The Binding of Arachidonic Acid in the Cyclooxygenase Active Site of Mouse Prostaglandin Endoperoxide Synthase-2 (COX-2)

The chemical mandates for arachidonic acid conversion to prostaglandin G2 within the cyclooxygenase (COX) active site predict that the substrate will orient in a kinked or l-shaped conformation. Molecular modeling of arachidonic acid in sheep COX-1 confirms that this L-shaped conformation is possible, with the carboxylate moiety binding to Arg-120 and the ω-end positioned above Ser-530 in a region termed the top channel. Mutations of Gly-533 to valine or leucine in the top channel of mCOX-2 abolished the conversion of arachidonic acid to prostaglandin G2, presumably because of a steric clash between the ω-end of the substrate and the introduced side chains. A smaller G533A mutant retained partial COX activity. The loss of COX activity with these mutants was not the result of reduced peroxidase activity, because the activity of all mutants was equivalent to the wild-type enzyme and the addition of exogenous peroxide did not restore full COX activity to any of the mutants. However, the Gly-533 mutants were able to oxidize the carbon 18 fatty acid substrates linolenic acid and stearidonic acid, which contain an allylic carbon at the ω-5 position. In contrast, linoleic acid, which is like arachidonic acid in that its most ω-proximal allylic carbon is at the ω-8 position, was not oxidized by the Gly-533 mutants. Finally, the ability of Gly-533 mutants to efficiently process ω-5 allylic substrates suggests that the top channel does not serve as a product exit route indicating that oxygenated substrate diffuses from the cyclooxygenase active site in a membrane proximal direction.

Prostaglandin endoperoxide synthase has two activities that are required for the production of PGH 2 , 1 the essential precursor of prostaglandins, thromboxane, and prostacyclin (1). The first activity catalyzes the oxygenation and cyclization of arachidonic acid within the cyclooxygenase (COX) active site to produce PGG 2 . PGG 2 then exits the cyclooxygenase active site and undergoes a two-electron reduction, yielding PGH 2 at the peroxidase active site (2). Two isoforms of COX exist, COX-1 and COX-2, that effect the same enzymatic reactions (3). They are approximately 60% identical in sequence and are highly homologous in both active site regions. Not surprisingly, their three-dimensional structures are nearly superimposable (4 -6). COX-1 and COX-2 are mediators of numerous physiological and pathological responses, and therefore considerable effort has been devoted to developing selective COX inhibitors. This is especially true of COX-2, which is a significant contributor to inflammation, hyperalgesia, and cancer (7). Structural analysis of COX-inhibitor complexes has provided a detailed understanding of their interaction with the enzyme and insight into the mechanism of isoform selectivity (4 -6, 8).
A less detailed picture is available for the interaction of the substrate arachidonic acid with the proteins. Crystal structures of COX-arachidonic acid complexes have not been reported, and so most of the available information has been developed by employing site-directed mutagenesis. It is generally agreed that Arg-120 2 ion pairs or hydrogen bonds to the carboxylate of the fatty acid (9 -12) and that Tyr-385 removes the 13-pro-S-hydrogen in the first step of oxygenation (13,14). However, the orientation of the rest of the substrate molecule, particularly the -end, is uncertain.
We have approached the problem of arachidonic acid-COX interaction by attempting to match the chemical mandates of the cyclooxygenase reaction to complementary regions on the protein. COX catalyzes the conversion of an achiral molecule into a product with five chiral centers (15). Noteworthy is the generation of the endoperoxide ring with the trans-substituted alkyl side chains. This stereochemistry is opposite that observed in the auto-oxidation of arachidonic acid in solution in which the alkyl side chains are oriented cis (16). Thus, the enzyme must orientate the arachidonic acid molecule in the COX active site to facilitate the formation of the trans-substituted ring. Any model for COX-arachidonate binding must accommodate this stereochemical mandate. Some time ago it was predicted that the enzyme holds arachidonate in a kinked or L-shaped conformation to facilitate cyclization to form a trans-disubstituted dioxabicycloheptane ring (17) (Fig. 1). Therefore, we modeled such a conformation into the COX-1 active site with the carboxylate adjacent to Arg-120 and the 13-pro-S-hydrogen adjacent to Tyr-385; this starting structure was then energy-minimized. The minimized conformation positioned the -end of arachidonate above Ser-530 projecting into a region that we term the top channel. The importance of the top channel was tested using site-directed mutagenesis of murine COX-2, which can be expressed at high levels in insect cells and purified to homogeneity. The results of the mutagenesis experiments are strongly supportive of a role for the top channel in binding the -end of arachidonic acid and are consistent with the predictions of the energy minimized model.
