Fatty Acid Binding to the Allosteric Subunit of Cyclooxygenase-2 Relieves a Tonic Inhibition of the Catalytic Subunit*

Prostaglandin endoperoxide H synthase-2 (PGHS-2), also called cyclooxygenase-2 (COX-2), converts arachidonic acid to PGH2. PGHS-2 is a conformational heterodimer composed of allosteric (Eallo) and catalytic (Ecat) subunits. Fatty acids (FAs) bind to Arg-120 of Eallo increasing to different degrees, depending on the FA, the Vmax of its Ecat partner. We report here that movement of helical residues 120–122 and loop residues 123–129 of Eallo underlies the allosteric effects of FAs and allosteric COX-2 inhibitors, including naproxen and flurbiprofen. An S121P substitution in both PGHS-2 monomers yields a variant (S121P/S121P PGHS-2) that has 1.7–1.8 times the Vmax of native PGHS-2 and is relatively insensitive to activation by FAs or inhibition by allosteric inhibitors. The S121P substitution in Eallo is primarily responsible for these effects. In X-ray crystal structures, the Cα atoms of helical residues 119–122 of S121P/S121P PGHS-2 are displaced from their normal positions. Additionally, the S121P/S121P PGHS-2 variants in which Pro-127 and Ser-541 are replaced by cysteines spontaneously forms Cys-127 to Cys-541 cross-links between monomers. This is unlike the corresponding native PGHS-2 variant and suggests that S121P substitutions also unhinge the loop involving residues 123–129. We conclude the following: (a) the region involving residues 120–129 of unoccupied Eallo tonically inhibits Ecat; (b) binding of an activating FA (e.g. arachidonic, palmitic, or oleic acid) to Eallo or an S121P substitution in Eallo repositions this region to increase Ecat activity; and (c) allosteric COX inhibitors act by preventing FA binding to Eallo and additionally by relocating Eallo residues to inhibit Ecat.

Prostaglandin endoperoxide H synthases (PGHSs), 4 also known generically as cyclooxygenases (COXs), convert arachidonic acid (AA), two O 2 molecules and two electrons from an unknown reductant(s) to prostaglandin H 2 (PGH 2 ) (1, 2). The formation of PGH 2 from AA involves two enzymatic steps. First, the COX activity of PGHSs catalyzes the introduction of two O 2 molecules into the AA backbone forming PGG 2 , a bicyclic compound with an endoperoxide group bridging C-9 to C-11 and having a hydroperoxyl group at C-15. This latter group is reduced to an alcohol by the peroxidase (POX) activity of PGHSs in forming PGH 2 . PGH 2 is the precursor for enzymes that produce what are considered to be the most important biologically active prostanoids, which include PGD 2 , PGE 2 , PGF 2␣ , PGI 2 (prostacyclin), and thromboxane A 2 (2). There is a functional interplay between the COX and POX sites in that the heme group at the POX site needs to be oxidized by a peroxide to oxidize Tyr-385 in the COX site to a tyrosyl radical. Although apparently not rate-limiting (3), the Tyr-385 radical abstracts the 13-pro-S-hydrogen from AA to initiate the COX reaction (1).
PGHSs are homodimers with subunit molecular masses of about 72 kDa (2). Previous studies have established that although PGHSs are sequence homodimers, they function as conformational heterodimers composed of a regulatory allosteric subunit designated E allo and a catalytic subunit denoted E cat (4 -12). Compared with E allo , E cat binds heme with high affinity, and maximal activity occurs with one heme/dimer (13)(14)(15). The COX-binding sites of E allo and E cat exhibit different binding and functional properties. For example, with both PGHS isoforms, AA binds with higher affinity to E allo than E cat , but AA is only oxygenated by E cat (11,14,15). In the case of PGHS-2, numerous fatty acids (FAs), both substrate and nonsubstrate (ns) FAs, and some nonsteroidal anti-inflammatory drugs (NSAIDs), including ibuprofen (IBP), naproxen, and flurbiprofen (FBP), also bind with higher affinity to E allo than E cat and allosterically regulate E cat activity (5)(6)(7)(8)(9)(10)(11)(12)14). Some FAs allosterically stimulate COX activity, whereas IBP, naproxen, and FBP allosterically inhibit activity.
