Identification of Two Cyclooxygenase Active Site Residues, Leucine 384 and Glycine 526, That Control Carbon Ring Cyclization in Prostaglandin Biosynthesis*

The cyclooxygenase (COX) reaction of prostaglandin (PG) biosynthesis begins with the highly specific oxygenation of arachidonic acid in the 11 R configuration and ends with a 15 S oxygenation to form PGG 2 . To ob- tain new insights into the mechanisms of stereocontrol of oxygenation, we mutated active site residues of human COX-2 that have potential contacts with C-11 of the reacting substrate. Although the 11 R oxygenation was not perturbed, changing Leu-384 (into Phe, Trp), Trp-387 (Phe, Tyr), Phe-518 (Ile, Trp, Tyr), and Gly-526 (Ala, Ser, Thr, Val) impaired or abrogated PGG 2 synthesis, and typically 11 R -HETE was the main product formed. The Gly-526 and Leu-384 mutants formed, in addition, three novel products identified by LC-MS, NMR, and circular dichroism as 8,9–11,12-diepoxy-13 R -(or 15 R )-hy-dro(pero)xy derivatives of arachidonic acid. Mechanistically, we propose these arise from a free radical intermediate in which a C-8 carbon radical displaces the 9,11-endoperoxide O–O bond to yield an 8,9–11,12-diep-oxide that is finally oxygenated stereospecifically in the 13 R or 15 R configuration. Formation of these novel products signals an arrest

The oxygenation steps in prostaglandin synthesis involve a series of free radical reactions that convert the arachidonic acid substrate into the prostaglandin endoperoxide product, PGG 2 (1,2). Two molecules of oxygen are incorporated into the fatty acid and five new centers of chirality are created in the process (for recent reviews, see Refs. [3][4][5][6]. Steric control by the cyclooxygenase (COX) 1 enzyme is almost perfect, such that the only significant products in addition to PGG 2 are small percentages of the mono-hydroperoxy fatty acids 11R-hydroperoxyeicosatetraenoic acid (11R-HPETE), 15R-HPETE, and 15S-HPETE (7,8). These by-products are formed at the two positions on arachidonic acid that are normally oxygenated during the transformation to prostaglandins. At the separate peroxidase site on the enzyme, PGG 2 is ultimately reduced to its 15-hydroxy analog, PGH 2 , and similarly, the HPETEs to their hydroxyl analogs, HETEs. The order of the individual steps in the main reaction is well understood. What is not so well understood is the structural or mechanistic basis for the exquisite control of stereochemistry throughout the process. Different experimental approaches are being pursued to answer this question including the use of EPR to probe radical intermediates (9,10), protein crystallization and x-ray analysis (11,12), computational modeling (13), and site-directed mutagenesis (14,15).
Most studies using site-directed mutagenesis of amino acids in the active site of COX enzymes have been concerned primarily with identifying residues intimately involved in substrate or inhibitor binding, or in initiating catalysis (16 -20). Nonetheless, as a result of all these studies a substantial body of data has accumulated on the effects of mutations on the profile of oxygenated products (14). One remarkable conclusion stemming from these mutagenesis experiments is how well the enzyme can tolerate mutations and still maintain synthesis of the normal prostaglandin stereochemistry. The typical consequence of mutations, if they alter the products, is to increase the proportion of one or other of the HPETE by-products; at the same time, any PG endoperoxide formed is PGG 2 with the usual stereochemistry. An exception to this rule, that the process of PG synthesis remains unaffected, is our recent report that small but specific changes to residues surrounding C15 of arachidonic acid in the COX active site could invert the configuration of the 15-hydroperoxy group of PGG 2 from 15S in wild-type to 15R in mutant COX enzymes (15). Of the two residues identified, Ser-530, 2 the target of acetylation by aspirin, had the more crucial role in preserving the usual 15S stereochemistry. Its replacement with the larger residues methionine, valine, or threonine led to formation of a prostaglandin product that was 82-95% of the 15R configuration. The second residue identified, Val-349, upon specific mutation only to isoleucine led to a 40 or 60 -65% inversion of the C15 configuration in COX-1 and COX-2, respectively (15).
With this precedent in mind, the initial objective of the studies reported here was to identify active site residues involved in stereocontrol of the initial oxygenation at C11. Five amino acids were selected for mutation based on their proximity to C11 of arachidonic acid as it appears when bound in the active site of the COX-1 isozyme (11). Although several of these amino acids have been targeted previously for mutation, the effects on product stereochemistry have not been reported. The approximate location of the five selected residues in relation to the bound arachidonic acid is depicted in schematic form in Fig.  1 (encircled residues). According to the previous findings on stereocontrol at C15 we hypothesized that either increasing space or constricting space around C11 of arachidonic acid would lead to partial or even complete inversion of the oxygenation from the wild-type 11R to the 11S configuration. As noted earlier, 11R-HETE is a regular by-product during the COX reaction. It is thought to arise from a conformation of arachidonic acid bound in the active site that is slightly different from the conformation necessary to yield PGG 2 (21). Whether PGG 2 or 11-HETE is the product, the oxygenation always occurs with high stereospecificity in the 11R-configuration (7,22). As reported here, our attempts to date to influence the stereocontrol at C11 have met with the same extraordinary resistance of the COX enzyme to producing anything other than the natural 11R oxygenation. In the course of these mutagenesis experiments, nevertheless, we identified two residues in the COX-2 active site that, upon mutation, arrest prostaglandin ring cyclization after formation of the 9,11-endoperoxide, and instead complete the catalytic cycle by formation of three novel diepoxyalcohol derivatives of arachidonic acid. Despite the arrest in cyclopentane ring formation, synthesis of these products is completed in the COX active site by a second highly specific oxygenation on the carbon chain at either C13 or C15.