Modeling-All modeling of arachidonic acid and sheep COX-1 was performed using InsightII (Biosyms Technologies, San Diego, CA) in the manner described previously (18). Arachidonic acid was built using the Builder module and then positioned within the cyclooxygenase active site with the carboxylate group in close proximity to Arg-120 and Tyr-355. The arachidonate main chain was then oriented upward toward the apex of the channel, with carbon 13 placed next to Tyr-385. The -end of arachidonate was then placed above Ser-530, protruding into the top channel toward Asn-375. Using the Discover module, several rounds of energy minimization were performed, maintaining the sheep COX-1 structure as a fixed entity and allowing arachidonate to freely rotate into its most energetically favorable orientation. Several rounds of energy minimization were necessary to obtain a structure that was energetically favorable and consistent with the predicted stereochemical requirements of PGG 2 production, i.e. maintenance of all cis double bonds and positioning of the 13-pro-S-hydrogen in close proximity to Tyr-385.
Mutant Construction-Site-directed mutagenesis was performed on a mCOX-2 pBS(ϩ) vector (Stratagene, La Jolla, CA) using the Quick Change site-directed mutagenesis kit (Stratagene). Mutant containing regions were subcloned into the mCOX-2 pVL1393 baculovirus expression vector (PharMingen, San Diego, CA) using the StuI restriction site in mCOX-2 and the XbaI restriction site present in both the pBS(ϩ) and pVL1393 vectors. The subcloned region was fully sequenced to ensure that no accidental mutations were incorporated.
Protein Expression and Purification-Wild-type and mutant protein was expressed by homologous recombination of the mCOX-2⅐pVL1393 vector with the Baculogold vector (PharMingen) in SF-9 cells (Novagen, Madison, WI). After virus amplification, 4 liters of SF-9 cells (95-100% viable) were grown in TNM-FH medium supplemented with 10% fetal bovine serum (HyClone, Logan, UT), 1% L-glutamine, and 0.1% (v/v) pluronic F68 and then infected with fresh viral stock. Upon reaching 65-70% viability, the 4-liter total volume was harvested by centrifugation at 2500 rpm in a Sorvall RC-3B, and the pellet was washed in ice-cold phosphate-buffered saline and recentrifuged. The final cell pellet was stored at Ϫ70°C.
Purification of wild-type and mutant COXs were performed at 4°C in a manner similar to that described previously (19). Frozen cells were resuspended to 30 ϫ 10 6 cells/ml in 80 mM Tris-HCl, 2 mM EDTA, 0.5 mM phenylmethylsulfonyl fluoride, and 0.1 mM diethyldithiocarbamic acid, pH 7.2. After centrifugation at 100,000 ϫ g for 45 min, the pellet was resuspended using a Dounce homogenizer to a final volume of 72 ml. Solubilization of the COX protein from the membrane was initiated by the dropwise addition of 8 ml of 11% (w/v) CHAPS. After stirring for 1 h, the sample was recentrifuged as described above and the supernatant removed and then diluted 4-fold by the addition of 20 mM Tris-HCl, 0.4% CHAPS, 0.1 mM EDTA, and 0.1 mM diethyldithiocarbamic acid, pH 8.0 (Buffer B). The diluted sample was then loaded onto a 25-ml Macro-prep High-Q ion exchange column equilibrated with Buffer B. COX enzyme was eluted with a linear gradient (500 ml) of increasing KCl to 0.3 M. An analytical 7.5% SDS-polyacrylamide gel electrophoresis was run of candidate COX-containing fractions to determine the fractions to pool for the gel filtration procedure. Appropriate tubes were concentrated in an Amicon concentrator (Amicon, Beverly, MA) to a final volume of 1.5 ml. The sample was then loaded onto a 90-ml Superose-12 column that was pre-equilibrated with 20 mM Tris-HCl, 0.4% CHAPS, 0.15 M NaCl, pH 8.0. Fractions containing COX enzyme, as determined from SDS-polyacrylamide gel electrophoresis analysis (as described above), were concentrated to approximately 2 mg/ml and stored at Ϫ70°C. The purity of wild-type and mutant COX-2 proteins was evaluated by analysis of a Coomassie-stained 7.5% SDS-polyacrylamide gel using an E-C Apparatus Model EC910 scanning densitometer.