Binding of allosteric ligands involves the interaction of the carboxylate groups of the ligands with the arginino group of Arg-120 of E allo (9,11). Arg-120 is two residues in from the partially deformed C terminus of helix D of huPGHS-2 (2,16). The N-terminal end of this helix D forms part of the membrane binding domain of PGHS-2 (17)(18)(19). Arg-120 of E cat is also important for -3 and -6 polyunsaturated FA substrate binding to PGHSs (9, 20 -27). This residue lies at the opening of an L-shaped hydrophobic core of the COX active site of E cat (27).
Cross-linking studies with huPGHS-2 (5) and structural studies with ovine PGHS-1 (7,14,28) have suggested that ligand binding to Arg-120 of E allo elicits changes in the position of a flexible loop immediately downstream of Arg-120 involving residues 123-129 (Fig. 1). To simplify our presentation, we designate the loop that contains residues 123-129 as the allosteric loop. Repositioning of the allosteric loop upon ligand binding changes its interaction with a region in the partner E cat monomer involving residues 541-543 of E cat . Thus, substitution of residues 126 and 543 or 127 and 541 with cysteines leads to oxidant-induced cross-linking between monomers comprising a dimer. Importantly, the cross-linking can be blocked by ligands that can bind simultaneously to E allo and E cat (5,11). This suggests that simultaneous binding of E allo and E cat ligands causes a relative repositioning of the allosteric loop.
Although it is clear from studies of huPGHSs in solution that E allo and E cat must be structurally different, X-ray crystallographic studies performed to date have not clearly distinguished E allo from E cat . PGHSs tend to crystallize as symmetric monomers even under conditions in which the ratios of protein monomers to ligand are less than one (28). In a few cases involving a relative excess of FA ligands, FA substrates are found in different orientations in the two subunits of murine (mu) PGHS-2; however, there are no differences in the structures of the bound subunits (27). Additionally, with muPGHS-2, palmitic acid (PA) binds only one of the two subunits, presumably E allo , but there is no difference between the structures of the occupied and unoccupied monomers (14). With ovine PGHS-1, the loop involving residues 121-129 was determined to be in two orientations in a structure showing partial occupancy of one subunit by celecoxib (7) and FBP (Fig. 1C) (28). The largest differences in positions between the two orientations of the loops involved residues 123-126. It was for these many reasons that we investigated residues within, adjoining, and interacting with this allosteric loop in huPGHS-2 to learn more about the how E allo regulates E cat . Our enzymologic studies were performed with huPGHS-2, whereas muPGHS-2 was used for the X-ray crystallography. Previous enzymologic studies have been performed with both muPGHS-2 and huPGHS-2, and the results are qualitatively similar (10 -12). Until very recently, X-ray crystallographic structures have been reported in the PDB only with muPGHS-2 (29,30). Crystallographic studies that post-date the structure determinations reported here indicate that the murine and human crystal structures are virtually identical (31).

Effects of Substitutions of Residues 120 -127 and Neighboring
Residues of huPGHS-2-In a first set of experiments, we examined the effects of mutations in the huPGHS-2 dimer (a) near the C terminus of helix D, (b) in the loop downstream of helix D, (c) of Arg-44 which forms a hydrogen bond with Asp-125 of the loop, and (d) of Ser-541 and Ala-543 that neighbor this allosteric loop ( Fig. 1  C. FIGURE 1. Position, sequence, and structure of the allosteric loop of PGHSs. A, schematic of the PGHS-2 dimer. The membrane binding domain of each monomer is highlighted in blue, with helix D labeled and the membrane bilayer denoted by red lines. The side chain residues located at the opening of the cyclooxygenase channel in each monomer are shown as sticks and labeled accordingly. The allosteric loop is shown in orange, with the location of Thr-129 depicted as a red sphere (for clarity). Similarly, the location of residues Ser-541 and Ala-543 at the dimer interface are depicted as red spheres. AA (pink) is shown bound to E cat . B, comparison of amino acid sequences of a portion of helix D that extends to residue 122, and the allosteric loop that extends from residue 123 to 129 of PGHS-1 and PGHS-2 from different species (2). C, stereoview reproduced from Ref. 28 with permission of the allosteric loop of a flurbiprofen/ovine R120Q/Native PGHS-1 co-crystal structure showing alternative conformations of the allosteric loop with blue and orange sticks representing the "up" and "down" configurations of amino acids (PDB code 3N8W).