EXPERIMENTAL PROCEDURES
Generation and Expression of Mutant COX-2 Enzymes-Mutant enzymes were prepared using human COX-2 cDNA in the pcDNA3.1 vector (Invitrogen) and appropriate oligos in the Quikchange site-directed mutagenesis kit (Stratagene) essentially as described before (15). All mutants and wild-type COX-2 were transiently transfected into HeLa cells together with co-transfection of a recombinant virus containing the T7 RNA polymerase gene (23).
The G526S mutant and wild-type COX-2 were also prepared in the pVL1393 baculovirus transfer vector (BD Biosciences) and expressed in Sf9 insect cells. The human COX-2 cDNA used for this expression system contained a His 6 tag located after the signal cleavage site and was kindly provided by Dr. David DeWitt, Michigan State University (24). At 72-h post-viral infection, Sf9 cells were pelleted by centrifugation and disrupted by freeze-thaw lysis. The pellet was resuspended in 20 ml of resuspension buffer consisting of 80 mM Tris-HCl, pH 7.2, 100 mM NaCl, 0.1 mM diethyldithiocarbamate, 0.5 mM phenylmethylsulfonyl fluoride, and 20 l of protease inhibitor mixture. After centrifugation at 100,000 ϫ g for 45 min at 4°C the pellet was resuspended in 30 ml of the same buffer and passed through a glass homogenizer several times. While stirring at 4°C, an 11% solution of CHAPS was added dropwise to the homogenate to a final concentration of 1%, and stirring was continued for an additional 2 h. After centrifugation at 100,000 ϫ g for 1 h at 4°C the supernatant was mixed overnight with 2 ml of Ni-NTA resin (Qiagen), pre-equilibrated in the resuspension buffer containing 20 mM imidazole. The slurry was poured into a column support and washed twice with 5 ml of the resuspension buffer each containing 0.5% CHAPS and 20 mM or 50 mM imidazole, respectively. The COX-2 protein was eluted with 10 ml of the washing buffer containing 200 mM imidazole. Fractions of 2 ml were collected and analyzed by SDS-PAGE. Imidazole was removed by dialysis of the pooled fractions (typically the first two fractions) against 25 mM Tris-HCl, pH 8.0, containing 100 mM NaCl and 0.1% CHAPS. This purification protocol yielded Ͼ85% pure protein as estimated by SDS-PAGE analysis. The final concentration of the enzyme preparations was 1.6 g/l (23 M) for wild-type and 2.7 g/l (39 M) for G526S COX-2.
Incubations and Extraction of Products-HeLa cells were harvested 24 h post-transfection, washed with phosphate-buffered saline, and sonicated on ice in incubation buffer (100 mM Tris-HCl, 0.5 mM phenol, pH 8.0). Protein concentration was determined using the Bradford protein assay (BioRad). To determine the relative conversion of arachidonic acid by wild-type and mutant COX-2, equal amounts of cells in 100 l incubation buffer were reacted with 50 M [ 14 C]arachidonic acid for 10 min at room temperature in the presence of 1 M hematin. The reaction was terminated by the addition of 250 l of methanol and 125 l of methylene chloride. After centrifugation the supernatant was removed and evaporated from the organic solvents, dissolved in 50 l of methanol and 900 l of 0.1% acetic acid. The samples were loaded onto preconditioned 100 mg Oasis HLB cartridges (Waters). After washing with water, products and unreacted substrate were eluted using 1 ml of methanol.
Large scale incubations of purified G526S COX-2 from the baculovirus expression system were conducted in 10 parallel reactions. Each reaction consisted of 2 ml of incubation buffer containing 1 M hematin and 50 l of 39 M G526S COX-2. Reactions were started by addition of 50 g of arachidonic acid dissolved in 10 l of ethanol. To one of the reactions 3 ϫ 10 6 cpm [ 14 C]arachidonic acid were added additionally. After 10 min at room temperature, the reactions were terminated by addition of 200 l of methanol and 16 l of glacial acetic acid. Each reaction was loaded immediately onto a preconditioned 100 mg Oasis HLB cartridge (Waters), washed with water, and products and substrate were eluted with 1 ml of methanol. In some of the experiments the pooled eluates were treated with 0.2 mg/0.5 ml triphenylphosphine.
Product Analyses-Products extracted from HeLa cell incubations were analyzed by RP-HPLC using a Waters Symmetry C18 5-m column (0.46 ϫ 25 cm) eluted at a flow rate of 1 ml/min with a stepwise gradient of the solvents: acetonitrile/water/acetic acid 37.5:62.5:0.01 (by volume) for 15 min, then acetonitrile/water/acetic acid 70:30:0.01 (by volume) for 15 min, and finally with methanol. Prior to HPLC analysis unlabeled standards of HETEs (0.2 g each) and PGE 2 (2 g) were added to facilitate UV-detection of the peaks. Chromatography was monitored using an Agilent 1100 series diode array detector coupled on-line to a Packard A140 Radiomatic liquid scintillation counter. Initially, only 20% aliquots of the samples were analyzed by RP-HPLC to monitor the product profile. For chiral analysis of PGE 2 and 11-HETE the remaining 80% of each sample were injected in a second RP-HPLC run under identical conditions. Products were collected and further analyzed using SP-HPLC and chiral phase HPLC as described (15,25).