Cyclooxygenase Activity-Oxygen consumption was measured with a Gilson Model 5/6 oxygraph (Gilson Medical Electronics, Inc., Middletown, WI) fitted with a Clark electrode and a thermostatted cuvette set to 37°C in a 100 mM Tris-HCl, 500 M phenol, pH 8.0, buffer. The rate and magnitude of oxygen uptake were determined in the manner previously described (20).
Peroxidase Assays-The peroxidase activity of all purified proteins was measured using the guaiacol peroxidase assay. Purified wild-type mCOX-2 or mCOX-2 mutants were suspended in 1 ml of 118 mM Tris-HCl, pH 8.0, at a final concentration of 100 nM. Guaiacol at 500 M was added, and the dual cuvettes were auto-zeroed in a Shidmadzu UV 160U spectrophotometer before the addition of 400 M H 2 O 2 to the same cuvette. Oxidation of guaiacol was monitored at 436 nm (E 436 ϭ 6390 M Ϫ1 cm Ϫ1 ), and the initial rate was measured from the linear portion of the curve.
Cyclooxygenase Product Assays-Cyclooxygenase product assays were performed with purified protein reconstituted with 2 eq of hematin. Reactions were initiated by the addition of 100 M [ 14 C]arachidonic acid. Conditions such as the concentration of phenol, supplementation with H 2 O 2 , and time were varied and are indicated in the figure captions. All assays were terminated and analyzed by thin layer chromatography in the manner described previously (21).

RESULTS
Modeling-Arachidonate was anchored at the carboxyl end through interactions with Arg-120 and Tyr-355 and by positioning the 13-pro-S-hydrogen adjacent to Tyr-385. The -end was placed into the region we term the top channel. Several rounds of energy minimization and refinement were carried out to optimally position the substrate in the sheep COX-1 active site. In the final model (Fig. 2), the carboxylic acid moieties of arachidonic acid are 3.0 and 3.1 Å from Arg-120 and Tyr-355, respectively, and the 13-pro-S-hydrogen is 2.4 Å from the hydroxyl of Tyr-385. The -end of arachidonate protrudes into the top channel region, and although the opening to this channel is narrow, there is sufficient room for carbons 17-20 to reside above Ser-530 and Leu-534 and project into the solvent-accessible cavity toward Gly-533. Carbon 20 resides 3.3 Å from the ␣ carbon of Gly-533.
Mutant COX-2 Characterization-To test this hypothesis, we constructed a series of site-directed mutants at Gly-533. This residue is conserved in all COX sequences, and we anticipated that increasing the steric bulk at this position would reduce arachidonate binding. Both wild-type and mutant COX-2 cDNAs were expressed in insect cells from baculovirus vectors, and recombinant proteins were purified by ion exchange and gel filtration chromatography as described under "Experimental Procedures." All of the purified proteins were shown by densitometric scanning of a 7.5% SDS-polyacrylamide gel to be equal to or greater than 80% pure (Table I). Table I lists both the peroxidase activity and the cyclooxygenase activity of the three Gly-533 mutants. The similarities of the peroxidase activities to that of wild-type COX-2 demonstrate that the mutations did not introduce gross structural perturbations. However, the cyclooxygenase activities of all three mutants were affected significantly. The G533L and G533V mutants were unable to convert arachidonate to PGG 2 , whereas the G533A mutant had a much slower initial rate for substrate conversion and demonstrated just 26% of wild-type total COX-2 activity.
One possibility that could explain the low turnover rate observed with G533A could be a decreased activation of cyclooxygenase catalysis. PGG 2 released from the cyclooxygenase active site is converted to PGH 2 at the peroxidase active site with concomitant formation of a heme-oxo complex (22). This complex oxidizes Tyr-385 and generates the catalytically active tyrosyl radical. As the G533A substitution compromises conversion of arachidonate to PGG 2 , this lower activity could be further pronounced because of less PGG 2 available to maximally activate all the COX molecules in solution. To address the concern of insufficient peroxide activation, three different approaches were employed: 1) lowering the concentration of reducing substrate (phenol) in the assay buffer to slow down the reduction of exogenous peroxides and cyclooxygenase-synthesized fatty acid hydroperoxides; (2) adding H 2 O 2 to generate higher oxidation states of the peroxidase; and (3) adding enzyme-synthesized fatty acid hydroperoxide as an activator.