TABLE 1
Properties of huPGHS-2 variants having substitutions of residues adjoining Arg-120 or in the allosteric loop immediately downstream of Arg-120 or of residues that interact with this loop huPGHS-2 variants were prepared and purified, and COX activities were assayed as described under "Experimental Procedures." Kinetic values are derived from the average of triplicate determinations that varied by less than Ϯ5% from the average.   duced into both monomers by substituting alanine and/or the amino acid present at the corresponding position of PGHS-1. Note that because the mutations are in both subunits of the huPGHS-2 dimer, the results observed are a composite of effects occurring in both E allo and E cat . As reported previously, substitution of Arg-120 with alanine abrogates the allosteric effects of PA, oleic acid, and 20:19 that are not COX substrates (i.e. non-substrate fatty acids (nsFAs)) ( Table 1) (9). This is because FAs do not efficiently bind to E allo of R120A/R120A huPGHS-2. Moreover, E cat of this mutant binds AA with a 5-10-fold lower affinity than does E cat of native huPGHS-2 (9). S121P/S121P huPGHS-2 has quite different properties than native huPGHS-2, notably an almost 2-fold higher specific COX activity with AA (Table 1). S121P/S121P huPGHS-2 also exhibited substantially diminished allosteric responses to PA, oleic acid, and 20:19. PA caused a modest 20% allosteric activation of S121P/S121P huPGHS-2 compared with a 75% activation of native huPGHS-2 under the same conditions. Other amino acid substitutions intended to destabilize the N terminus of helix D either had no effect on COX activity (i.e. S121G) or significantly decreased COX activity (i.e. H122P). A relatively neutral H122N substitution had little effect on huPGHS-2.
Substitutions of allosteric loop residues 123, 124, and 125 diminished allosteric responses to PA and oleic acid. Residue 543, which neighbors residue 126 in huPGHS-2 (5), is a glutamate in PGHS-1 and an alanine in PGHS-2 (2). A543E substitutions reduced the response to PA but had no effect on the response to oleic acid and even enhanced the response to 20:19.
The binding of [1-14 C]AA and -PA to E allo was estimated using the method developed previously (14) and described under "Experimental Procedures" that takes advantage of the ability to measure binding at high enzyme to substrate ratios of unreacted [1-14 C]AA to E allo of huPGHS-2. Except with R120A/R120A huPGHS-2, [1-14 C]AA binding to E allo of the mutants tested in Table 1 was comparable with that observed for E allo of native huPGHS-2 (9). Details are provided below. Although not definitive, the results in Table 1 are consistent with the idea that several of the residues tested, except perhaps His-122, contribute to the allosteric regulation of huPGHS-2 by nsFAs and thus, we suspected, in various ways to the positioning of the allosteric loop.