The products formed by the purified G526S mutant COX-2 were analyzed and isolated by RP-HPLC using a Waters Symmetry C18 5-m column (0.46 ϫ 25 cm) eluted with a solvent of acetonitrile/water/ acetic acid 50:50:0.01 (by volume) at a flow rate of 1 ml/min. After 15 min, the solvent was changed to methanol to elute 11-HETE and unreacted arachidonic acid. Chromatography was monitored using the diode array detector coupled on-line to the liquid scintillation detector. Following reduction with triphenylphosphine, two peaks were collected from RP-HPLC with retention times of Ϸ10 min and Ϸ13 min, extracted using methylene chloride, and evaporated to dryness under a stream of nitrogen. SP-HPLC of the early eluting peak gave a single peak at retention time 26.5 min using a Beckman Ultrasphere Si 5-m column (0.46 ϫ 25 cm) eluted with a solvent of hexane/isopropyl alcohol/acetic acid 100:5:0.1 (by volume) at a flow rate of 0.5 ml/min. The second peak from RP-HPLC was either methylated using ethereal diazomethane or chromatographed without methylation. Diazomethane was prepared using the Aldrich generator (www.sigmaaldrich.com/Brands/Aldrich/ Technical Bulletins.html). Using identical SP-HPLC conditions as above, the second RP-HPLC peak (after methylation) resolved into two products with retention times of Ϸ18 and Ϸ21 min.
Kinetic Analyses-Rates of reaction of wild-type and G526S COX-2 enzymes were determined from oxygen consumption. A Clark-type oxygen electrode was used. The reactions contained 2 ml of 100 mM Tris-HCl, 0.5 mM phenol, pH 8.0 supplemented with 1 M hematin maintained at 27°C. Wild-type and mutant enzyme was added at a final concentration of 0.12 M in the assays (16 g of protein). The reaction was initiated by addition of 100 M arachidonic acid, and oxygen uptake was monitored for 5 min. Rates were calculated from the slope of the linear curve over the first 15-30 s, and assuming an initial oxygen concentration of 240 M in the buffer.
LC-ESI-MS and NMR-For LC-ESI-MS analyses a Thermo Finnigan LC Quantum system was used. Samples isolated by SP-HPLC were introduced via a Waters Symmetry C18 3-m column (0.2 ϫ 10 cm) eluted with a water/acetonitrile gradient containing 10 mM ammonium acetate at a flow rate of 0.2 ml/min. The heated capillary ion lens was operated at 220°C. Nitrogen was used as a nebulization and desolva-tion gas. The electrospray potential was held at 4 kV. Mass spectra were acquired over the mass range m/z 100 -1,000 at 2 s/scan. 1 H NMR and H,H COSY NMR spectra were recorded on a Bruker Avance DRX 400 MHz or 500 MHz spectrometer. The ppm values are reported relative to residual non-deuterated solvent (␦ ϭ 7.24 ppm for C 6 H 6 ; ␦ ϭ 1.92 ppm for CH 3 CN). 10 l of D 2 O were added to some of the samples dissolved in benzene-d 6 , and the samples were analyzed again.
CD Spectroscopy-CD spectra were recorded on a JASCO J-700 spectropolarimeter. The configuration of the hydroxyl group was assigned based on Cotton effects in the spectrum of the benzoate ester derivative. The benzoate group has a transition moment that is sufficiently close to that of the double bond adjacent to the hydroxyl group. This allows efficient coupling of the two chromophores, however, due to the relatively low max of the chromophores only the Cotton effect at higher wavelength is observed which is still sufficient to determine the absolute configuration of the molecule (26,27). The methyl ester derivatives of the three diepoxyalcohol products 1-3 (about 25 g each) were dissolved in 50 l of dry acetonitrile and reacted with 1 l of benzoyl chloride in the presence of 1 l DBU and a few grains of dimethylaminopyridine at room temperature overnight. After evaporation of the solvent, 500 l of water were added and the products were extracted using 1 ml of dichloromethane. The methyl ester, benzoate derivatives were isolated by RP-HPLC using a Waters Symmetry C18 5-m column (0.46 ϫ 25 cm) eluted with a solvent of methanol/water/acetic acid (95:5:0.01 by volume) at a flow rate of 1 ml/min and UV detection at 235 nm. The products eluting at 5 min were collected and extracted from the HPLC solvent using dichloromethane. 1 H NMR (400 MHz, in CD 3 CN) was used to confirm the structure of the derivative and to determine the coupling constants between H14 and H15 (product 1) and between H13 and H14 (products 2 and 3). In each case the existence of the conformer in which the geminal hydrogen at the carbon bearing the hydroxyl group is eclipsed with the hydrogen of the double bond was evident from the large coupling constant of 6.3 Hz (product 1) and 9.1 Hz (products 2 and 3). The methyl ester, benzoate derivatives were dissolved in acetonitrile to a final OD of 1.2-1.4 AU at 228 nm for recording of the CD spectra.