Reducing the phenol concentration from 500 to 100 M resulted in an increase in cyclooxygenase activity of the G533A mutant from 5.5 to 18.9% converted substrate (Fig. 3A). Likewise, the addition of 15 M H 2 O 2 increased the conversion of arachidonate from 3.4 to 22% (Fig. 3B). However, changing these conditions did not restore cyclooxygenase activity to the level observed with an equivalent amount of wild-type mCOX-2 nor did it restore any activity to either G533V or G533L. To evaluate whether PGG 2 could enhance G533A turnover, G533A mCOX-2 was incubated with a 13-fold lower concentration of Mn-PPIX reconstituted ovine COX-1. Mn-PPIX reconstituted ovine COX-1 has full cyclooxygenase activity but only 0.8% of the peroxidase activity of Fe-PPIX reconstituted COX-1 (23,24). Thus, G533A mCOX-2 should be able to use the COX-1derived PGG 2 to increase the rate of tyrosyl radical formation and cyclooxygenase activity. After subtracting the cyclooxygenase activity resulting from COX-1 turnover, it was found that the G533A activity was equivalent to that observed when G533A was incubated with 15 M H 2 O 2 as an activator. 3 These results suggest that the reduced cyclooxygenase activity of the G533A mutant leads to a slower rate of auto-activation. However, even when maximally activated, the cyclooxygenase activity of this mutant is reduced by 80% compared with the wild-type enzyme, and the G533V and G533L mutants are completely inactive.
Analysis of the arachidonate/sheep COX-1 model suggests that the most likely reason for the reduced activity is steric hindrance between arachidonate carbon 20 and the introduced side chains at position 533. Therefore, the Gly-533 mutants were compared with wild-type enzyme for their ability to metabolize two fatty acids that are substrates for COX-2 but contain an abstractable hydrogen closer to the -end (-5 po-3 B. C. Crews, unpublished observation.

TABLE I Characterization of Gly-533 mutant cyclooxygenases
The purity of wild-type and mutant COX-2 proteins was determined from Coomassie-stained 7.5% SDS-polyacrylamide gels. Peroxidase activities were analyzed using the guaiacol peroxidase assay, and the cyclooxygenase activity initial rates and total product production were determined in the manner described under "Experimental Procedures." Values are the average of three determinations Ϯ S.E. wt, wild type.  (Table II). All three Gly-533 mutants were able to oxidize linolenic and stearidonic acid. In fact, G533A was more efficient than wild-type at converting linolenic acid, and both G533A and G533V metabolized more linolenic acid on a molar basis than did the wild-type enzyme. With stearidonic acid, the rate of substrate conversion for G533A was similar to wild-type mCOX-2, but it decreased upon increasing the chain length at position 533 to valine and leucine. As seen with linolenic acid, the total turnover was greater with G533A and G533V than with wild-type. As a control, incubations were performed with linoleic acid, which contains two fewer carbons at the carboxylate end of the substrate but the same number of carbons as arachidonate between its abstractable hydrogen at its -end.
None of the Gly-533 mutants oxidized linoleic acid.
To establish whether exogenously added peroxide was necessary to obtain maximal cyclooxygenase activity with linolenic and stearidonic acid, 4 assays were performed in the absence of H 2 O 2 but were then supplemented with H 2 O 2 after 2 min to complete the reaction (Tables III). Comparison of wild-type mCOX-2 treated with or without 15 M H 2 O 2 (Tables II and III) showed that the addition of H 2 O 2 had little effect on either the rate or the total substrate conversion. The Gly-533 mutants, in contrast, showed much more dependence on H 2 O 2 , as both the rate and amount of substrate converted were reduced in its absence. The addition of 15 M H 2 O 2 to the reaction was able to reactivate substrate conversion for the Gly-533 mutants. DISCUSSION We propose a model for the binding of arachidonic acid in the cyclooxygenase active site that is consistent with the chemical mandates for its oxidation to PGG 2 and with existing information on its interaction with individual protein residues. The carboxylate of arachidonate is anchored to Arg-120 and Tyr-355 by ionic and hydrogen bonding interactions, respectively (Fig. 2). The substrate backbone then projects upward into the apex of the cyclooxygenase active site where it bends around the 9,10-single bond into an L-shape. Carbon 13 is positioned under Tyr-385 with the 13-pro-S-hydrogen 2.4 Å from the phenolic hydroxyl group. Finally, the -end of arachidonate extends above Ser-530 and Leu-534 into the top channel. Maintenance of this L-shaped conformation is essential for generating the dioxobicycloheptane ring of PGG 2 with the pendant alkyl chains oriented trans to each other. Considering the chemical steps necessary to generate PGG 2 from the perspective of this structural model suggests that the individual reactions can occur with minimal motion of the bound intermediates. Furthermore, the model predicts the generation of PGG 2 with all five stereocenters in the correct absolute configuration, because once the arachidonate is bound as indicated in Fig. 2, the only available space through which O 2 can approach the radical intermediates is through the center of the active site channel. Thus, O 2 approaches the bound fatty acid from the opposite side from which the 13-pro-S-hydrogen is removed. This antarafacial relationship is consistent with the stereochemistry of PGG 2 .