Characterization of S121P Substituted PGHS-2-The most pronounced effect we observed in testing the mutants shown in Table 1 was the heightened activity of the huPGHS-2 variant having S121P substitutions. This led us to characterize this mutation in detail. Initially, we examined S121P/S121P huPGHS-2, which carries the S121P substitution in both subunits. We then analyzed mutants having this substitution individually in E allo or E cat . Fig. 2 and Table 2 provide comparisons of the specific activities of native huPGHS-2 versus S121P/S121P huPGHS-2 with a variety of polyunsaturated FA substrates. S121P/S121P huPGHS-2 had markedly increased activity with AA and somewhat increased activity with dihomo-␥-linolenic acid. There were no appreciable effects caused by the S121P substitution on the K m values for these two 6 FA substrates. In contrast, this mutant enzyme had somewhat reduced activity toward a variety of other substrates tested, including docosahexaenoic acid, linoleic acid, and 11,14-eicosadienoic acid; in the case of 11,14eicosadienoic acid, there is a significant increase in the K m value. None of these latter three substrates efficiently bind to E allo of native huPGHS-2 (11). S121P/S121P huPGHS-2 had essentially the same activity as native huPGHS-2 with 5,8,11,14,17-cis-eicosapentaenoic acid (EPA). EPA is an unusual PGHS-2 substrate. Although it does bind to E allo of native huPGHS-2 (11), binding of EPA to E allo of native huPGHS-2 appears to suppress its own oxygenation. 5 When incubated with [1-14 C]AA, S121P/S121P huPGHS-2 formed about the same proportion of PGH 2 -derived products as native huPGHS-2 (i.e. 89% versus 84% for native and S121P/S121P huPGHS-2, respectively; data not shown). Thus, the S121P substitution does not cause a change in the PG product profile. When titrated with heme, S121P/S121P huPGHS-2 exhibited one high affinity heme-binding site with a K d ϭ 0.36 Ϯ 0.07 M (data not shown). This value is about twice that originally determined for huPGHS-2, but the same as that we previously determined for Y385F/Native huPGHS-2 (9).
A K d ϭ 0.44 M was determined for [1-14 C]AA binding to S121P/S121P huPGHS-2, which represents binding to the E allo subunit (Table 3). Also, as indicated in Table 3, bound [1-14 C]AA was effectively displaced by 5 M PA. The data 5 COX activity toward EPA is the same with native huPGHS-2 and Y385F/Native huPGHS-2 (9) but is increased about 2-fold with R120A Y385F/Native huPGHS-2 (L. Dong and W. L. Smith, unpublished data). E allo of the latter variant does not bind EPA (11) suggesting that EPA binding to E allo suppresses COX activity toward EPA. Oxygenation of fatty acid substrates by native huPGHS-2 and S121P/S121P huPGHS-2. Recombinant enzymes were prepared, and measurements of COX activities using an O 2 electrode were performed with 100 M concentrations of the indicated fatty acids as noted under "Experimental Procedures." The error bars indicate the average Ϯ S.E. A significant difference from the value with the Native/Native (versus S121P/S121P) huPGHS-2 as determined by the Student's t test (p Ͻ 0.05) is indicated with an asterisk. The experiment was performed with two different preparations of enzymes with the same results.
establish that although the magnitude of allosteric activation of this variant by PA was less than a third of that seen with native huPGHS-2, E allo of S121P/S121P huPGHS-2 binds AA and PA with similar affinities as native enzyme (Table 4).
Structural Studies of Murine and Human S121P/S121P PGHS-2-As noted in the Introduction, Ser-121 is located near the end of helix D of the membrane-binding domain, adjacent to Arg-120 at the opening of the COX channel. Arg-120, along with Tyr-355 and Glu-524 collectively form a water-mediated interaction network at the channel entrance to effectively bury substrates and inhibitors within the catalytic domain (17,18,22,27,32). As these interactions decrease the aperture of the channel entrance when ligand is bound, the conformation of COX in this state is generally referred to as the "closed" conformation (33). In addition to forming the constriction, these residues participate in the binding of acidic time-dependent inhibitors and substrates (18, 27, 32, 34 -39).
We determined the crystal structure of S121P/S121P muPGHS-2 in the absence of ligands (S121P/apo) bound in the COX channel using synchrotron radiation (Table 4 and Fig. 3). S121P/apo crystallized with two monomers in the asymmetric unit, with complete conservation of the global domain architecture observed in other crystal structures of muPGHS-2 (14,27,29). The root mean square deviation calculated between monomers is less than 0.22 Å, smaller than the mean positional error and indicating that there are no significant structural deviations between monomers. Closer inspection of the COX channel opening in the S121P/apo-crystal structure reveals that the S121P mutation disrupts the last helical turn of helix D, causing it to unwind (Fig. 3A). As a result, the side chain of Arg-120 is displaced and becomes disordered, as evidenced by the lack of electron density observed beyond the C␤ carbon of the side chain. The 3 Å positional displacement of the main chain in this area of helix D (Fig. 4), in comparison with its position in native enzyme crystal structures complexed with substrates and inhibitors, results in the breakage of the ionic interaction between the side chains of Arg-120 and Glu-524 (Fig. 3). The loss of the Arg-120/Glu-524 ionic interaction Kinetic properties of native and S121P/S121P huPGHS-2 with several 6 fatty acids that are COX substrates huPGHS-2 variants were expressed and purified, and COX activities were assayed using an O 2 electrode assay as described under "Experimental Procedures." Kinetic values are derived from the average of triplicate determinations Ϯ S.D. The experiments were performed two different enzyme preparations with comparable results.