RESULTS
Activity of COX-2 Mutants-We prepared a series of human COX-2 enzymes with mutations of residues Leu-352, Leu-384, Trp-387, Phe-518, and Gly-526 (Fig. 1). The carbons on the side chains of these amino acids are located within 3-6 Å of the carbon chain of arachidonic acid, as seen in the crystal structure of ovine COX-1 (11), and the associated hydrogen atoms are closer still (Table I). Mutations were selected with a view to changing the available space in the oxygenase active site while preserving the overall hydrophobic nature of the pocket. The mutations tested were a change of Leu-352 to Ala, Ile, Gln, Met, or Val; of Leu-384 to Phe or Trp; of Trp-387 to Phe or Tyr; of Phe-518 to Ile, Trp, or Tyr; and of Gly-526 to Ala, Ser, Thr, or Val. All mutants and the wild-type COX-2 were expressed in HeLa cells and subsequently incubated with 50 M [ 14 C]arachidonic acid and the product profile analyzed by RP-HPLC. Fig. 2 summarizes the analytical data for all mutants, showing the activity as assessed by the conversion of radiolabeled substrate and the product profile. All mutants were catalytically active, many with similar activity to wild type. For the purposes of this comparison, the products shown for wild-type COX-2 are PGE 2 , (quantitatively accounting for 80 -90% of the prostaglandins produced), 11-HETE and 15-HETE, taking their total as 100%. As can be seen on the left side of Fig. 2, the five different mutations to Leu-352 had almost no effect, either on activity or on the profile of products. In the middle part of Fig. 2, the three mutations to Phe-518 and two mutations to Trp-387 can be seen to increase the relative levels of 11-HETE and 15-HETE at the expense of PGE 2 . On the right side, illustrating the results with the Leu-384 and Gly-526 mutations, the appearance of two new HPLC peaks is indicated by the black bars and white bars; among the Leu-384 and Gly-526 mutant COX-2 enzymes, only the L384W and G526A mutants retain the ability to synthesize PGE 2 .
For several of these mutations, the identity and the C15 stereochemistry of the [ 14 C]PGE 2 product was analyzed using a straight-phase HPLC system that easily resolves 15R-PGE 2 and PGE 2 (15). This analysis was carried out on PGE 2 formed by the L352Q, F518I, F518Y, L384W, and G526A COX-2 enzymes, the latter two being the only ones that synthesized both PGE 2 and the new products to be identified below. All these PGE 2 analyses found exclusively natural (15S) PGE 2 (Fig. 3A). 11-HETE is prominent in all the mutants that have an altered profile from wild type, and it is the dominant product in each of the Gly-526 mutant enzymes. The stereochemistry of the 11-[ 14 C]HETE was measured by chiral column HPLC using a method that gives a 2-min baseline separation of the two enantiomers (25). In all cases with sufficient counts to give a strong signal on the on-line radiodetector, greater than 99% of the counts eluted with the 11R enantiomer and in several cases there was no visible 11S peak (Fig. 3B). These 11-HETE analyses indicated highly stereoselective 11R-HETE synthesis, often Ͼ99% 11R, the lowest measurement being 96% 11R for the F518I mutant. Fig. 4 shows a comparison of the RP-HPLC analyses of products formed by wild-type COX-2 and the G526A, G526S, L384F mutant enzymes. As noted earlier, formation of PGE 2 was reduced in the G526A and L384W mutants, absent in the G526S, G526T, G526V, and L384F mutants, and 11R-HETE was a major product. As seen in the middle of the chromatograms, two peaks of intermediate polarity are present that  were not formed by wild-type COX-2. (Subsequently, the second peak was further resolved by SP-HPLC to reveal a minor third product, the three being designated as 1, 2, and 3.) As the G526S COX-2 mutant gave a good yield of the novel products in these transient transfection studies in HeLa cells, this cDNA was selected for transfer into a baculovirus system for expression in Sf9 insect cells. The baculovirus COX-2 construct contained a His 6 -tag located immediately after the N-terminal signal peptide, allowing subsequent purification of the mutant enzyme by affinity chromatography (24). This enzyme preparation was used in large scale incubations with arachidonic acid to isolate sufficient material for subsequent spectroscopic analyses.

Detection of Novel Products from Mutant Cyclooxygenases-
Identification of Novel Products Formed by the G526S Mutant COX-2-Using the purified G526S mutant COX-2, a slightly more complicated profile of products was evident on RP-HPLC analysis of the [ 14 C]arachidonic acid metabolites. Two additional peaks at 14 and 17 min appeared (designated c and d in Fig. 5A). Since these peaks were absent in the incubations with the HeLa cell homogenates, they were likely the corresponding hydroperoxy precursors of the novel products. In support of this conclusion, triphenylphosphine reduction of the G526S reaction mixture eliminated the extra products and increased the size of the original two peaks (designated a and b in Fig. 5B). We surmise that the novel products are formed as hydroperoxides and that these are poor substrates for the peroxidase activity of the purified G526S COX-2, whereas the HeLa cell homogenates contain additional peroxidases that completed the reduction.
The peaks eluting at Ϸ10 min and Ϸ13 min upon triphenylphosphine reduction were collected from the RP-HPLC system, and further analyzed using SP-HPLC. The early peak from RP-HPLC eluted as single peak from SP-HPLC (Fig. 5C,  product 1), while the second peak from RP-HPLC resolved into two products in ϳ4:1 ratio (Fig. 5D, products 2 and 3, as the methyl esters). The novel products formed by the G526A and L384F mutants (Fig. 4, B and D) were also analyzed using RPand SP-HPLC and were found to co-chromatograph on both systems with the G526S products. NMR Analyses-An overall appraisal of the 1 H and H,H COSY NMR spectra of products 1, 2, and 3 established that they share several structural elements in common. Each contains the full-length carbon chain with two double bonds, one hydroxyl, and two epoxide groups. They each are linear molecules with no carbon ring. These structural features predict the correct molecular weight of 352. The detailed analysis of each product is described as follows, beginning with the quantitatively most abundant product 2, identification of which formed the basis for characterization of 1 and 3.
Structure of Product 2-In the 1 H NMR spectrum of the methyl ester of product 2 four signals appeared with a chemical shift compatible with epoxide protons (␦ ϭ 3.16, 2.96, 2.90, and 2.76 ppm) (Table II). H,H COSY analysis located the epoxides to carbons 8, 9 and 11, 12 (Fig. 6). In a NOESY experiment a cross-peak of the signals of H8 and H9 appeared which identified the configuration of the 8,9 epoxide as 8,9-cis. This is in accord with the coupling constant of J 8,9 ϭ 4.1 Hz. Assignment of the cis or trans epoxide configurations described here is supported by reference to several reports on the NMR properties of compounds with related partial structures (28 -38). Lack of a NOESY cross-peak between protons H11 and H12 identified the epoxide configuration as 11,12-trans, again in accord with the coupling constant, J 11,12 ϭ 2.2 Hz (28,38). In the H,H COSY spectrum, H12 was connected to the geminal proton of a hydroxyl group (␦ ϭ 4.56 ppm), located at C13. The geminal H13 was connected to the 14,15-cis double bond protons, and also to the OH proton of the C13 hydroxyl group located at ␦ ϭ 1.64 ppm. (The relatively low chemical shift of the OH may be characteristic of using benzene as solvent, as we have encountered another such example in analysis of 11-hydroxylinoleate (39)). Both H11 and H9 coupled to the two methylene protons at C10, H10a, and H10b. The protons at C7 were split into two signals with discrete chemical shifts (␦ ϭ 2.11 ppm and 2.25 ppm), and both were coupled to H6 of the 5,6-cis double bond.
The relative configuration of the hydroxyl at C13 was established using two independent approaches. For determination by 1 H NMR, the sample in d 6 -benzene was re-analyzed after addition of 10 l of D 2 O to the solution. On addition of this trace of D 2 O, the 13-OH signal disappears due to exchange, and the corresponding splitting of the geminal H13 signal is simplified to a dd multiplicity, allowing calculation of the coupling constant J 12,13 . The coupling constant of J 12,13 ϭ 3.2 Hz indicates an erythro arrangement of the protons at carbons 12 and 13 (30), and hence the R-configuration at C13. The configura-

FIG. 2. Product formation and relative conversion of arachidonic acid by wild-type and mutant human COX-2 enzymes expressed in HeLa cells.
Incubations were conducted and analyzed by RP-HPLC and chiral-phase HPLC as described under "Experimental Procedures." The percentage (%) of the total of the products PGE 2 , 15-HETE, 11-HETE, and the novel products 1, 2, and 3, is given. Under the bar graph, the relative activity of the mutants is scored as the conversion of arachidonic acid compared with the wild-type set at 100%; ϩϩϩϩ, 75-100%; ϩϩϩ, 50-75%; ϩϩ, 25-50%; ϩ, Ͻ25%.
tion of the 13-hydroxy group was independently determined by CD spectroscopy using the benzoate derivative (Fig. 7) (40, 41). The CD spectrum showed a positive first Cotton effect at 228 nm (⌬⑀ ϭ 30.0) demonstrating positive chirality between the 13-hydroxy group and the 14,15 double bond, and predicting the R-configuration for C13 (Fig. 5), completely consistent with the results from NMR.
The absolute configuration of the epoxide carbons in product 2 was deduced from the known configuration of the intermediate 9R,11R-endoperoxide (according to the mechanism of formation proposed in Discussion and in Fig. 9). Both stereocenters do not change their absolute configuration during the formation of the epoxide, and therefore the 11,12trans epoxide is 11R,12R, and the 8,9-cis epoxide is 8R,9S (there is a reversal in the assignment of priorities at C9 once the epoxide is formed leading to a change in the designation of 9R to 9S) (Fig. 8). The spectroscopic data identified product 2 as 13R-hydroxy-8R,9S,11R,12R -diepoxyeicosa-5Z,14Z-dienoic acid.
Structure of Product 1-The 1 H and H,H COSY NMR analyses of product 1 (free acid) showed a similar 8,9-cis,11,12trans epoxide structure, but C12 was directly connected to the 13,14 double bond, and the hydroxyl group (␦ ϭ 3.94 ppm) was located at C15 (Table III). The presence of the 15-hydroxyl group shifted the signal for H14 downfield (␦ ϭ 5.87 ppm) from the 5,6 double bond signals allowing determination of the coupling constant as J 13,14 ϭ 15.6 Hz. Accordingly, the configuration of the 13,14 double bond is trans. CD spectroscopy of the benzoate derivative was used to determine the C15 configuration which was found to be 15R (Fig. 7). This is the opposite configuration as compared with the 15-hydroxy in PGH 2 . Using the same lines of evidence for deducing the absolute configuration of the other stereocenters mentioned above, product 1 was identified as 15R-hydroxy-8R,9S(cis)11R,12R(trans)-diepoxyeicosa-5Z,13E-dienoic acid. Consistent with the 11,12-epoxy-13-ene-15-hydroxyl moiety in product 1, this epoxyalcohol was found to be chemically the most labile of the three novel products, and prone to acid-catalyzed hydrolysis.
Structure of Product 3-The 1 H NMR spectrum of the methyl ester of product 3 was very similar to the spectrum of product 2, except for a notable difference in chemical shifts for the signals of H8 and H9 (Table IV). Lack of a NOESY cross-peak between the two signals and the small coupling constant of J 8,9 ϭ 2.0 Hz implicated trans configuration of the 8,9-epoxide. The hydroxyl was located at C13 (␦ ϭ 4.56 ppm), adjacent to the 14,15-trans double bond. Addition of D 2 O to the NMR sample dissolved in d 6 -benzene simplified the coupling pattern of H13 to a dd multiplicity, and the J 12,13 coupling constant was determined as 3.4 Hz. This is compatible with an erythro arrangement of H12 and H13 and implicates R-configuration of C13 (30). Again, the configuration of the C13 hydroxyl group was independently determined using CD spectroscopy of the benzoate derivative, and  6. H,H COSY analysis of product 2 formed by the G526S human COX-2 mutant. The methyl ester of product 2 was isolated using RP-and SP-HPLC and analyzed in benzene-d 6 using a Bruker DRX 400 MHz spectrometer ("Experimental Procedures"). The structure of the methyl ester of product 2 (13R-hydroxy-8R,9S,11R,12Rdiepoxy-5Z,14Z-eicosadienoic acid) is shown at the top.

FIG. 7.
Determination of the absolute configuration of the hydroxy group in the diepoxyalcohols 1, 2, and 3 using CD spectroscopy. UV and CD spectra of the benzoate derivatives of diepoxyalcohols 1, 2, and 3 were recorded in acetonitrile. For products 2 and 3 the carbon close to the viewer shown in the Newman projection is C13, the carbon in the back is C14, and R 1 represents the proximal carbon chain from C1 to C12. For CD-spectroscopy, the 13-hydroxyl group derivatized to the benzoate (checkered box) is the first chromophore and the 14,15 double bond is the second chromophore. The positive first Cotton effect at 228 nm in the CD spectrum indicates positive chirality between the first and second chromophore in the Newman projection, equivalent to the R-configuration at the chiral center.
FIG. 8. Deduction of the absolute configuration of the 8,9-and 11,12-epoxides in product 2. Top, the endoperoxide in prostaglandin H 2 has the 9R,11R-configuration. Bottom, in product 2, the configuration of carbons 9 and 11 is not changed allowing the designation of the 11,12-trans epoxide as 11R,12R; the absolute configuration of the 8,9cis epoxide is 8R,9S, due to a change in assigning the priorities of the substituents at C9. it was confirmed to be 13R (Fig. 7). The spectroscopic and chromatographic data, together with the implications from the configuration of the 9,11-endoperoxide, identified product 3 as 13R-hydroxy-8S,9S(trans)11R,12R(trans)-diepoxyeicosa-5Z,14Z-dienoic acid.

Rates of Reaction of Wild-type and G526S Mutant COX-2-
The initial rates of reaction of wild-type and G526S COX-2 enzymes were determined using an oxygen electrode. At 100 M arachidonic acid concentration, wild-type COX-2 (0.12 M) consumed 62 M O 2 /min while the G526S mutant (0.12 M) consumed 30 M O 2 /min. Direct comparison of the rates obtained by measuring oxygen uptake is problematic due to the fact that the main product PGG 2 in the wild-type COX-2 reaction incorporates two molecules of oxygen while the main product 11R-HPETE of the G526S COX-2 incorporates only one molecule of oxygen. Assuming that both enzymes form only one major product, PGG 2 by wild-type and 11-HPETE by the mutant, both rates are equivalent to turnover numbers of about 4/sec for the fatty acid substrate. DISCUSSION We began these experiments with the aim of disrupting the normal course of prostaglandin biosynthesis with mutations in the cyclooxygenase active site, in particular with a view to uncovering elements of the stereocontrol of the initial oxygenation of arachidonic acid at C11. So far it has proved impossible to perturb the stereochemistry of the C11 oxygenation, even though there are many mutations that alter the balance of COX products (14). Mutations to Gly-526 and Leu-384 did successfully interrupt the reaction at a later point in catalysis. We can infer a critical role for these two residues in holding the reacting fatty acid radical in place, and allowing reaction from C8 across to C12 with closure of the characteristic 5-membered ring of prostaglandins. Although the reaction is thrown off its normal course, it is significant that, as discussed later, the final products are formed with complete stereocontrol at the last step of oxygenation.
Mechanism of Synthesis of the Diepoxyalcohols-There is a chemical precedent for the formation of the novel diepoxyalcohols we have characterized. This comes from studies on mercuric ion-catalyzed generation of a fatty acid carbon radical ␣ to an endoperoxide (42). Demercuration of a ␥-linolenate derivative produced the equivalent in a C20 fatty acid of a C8 carbon radical next to a 9,11-endoperoxide and 12,14-conjugated diene. This product, directly analogous to an intermediate of prostaglandin biosynthesis, then spontaneously either reacted with molecular oxygen at C8 (again, numbering according to a C20 fatty acid), cyclized to a prostaglandin analog, or rearranged with oxygenation to 8,9 -11,12-diepoxy-15-hydroperoxide diastereomers directly analogous to our product 1. The mechanism of formation through the chemical route is quite clear (42), and provides a basis for explaining the biosynthesis of the diepoxyalcohols in our experiments (Fig. 9). The diepoxyalcohols arise from a free radical intermediate after the step of forming the endoperoxide from C11 to C9, and prior to closure of the 5-membered carbon ring (Fig. 9). Their presence, therefore, signals an arrest in the process of prostaglandin synthesis and uncovers a critical role for Gly-526 and Leu-384 in the normal stereocontrol of the cyclooxygenase reaction.
Location of Residues in the COX Active Site-Gly-526 is depicted on the left hand side of Fig. 10 as the magenta residue with the space-filling "balls" signifying the hydrogen atoms of the ␣-chain CH 2 group. In this three-dimensional structure of COX-1 with bound substrate, carbons 8 through 10 of arachidonic acid are kinked toward Gly-526. Fig. 10B, showing bound PGG 2 in the active site of COX-2 (12), emphasizes the proximity of Gly-526 to the bicyclic ring of the product, and thus, presumably to the partially transformed fatty acid radicals during the cyclooxygenase reaction. Gly-526 is located on a helix between Val-523, a residue with a well known role in the binding of COX-2 selective inhibitors, and Ser-530, the target for acetylation by aspirin (3). Leu-384, the second significant residue highlighted by our results, forms part of the top of the COX substrate binding channel. Its immediate neighbor, Tyr-385, initiates the free radical reaction that transforms arachidonic acid to products (3,5). Leu-384 has closest contacts with carbons 8 through 11 of arachidonic acid. Based on their location, it is reasonable to expect that mutations to Gly-526 or Leu-384 can shift the alignment of the partially reacted substrate from its optimal positioning and interfere with closure of the cyclopentane ring; formation of the carbon-carbon bond from C8 across to C12 becomes impaired or impossible. The mutations of Gly-526 or Leu-384 do not appear to interfere with the initial hydrogen abstraction and the first oxygenation. Changes to the other three residues under study (Leu-352, Trp-387, Phe-518; in yellow in Fig. 10) were tolerated either without significant changes to the catalytic reaction or they led to the preferential formation of HETE products over PGG 2 as observed previously with other mutations (14).
Control of Oxygenation-Despite our attempts to induce the initial oxygenation of arachidonic acid in the 11S configuration by site-directed mutagenesis, all the evidence indicates that the normal 11R specificity is retained in our mutant COX-2 enzymes. The enzymes formed 11R-HETE and further products of the 11R-peroxyl radical, i.e. prostaglandins or the novel diepoxyalcohols. We deduce also that the final C15 oxygenation is unaffected by the mutations. This is evidenced by the formation of normal PGG 2 (and ultimately PGE 2 ) with the usual 15S configuration in two of the mutants that form the novel products and PGG 2 (L384W and G526A) (see "Results" and Fig. 3). The fact remains that the same active site that forms (15S) PGG 2 forms the novel 15-hydroxy diepoxyalcohol with the 15R configuration! This raises an interesting and mechanistically significant issue regarding the mechanism of stereocontrol of the oxygenation process. If the radical intermediates reacting through either pathway were considered to be "substrates" for the final oxygenation step, then the two different substrates are optimal for either 15S or 15R oxygenation. Previously we have shown that specific changes to the COX active site can alter the C15 stereochemistry of the PG product; while wildtype COX-2 makes 15S-products, the S530T mutant makes 15R (15). We postulated that a change in conformation of the radical intermediate induced by the mutated enzyme alters the chirality of reaction with molecular oxygen in the two enzymes (15). Here, in the same enzyme, two reaction pathways form distinct intermediates, and this is also associated with a switch in configuration at C15. Distinct reaction intermediates may acquire distinct conformations in the active site that favor either R or S oxygenation.
Lack of Dioxygenation at C8 -Since the Gly-526 and Leu-384 mutations arrested prostaglandin synthesis before the C8 radical closes the cyclopentane ring to C12, we would have expected to see C8 hydroperoxide products formed as an alternative pathway. This type of reaction is well precedented in the formation of hydroperoxy-endoperoxides during lipid peroxidation (42,43). Remarkably, C8 dioxygenation products were not observed. The mutants match the wild-type enzyme in this regard. C8 hydroperoxides are not formed within the constraints of the cyclooxygenase active site, either because of limited access of oxygen to C8, or because any C8 peroxyl radicals quickly revert to carbon radicals due to lack of a FIG. 9. Proposed mechanism of formation of arachidonic acid diepoxyalcohols 1, 2, and 3 by the G526S mutant human COX-2. The initial hydrogen abstraction, first oxygenation to the 11R-peroxyl radical, and formation of the 9,11-endoperoxide are identical in wild-type and G526S COX-2. In wild-type COX-2 the C8 radical closes the 5-membered prostaglandin ring to C12 (5-exo cyclization) followed by subsequent oxygenation in the 15S-configuration to form PGG 2 . In the G526S mutant ring closure of the C8 radical with C12 is sterically hindered by misalignment of the carbon chains. At this stage, there is partial isomerization of the carbon chain from the 8,9-cis to the 8,9-trans arrangement. Instead of PG ring closure the C8 radical displaces the O-O bond of the endoperoxide to form the 8,9-epoxide. Homolytic cleavage of the endoperoxide yields a C11 alkoxyl radical that forms a trans epoxide with C12 and an allyl radical delocalized from C13 to C15. A second stereospecific oxygenation in the 13R or 15R-configuration forms the primary hydroperoxide products that are subsequently reduced to the diepoxyalcohols 1, 2, and 3.
suitable hydrogen donor that would stabilize the oxygen moiety as the hydroperoxide.
Appearance of 8,9-cis and 8,9-trans Epoxy Products-In normal COX catalysis, the C8 radical reacts rapidly at C12 to form the cyclopentane ring. In the Gly-526 and Leu-384 COX-2 mutants, this ring closure becomes less favorable and therefore other reactions with slower rates become significant. The most obvious of these is attack of the C8 radical on the endoperoxide moiety with subsequent closure of the oxygen bound at carbon-9 to the 8,9-epoxides, both 8,9-cis (major, products 1 and 2) and 8,9-trans (minor, product 3). Formation of both cis and trans epoxides indicates another competing reaction is occurring, namely "flipping" of the pyramidal C8 carbon radical to the alternate configuration with the main carbon chain now lying trans. Normally, the fast C8-C12 ring closure all but precludes the ability to observe this slower isomerization at C8. It is evidenced only by a tiny percentage of 8-iso-PGH 2 as product (44,45).
Comparison with Previous Reports of Unusual COX Products-Oliw et al. (46) characterized the products of oxygenation of 5Z,8Z,11Z-eicosatrienoic acid (Mead acid) by ovine COX-1 and identified 13R-hydroxy-5,8,11-eicosatrienoic acid (13R-HETrE) and the epoxyalcohol 8,9-epoxy-11-hydroxy-5,12-eico-sadienoic acid among the products. Formation of 13R-HETrE involves an oxygenation antarafacial to the normal C13 pro-S hydrogen abstraction of COX catalysis and therefore the 13R-OH can be considered the expected and predicted product (2,47). The implication for our finding of 13R-hydroxy-diepoxyalcohols 2 and 3 is that their stereochemistry is also the norm for COX catalysis. The formation of 13R-hydroxy-diepoxyalcohol products 2 and 3 is also reminiscent of the occurrence of the 13-hydroxy isomer of PGH 2 isolated by Hecker et al. (7) as a minor by-product from large scale oxygenations of arachidonic acid with the purified ovine COX-1. It is likely that the 13hydroxy-PGH 2 was also formed in the R configuration.
The 8,9-epoxy-11-hydroxy derivative of Mead acid (46) is a closely analogous product to our diepoxyalcohols. The only difference arises in the final transformation of the C11 alkoxyl radical (derived from opening of the 9,11-endoperoxide) to 11hydroxyl (in Mead acid) or to 11,12-epoxy-13-hydro(pero)xide (or 15-hydro(pero)xide) as in our products. The reactions we describe are completed by a stereospecific oxygenation, proving the involvement of the enzyme, whereas Oliw et al. (46) postulated the formation of the epoxyalcohol (and a triol) by nonenzymatic processes. One reasonable line of argument to support of their conclusion was the finding of triols derived from FIG. 10. Location of residues Gly-526 and Leu-384 in the COX active site (stereoview). A, crystal structure of ovine COX-1 with arachidonic acid bound in the COX active site (11). Relative to the view in Fig. 1, the picture is turned by about 120°to the right. The carbon atoms of arachidonic acid are shown in green with the carboxylate coordinated to Arg-120 and Tyr-355 (red). The amino acid residues under study are boxed. Gly-526 and Leu-384 (both in magenta) make close contact to C8 through C11 of the substrate. For better identification, the hydrogen atoms of Gly-526 are depicted as balls. Mutations of both residues partially or completely arrested PG synthesis. Mutations of Leu-352, Trp-387, and Phe-518 (all in yellow) did not interfere with PG synthesis but some had an influence on the ratio of PG to HETE products. Tyr-385 (red) is the residue that initiates COX catalysis by abstraction of the pro-S hydrogen from C13 (gray) of arachidonic acid. Ser-530 and Val-349 (both in blue) shown in the back are the determinants of the chirality of oxygen addition at C15 of PGG 2 (15). B, crystal structure of mouse COX-2 with PGG 2 modeled into the active site (PDB accession code 1DD0) (12). The cyclopentane ring (C8-C12) of the PGG 2 product carrying the 9,11-endoperoxide (red) occupies a small groove above Gly-526 with Tyr-385 and Trp-387 forming the top. In this model it appears that Leu-384 is shifted away in order to allow access of the product to the groove. C8 oxygenation, a reaction that is not observed in the cyclooxygenase active site.
Concluding Remarks-As we have discussed, the stereochemistry of the novel epoxyalcohols provides strong evidence of their synthesis under enzymatic control within the active site of the cyclooxygenase. It is remarkable that their synthesis is initiated with perfect control of oxygenation at C11, the reaction then diverges from the normal pathway, and then it returns "on course" to produce pure 13R or 15R oxygenated product at the final step. Any oxygenation of substrate that occurs at C11 invariably has the 11R stereochemistry, whereas the second reaction with O 2 can be in the 15S, 13R, or 15R configurations. The latter can be varied by mutation of the enzyme, either at Val-349 and Ser-530 (15), or at Leu-384 and Gly-526, the two residues highlighted here. Because Gly-526 and Leu-384 are remote from C13 and C15 it seems unlikely that the participation of molecular oxygen is influenced directly. This is in line with previous suggestions on the mechanisms of 15R versus 15S stereocontrol in COX-2. To explain the inversion of oxygenation stereochemistry by the acetylated Ser-530 in the COX-2 active site giving 15R-HETE as product, it was proposed that the increase in size of the residue triggers a realignment of the arachidonic acid -chain (8,47,48). It seems reasonable to propose a similar twisting over of the side chain as the basis for the 15R stereochemistry of diepoxyalcohol product 1. The appearance of the prominent 13R oxygenations in products 2 and 3 implies, in addition, the efficient trapping of an allyl radical at the 13-carbon, normally an exceedingly minor event in COX catalysis.