The key feature of our model is that the -end of arachidonate projects into an area of the protein we term the top channel, which is located above Ser-530 and Leu-534. There are two major pieces of experimental information consistent with this hypothesis. The first is that introduction of steric bulk at position 533 by site-directed mutagenesis significantly reduces or completely abolishes the ability of the mutant COX-2s to oxidize arachidonic acid. The loss of oxygenase activity is not due to a major structural change in the proteins, because the peroxidase activity of each of the position 533 mutants is identical to that of wild-type enzyme. Likewise, the loss of activity is not due to an inability of the mutant proteins to activate the cyclooxygenase activity, although activity is stimulated to some extent by addition of H 2 O 2 . The second piece of experimental information consistent with a role for the top channel in arachidonate binding is that all of the position 533 mutants are able to oxidize unsaturated fatty acid substrates that contain three less carbons at their -end than arachidonate relative to the position of the hydrogen abstracted by the tyrosyl radical of Tyr-385. By contrast, the mutants are unable to oxidize linoleic acid, which contains two fewer carbons on the carboxyl end than arachidonate but the same distance between its abstractable hydrogen and its -end.
The ability of the Gly-533 mutants to oxidize linolenic and stearidonic acids but not arachidonic or linoleic acids establishes that the distance between the hydrogen abstracted by the tyrosyl radical of Tyr-385 and the -end of the fatty acid is an important determinant of COX-fatty acid interactions. Positioning of fatty acids in the top channel may be as important for determining substrate specificity as the positioning of the carboxylate adjacent to Arg-120 at the mouth of the channel. There appears to be flexibility in the distance from the hydrogen abstracted by Tyr-385 to the carboxylate of the fatty acid, especially in the case of COX-2. Substrates with an abstractable hydrogen at the 11, 13, or 14 carbon from the carboxylate are oxidized by COX-2. Laneuville et al. (25) have proposed that the substrate adapts to the accessible space between Tyr-385 and Arg-120 by adopting a linear conformation (11-carbon length) or a kinked conformation (13-or 14carbon lengths). The extra space in the side pocket off the main channel of COX-2 relative to COX-1 may permit the kinking required to accommodate different length substrates.
A corollary of the ability of the position 533 mutants to oxidize linolenic and stearidonic acid is that the top channel does not represent part of an exit route to the top surface of the protein. An examination of the crystal structures of both COX-1 and COX-2 indicates that the channel around Gly-533 connects to another channel that eventually leads to an opening on the surface of the protein near the dimer interface (6). Conceivably, this opening could represent an exit port through which product escapes the cyclooxygenase active site. Indeed, this could be an explanation for the inability of sterically blocked mutants (e.g. G533V and G533L) to oxidize arachidonic acid. The first molecules of PGG 2 synthesized from arachidonic acid would be unable to exit and would prevent the binding of additional molecules of arachidonic acid into the cyclooxygenase active site. However, blocking the exit channel should also prevent the release of the products of linolenic acid and stearidonic acid, thereby stopping turnover. The high activity of the Gly-533 mutants toward these carbon 18 fatty acids eliminates the possibility that the top channel is part of an exit route for products.
Our model is consistent with the chemistry of the production of PGG 2 from arachidonic acid and with all the currently available site-directed mutagenesis results. Thus, it is likely that this is the conformation by which COX enzymes convert arachidonic acid to its major enzymatic product. This conclusion does not rule out the possibility that arachidonate binds in the cyclooxygenase active site in alternate conformations. For example, arachidonate is oxygenated to a series of hydroxy acids by cyclooxygenase including 11-(S)-HETE, 15-(S)-HETE, and 15-(R)-HETE (15,26). The production of 11-(S)-HETE and 15-(S)-HETE have been assumed to result from O 2 trapping of carbon radicals produced following removal of the 13-pro-Shydrogen. However, 15-(R)-HETE is produced by aspirin-acetylated COX-2 and is likely to arise from an alternate conformation of arachidonic acid (26). Indeed, Xiao et al. (27) have recently shown that 15-(R)-HETE also can be made by unacetylated COX-2 at high concentrations of arachidonic acid. Thus, 15-(R)-HETE appears to represent an alternate product that results from a less favored substrate conformation.