TABLE 3
Oxygenation of [1-14 C]AA by S121P/S121P huPGHS-2 variants at high enzyme to AA ratios  results in the opening of the constriction at the base of the COX channel. R120A/R120A and R120Q/R120Q PGHS-2 constructs are refractory to inhibition by carboxylic acid containing NSAIDs (9,21,39). Given the large observed displacement of Arg-120 in the S121P/apo-crystal structure, we hypothesized that NSAIDs requiring an interaction with Arg-120 would not bind to S121P/ S121P muPGHS-2. To test this hypothesis, we employed a thermal shift assay to monitor the binding of FBP, IBP, naproxen, and celecoxib to native, R120A/R120A, and S121P/S121P muPGHS-2 variants (Fig. 5). With the exception of celecoxib, all the drugs form an ionic interaction with Arg-120 to facilitate binding. Melting temperatures (T m ) were calculated for all three constructs in the absence and presence of inhibitor. The change in T m (⌬T m ) was then calculated to provide a measure of inhibitor binding. As expected, all of the drugs bound to native muPGHS-2, inducing ⌬T m shifts of 10 -17°C. Conversely, FBP, IBP, and naproxen failed to induce a significant ⌬T m shift when FIGURE 3. Conformational differences between residues Ser-119 and Asp-125 in the absence and presence of inhibitor. Stereo views that highlight the main chain conformational differences for residues Ser-119 through Asp-125 at the entrance of the cyclooxygenase channel in the crystal structures of S121P/S121P muCOX-2 in the absence and presence of bound inhibitor. A, comparison between the S121P/apo-crystal structure and native muCOX-2 (PDB code 3HS5) depicting the unwinding of the last turn of helix D. Carbon, oxygen, and nitrogen atoms for residues Ser-119 through Asp-125 are colored yellow, red, and blue, respectively. A green ribbon depicts the location of the C␣ positions for the equivalent residues in the wild type muCOX-2 crystal structure. B, comparison between the S121P/apo-and S121P/cbx and C, S121P/apo-and S121P/fbp crystal structures highlighting the fact that the binding of FBP induces a rewinding of the last turn of helix D, whereas the binding of celecoxib does not. For all three panels, F o Ϫ F c simulated annealing omit electron density is shown in blue mesh contoured at 2.5. The side chains for Tyr-355, Arg-513, and Glu-524, which participate in the water-mediated interaction network at the channel entrance and membrane-binding domain helices A, B, and D, are also labeled accordingly.
tested with the R120A/R120A muPGHS-2 construct (Fig. 5). Surprisingly, we observed ⌬T m shifts for all the drugs when tested with the S121P/S121P muPGHS-2 variant, suggesting that inhibitor binding can rescue or reverse the observed displacement of Arg-120 in the S121P/S121P muPGHS-2 mutant.
To follow up on our thermal shift assay observations, we determined the crystal structures of S121P/S121P muPGHS-2 in complex with FBP (S121P-fbp) and celecoxib (S121P-cbx) (Fig. 3). Importantly, NSAID-bound structures were obtained by soaking crystals of the unliganded S121P/S121P muPGHS-2 in cryoprotectant containing 1 mM inhibitor. As with the S121P/apo-crystal structure, the liganded structures exhibited the same domain architecture as other PGHS-2 crystal structures, with no differences between monomers. Celecoxib binds within the cyclooxygenase channel of the S121P/ S121P muPGHS-2 with its phenyl sulfonamide moiety inserted into the COX-2 side pocket, analogous to that observed in the native crystal structure (40). Moreover, celecoxib binding fails to induce a rewinding of the last turn of helix D, which exhibits the same conformation observed in the S121P/apo structure. Consistent with our thermal shift assay results and the crystal structure of the native enzyme, there are no interactions between the inhibitor and the side chain of Arg-120 (Fig. 3B). FBP also binds within the cyclooxygenase channel of the S121P construct in the same conformation as that observed in the native structure, with its carboxylate moiety forming ionic interactions with the side chain of Arg-120 (18,41). Importantly, the binding of FBP within the COX channel of the S121P construct results in the rewinding of the last turn of helix D and ordering of the Arg-120 side chain, thus restoring the closed conformation of the channel opening (Fig. 3C). Although crystal structures of the S121G and H122P mutants were not determined, we assume based on analysis of the huPGHS-2 variants ( Table 4) that neither caused the degree of distortion of helix D that is observed with S121P/S121P muPGHS-2.
The structure of the region involving residues 124 -129 is relatively unaltered in any of the crystal structures; however, crystal contacts present in this region of the protein likely constrain movement in the loop that could occur in solution. Therefore, we performed the cross-linking studies depicted in Fig. 6 to test for movement of this allosteric loop. As discussed next, the results provide evidence that at least part of the allosteric loop involving residues 123-129 is repositioned in S121P/ S121P huPGHS-2 in solution.
We had previously examined a cysteine-less huPGHS-2 variant denoted P127C A541C ⌬C huPGHS-2 in which the free cysteines of huPGHS-2 were replaced with serine residues, and new cysteine residues were introduced in the allosteric loop at residue 127 and in an adjoining loop at residue 541 from the partner subunit (5). Addition of an oxidizing agent, a cuprous⅐ophenanthroline complex, caused a time-dependent cross-linking of the monomers comprising P127C A541C ⌬C huPGHS-2. This is illustrated for comparison purposes in the top panel of Fig. 6 that we previously presented in Ref. 5. The uppermost band is a dimer having two disulfide cross-links; immediately below this is a band having one cross-link, and the lowest band is the monomer (see Ref. 5). The cross-links are between Cys-127 and Cys-541. Cross-linking could be prevented by certain FAs or COX inhibitors that bind both monomers, and we infer that binding of these ligands prevents close apposition of Cys-127 and Cys-541 (5,11). In the experiment depicted in Fig. 6A, formation of cross-linked monomers having one or two disulfide linkages is inhibited by FBP. Note that in the top panel at 0 time (Fig. 6) that about two-thirds of the control P127C A541C ⌬C huPGHS-2 template has become spontaneously crosslinked during expression and purification and prior to the addition of oxidant. We introduced R120Q or S121P substitutions into the P127C A541C ⌬C huPGHS-2 homodimer template to form the homodimeric variants denoted as R120Q P127C A541C ⌬C huPGHS-2 (Fig. 6B) and S121P P127C A541C ⌬C huPGHS-2 (Fig. 6C). Both of these latter proteins underwent spontaneous cross-linking during their preparation. With R120Q P127C A541C ⌬C huPGHS-2, which likely binds FBP only weakly (9), spontaneous cross-linking, although very substantial, was not quite complete. Not surprisingly, FBP was relatively ineffective in preventing oxidant-induced cross-linking.  Wild type, R120A, and S121P muCOX-2 constructs were incubated with the indicated inhibitor on ice for 30 min, followed by determination of the protein denaturation temperature as described under "Experimental Procedures." Results are reported as the change in melting temperature (⌬T m ) between liganded and unliganded states. Melting temperature measurements were made in triplicate. CBX is celecoxib; NPX is naproxen.
In the case of S121P P127C A541C ⌬C huPGHS-2, all the monomers spontaneously cross-linked. We were unable to prevent the complete spontaneous cross-linking even by adding the reversible COX inhibitor IBP to the insect cell culture medium during expression of S121P P127C A541C ