipitant solution and were equilibrated against a reservoir containing 500 l of precipitant solution. Crystals of the substrate-free form were also obtained by vapor-diffusion method at 23°C from 18% polyethylene glycol 5000 monomethyl ether/0.1 M Li 2 SO 4 /0.1 M MES, pH 6.5. Initial protein concentration was 20 mg/ml. Crystals of epothilone-bound P450epoK were grown at 23°C from 11% of polyethylene glycol 550 monomethyl ether/0.05 M glycine, pH 8.4, in the presence of a saturated amount of epothilone B or D. The microseeding method was also used for these crystal growths. The single crystal was carefully transferred to cryobuffer containing glycerol with the glycerol concentration brought up to 20% in four steps.
Data Collection and Processing-With the exception of MAD data, x-ray diffraction data were collected at the Advanced Light Source, Lawrence Berkeley Laboratory, beam line 8.2.2 with an Area Detector System Corporation Quantum 315 charge-coupled device detector at cryogenic temperature. MAD data sets of imidazole-bound P450epoK were collected at the Stanford Synchrotron Radiation Laboratory beam line 1-5 with an Area Detector System Corporation Quantum-4R charge-coupled device detector. Three wavelengths around the iron edge, 1.738 Å (peak), 1.741 Å (inflection), and 1.653 Å (remote), were chosen. The inverse-beam data collection procedure was used to ensure good completeness and redundancy for Bijvoet pairs. Data were processed and scaled using HKL2000 (14). Crystallographic statistics are given in Table I.
MAD Phasing-The iron position was located, and initial phases were refined with three-wavelength MAD data sets at 2.9 Å using SOLVE (15) enabling the location of several ␣-helices and bulky side chains. Solvent flattening and density modification using RESOLVE (16) resulted in a more clearly interpretable electron density map. Statistics for the MAD data sets are given in Table II. Model Building, Refinement, and Cavity Volume Calculation-A polyalanine model of P450eryF was used as a reference for the initial electron density map fitting using the graphical model building program O (17). Model refinement was carried out with CNS (18) and REFMAC (19) using the maximum likelihood protocol. Cavity volumes for substrate-free P450epoK and P450cam were calculated by the use of the program VOIDOO (20). All figures were prepared using MOL-SCRIPT (21), RASTER3D (22), or PYMOL (www.pymol.org).

RESULTS AND DISCUSSION
Comparison with Other P450s-To accommodate the bulky substrate, epothilone, P450epoK has a large substrate-binding cavity of 1060 Å 3 in volume. Despite the presence of the very large substrate binding site, the overall structure of P450epoK and its outer dimensions are similar to that of P450cam (23) whose substrate binding cavity is only 240 Å 3 in volume. Like other P450s, P450epoK exhibits the typical triangular prismshaped P450 fold (Fig. 2a) with a side 60 Å long and 30 Å thick and the heme prosthetic group embedded between the I and L helices. In addition, the substrate binding cavity is surrounded mainly by the heme and the I and F helices (Fig. 2b). Although the folds are the same, P450epoK and P450cam differ in the location of several helices (Fig. 3). The significant differences, which are important for substrate binding (24,25), include the BЈ, F, and G helices (Figs. 3 and 4a). The BЈ helix of P450cam is important for substrate binding, especially via an H-bond between Tyr 96 and the camphor carbonyl oxygen atom. In P450epoK the BЈ helix region is composed of two helices, BЈ1 and BЈ2. Both helices are farther from the active site than the BЈ helix in P450cam to accommodate the larger substrate. The BЈ2 helix is especially important, because residues from this helix help to form part of the substrate binding pocket. On the other hand, the F helix, which forms the roof over the substrate, is even closer to the heme than that of P450cam. The closer approach of the F helix to the active is possible, because the C-terminal end of the F helix has smaller side chains on the side facing the heme: Gly 176 and Ala 180 , which correspond to Thr 181 and Thr 185 of P450cam, respectively. This close approach of the F helix to the substrate or product enables the carbonyl O atom of Ala 180 in the F helix to accept an H-bond from the substrate/product C3 OH group, which will be considered in more detail below.
As in other P450s, the I helix experiences a kink near a conserved Thr residue, Thr 258 in P450epoK, that is important to the oxygen activation machinery (26,27). Another conserved structural element in P450s is the Cys thiolate heme ligand and its immediate surroundings. P450epoK is no exception with Cys 365 near the N-terminal end of the L helix coordinating the heme iron (Fig. 2b). Also similar to other P450s are the H-bond and ionic interactions between the heme propionates and His and Arg residues (Arg 107 , His 103 , and Arg 307 in P450epoK).
Of the known P450 structures, the closest homologue to P450epoK at 30% identity and 48% similarity is P450eryF (28), which also utilizes a polyketide, 6-deoxyerythronolide B (6-DEB), as a substrate. As seen for the substrate bound to P450eryF (28), epothilones D and B are oriented roughly perpendicular to the heme plane and I helix (Figs. 2b and 5). P450eryF has a long loop preceding the BЈ helix, which might provide flexibility required to accommodate the large 14member ring of 6-DEB. However, this loop is much shorter in P450epoK despite the larger size of epothilone. The F helix of P450epoK (Cys 168 -Leu 183 ), which forms the roof of the substrate binding site, exhibits a 4.7-Å r.m.s. difference in backbone compared with P450eryF (Glu166-Val176), and is positioned closer to the heme for protein-substrate interactions. The G helix (Glu194-Asn219) exhibits a 4.8-Å r.m.s. difference in backbone, and its N-terminal position is located farther from the heme than in P450eryF (Arg 181 -Glu 206 ). The relocation of these helices results in a P450epoK F/G loop that is 5 residues longer than the corresponding loop in P450eryF. The longer F/G loop could provide additional flexibility for substrate entry.
In addition to the size of the macrolide ring, another important difference between P450epoK and P450eryF is that in P450epoK the macrolide ring of the substrate is rotated about 90°relative to 6-DEB (Fig. 4b). This is necessary to accommodate the thiazole ring. The only other orientation of epothilone that would position the C12-C13 double bond for proper epoxidation would have the thiazole ring pointing toward the I helix.
Because the I helix is one of the most conserved regions in P450 and is critical for oxygen activation (26,27,29), large variations in the I helix cannot be tolerated. Therefore, the thiazole ring must point toward the BЈ2 helix thus requiring a reorientation of the macrolide ring relative to P450eryF. In P450epoK this region is farther from the active site and provides side chains such as Phe 96 that specifically interact with the thiazole ring.
Heme Coordination-Normally substrate-free oxidized P450s are low spin with a water molecule coordinated to the sixth coordination position. Upon substrate binding, this ligand is displaced giving a penta-coordinate high spin heme (30,31). P450epoK is somewhat different, because the epothilone B and D complexes give a low spin species (spectra not shown) like substrate-bound P450BS␤, which has a water molecule at the sixth coordination site (32). Consistent with a low spin heme, the structure reveals that a water molecule, which is H-bonded to the epothilone B epoxide O atom (Fig. 5b), is coordinated to the heme iron. The substrate-free and epothilone D-bound structures are low spin and expected to be hexa-coordinate. Although there is some electron density at the sixth coordination site, the limited resolution of the substrate-free data, 2.65 Å, precludes a clear identification of a single water molecule coordinated to the heme iron.
Conformational respectively. The plausible substrate entry channel is indicated. Secondary structures were assigned by using PROCHECK (46). b, active site structure of epothilone D-bound P450epoK. Thr 258 , which is important to oxygen activation, the proximal heme ligand, Cys 365 , and the residues involved in heme anchoring are displayed as stick models.  (38). P450epoK represents the second example where all three structures are known. In P450cam there is very little difference in structures between the three forms, whereas with P450BM-3, there are large changes between the substrate-free and -bound structures primarily due to motions of the F and G helices and the F/G loop (34), whereas with P450 2C5 similar but smaller changes are observed (37). Large changes in these regions also have been observed upon ligand binding for a thermophilic P450, CYP119 (39). Given that epothilone is the largest substrate of a P450 with known structure, we anticipated some significant changes in structure upon substrate binding. However, the differences between the substrate-bound and -free P450epoK structures are small and are confined to the F/G loop and BЈ1 helix regions (Fig. 6, a and b) generally thought to provide the entry point for substrates in P450s (24). The BЈ1 helix moves closer toward and the F/G loop moves away from the active site to optimize interactions with the substrate. These differences are similar to what was observed in the P450BM-3 substrate-bound (34) and -free (35) forms except the changes are much smaller in P450epoK. It appears that the plausible substrate access channel, indicated in Fig. 2a, is similar to what has been proposed for P450BM-3 and P450cam. However, P450epoK must be able to adopt an even more open conformation to allow substrate entry and product release. That the substrate-free P450epoK crystallized in the nearly closed form is very likely due to crystal packing forces favoring the substrate-bound closed conformation as opposed to the postulated open conformation. With the substrate-free P450BM-3 just the opposite occurred: crystal packing favored the open form (35).
In addition to a small adjustment in the backbone, various side chains move to accommodate the substrate. Most notable is Arg 71 in the loop between B and BЈ1 helix, the guanidium group of which moves about 4 Å closer to the active site (Fig.  6c). This movement enables Arg 71 to hold two water molecules close to the substrate, especially the thiazole ring. In addition, the movement of Arg 71 enables the thiazole ring to form astacking interaction with Phe 96 , whose side chain provides a key substrate contact.
Epothilone/P450epoK Interactions-As found for P450eryF (28), there are a number of water molecules around epothilone that form H-bonding bridges between epothilone and protein atoms (Fig. 7). The H-bond network involves 6 residues for the epothilone D-bound form and 7 residues for the epothilone B-bound form, respectively. Most of the residues involved in these backbone H-bonds are small, such as Gly and Ala, which provide the required room for the large substrate. Arg 71 , which, as already mentioned, moves into the active site upon epothi-  6. Structural changes upon epothilone D-binding. a, r.m.s. deviation between substrate-free and epothilone D-bound P450epoK plotted against residue number with the helical regions indicated. The location of Arg 71 is highlighted. b, superposition of C␣-traces of substrate-free (gray) and substrate-bound (green) P450epoK. The orientation is the same as in Fig. 2a. The heme and substrate are shown as stick models. c, stereo view showing conformational changes around the epothilone-binding site. The heme and amino acid residues, which are important for the epothilone binding and catalytic activity, are shown in gray for substrate-free and in green for epothilone D-bound forms. Epothilone D is shown as a stick model (carbon, cyan; nitrogen, blue; sulfur, orange; and oxygen, red). lone binding, is the only the side chain that forms part of the H-bond network with the substrate. Phe 96 , which formsstacking interactions with the thiazole ring and the guanidium group of Arg 71 , is very likely a critical residue for epothilone specificity. In addition to H-bonding and aromatic interactions, Leu 183 in the F helix and Ala 250 in the I helix form non-bonded contacts with the substrate/product thiazole ring (Fig. 7).
Similar to other P450s, the carbons to be oxidized are close to the heme iron (Figs. 5a and 7a). In this case, the C12-C13 double bond reacts with the iron-linked oxygen atom to give an epoxide. The distances of C12-Fe and C13-Fe are 4.7 and 5.0 Å, respectively, which are comparable to the distance between the iron and the carbon atom to be hydroxylated in P450cam and P450eryF (4.2 and 4.7 Å, respectively). The substrate orientation is suitable for synchronous oxygen transfer from an Fe(IV)-O species.
The conformation of P450epoK-bound epothilones is somewhat different from the crystal structure of free epothilone (2). The thiazole side chain is perpendicular to the macrolide ring in P450epoK, whereas free epothilone has the thiazole moiety in-plane with the macrolide ring. However, the conformation of the macrolide ring predicted to bind to tubulin (31) is the same as that bound to P450epoK.

Implication for Tubulin/Epothilone
Interactions-Both paclitaxel and epothilone bind to tubulin to stabilize microtubules, arresting cells in mitosis. Epothilones competitively inhibit the binding of paclitaxel to polymerized tubulin, suggesting that the two types of antitumor compounds share an overlapping binding site in tubulin (7,8). Therefore, it is of interest to see if the binding of epothilones to P450epoK mimic to any degree the interactions found between paclitaxel and tubulin (40). As shown in Fig. 8, the C2-phenyl ring of paclitaxel is the structural homologue to the epothilone thiazole ring. Both formstacking interactions with a neighboring aromatic ring (His 229 of ␤-tubulin and Phe 96 of P450epoK) in addition to other nonbonded contacts with aliphatic side chains (Leu 217 of ␤-tubulin and Ala 250 of P450epoK). The aromatic moiety of epothilone is critical, because its removal or direct attachment of the aromatic moiety to C15 results in the loss of tubulin binding and cytotoxic properties (41). Therefore, it is reasonable to expect that the thiazole ring of epothilone makes similarstacking interaction when it binds to tubulin. Structure-activity relationship studies on epothilones have also shown that the location of the N atom in the aromatic ring is important for tubulin binding activity, because the 2-pyridyl-and 2-thiazyl-containing compounds exhibit properties comparable to the natural product, but 3-pyridyl and 5-thiazyl compounds were much less active (41). In the epothilone-bound P450epoK structures, the N atom of the thiazole ring H-bonds to water molecules, which interact with the enzyme and macrolide ring. Therefore, it is interesting to note that the predicted site for the thiazole ring interaction in tubulin places the ring close to potential Hbonding groups such as Asp 226 , peptide groups, or a water molecule.
A greater challenge is predicting the macrolide ring binding site on tubulin, although there is a molecular modeling studies in epothilone/tubulin interaction (42). Drug sensitivity data for several paclitaxel-and epothilone-resistant cell lines provide some additional clues. 1A9PTX10 cells derived from a human ovarian carcinoma cells shows paclitaxel resistance (42). These cells have the mutation F272V in ␤-tubulin, which would be expected to affect paclitaxel binding, because Phe 272 is located near the taxane ring of paclitaxel (Fig. 8). On the other hand, it would be expected that epothilone is located farther away from this site, because these cells retained significant epothilone B sensitivity. Two other mutations, T276I and R284Q, impair epothilone sensitivity in 1A9/A8 and 1A9B10 cells, respectively (43), while exhibiting only a minor effect on paclitaxel sensitivity further indicating the taxane ring of paclitaxel and macrolide ring of epothilone bind differently. From these arguments, we propose that the macrolide ring of epothilone binds close to Thr 276 and Arg 284 while the thiazole ring is in the crevice between His 229 and Leu 217 . To locate the macrolide ring near Thr 276 and Arg 284 , the conformation of the macrolide ring must flip by about 90°relative to the thiazole ring observed in the P450epoK structure. Such an adaptation of epothilone to the putative binding site would be possible owing to flexibility of epothilone as indicated by the quite different conformations of free epothilone and epothilone bound to P450epoK.
Another class of tubulin-stabilizing anticancer drug, sarcodictyin and eleutherobin (Fig. 1, c and d), also have a fivemembered aromatic group, methyl imidazole, attached to the main frame with the linker (-OCOCHϭCH-), which is of similar length to the epothilone linker (-CHϭC(CH 3 )-). In addition, as found for epothilone, eleutherobin is a competitive inhibitor of the paclitaxel binding to tubulin (44). These common structural features of drugs, 5-or 6-member aromatic rings, together with the similarstacking interactions for paclitaxel/tubulin and epothilone/P450epoK, suggest that these aromatic rings are indispensable for tubulin binding and/or stabilization of polymerized tubulin.

CONCLUSIONS
Although epothilone is the largest natural substrate for P450 with known structure, the overall fold of P450epoK is very similar to other P450s. Unexpectedly, the structures of the substrate/product-bound P450epoK are very similar to the substrate-free structure with no major changes in the substrate access channel. Because P450epoK must undergo a large open/ close motion to allow substrate to enter, we attribute crystal packing forces as the reason substrate-free P450epoK crystallized in the closed form. The largest change observed is in Arg 71 , which moves into the active site to help critical water molecules maintain H-bonds to the substrate/product. Similarities of how the thiazole ring of epothilone interacts in the P450epoK active site to how paclitaxel interacts with tubulin together with the mutant data suggest a possible binding mode for epothilone to tubulin. FIG. 8. Electron crystallographic structure of paclitaxel bound to ␤-tubulin (40). Paclitaxel and ␤-tubulin are shown as stick (yellow, carbon; blue, nitrogen; and red, oxygen) and surface models, respectively. Orange surface shows the position of Phe 272 , which is replaced with Val in ␤-tubulin of paclitaxel-resistant cells (42). Green surface shows the position of Thr 276 (lower) and Arg 284 (upper), which are replaced with Ile and Gln, respectively, in ␤-tubulin of epothiloneresistant cells (43).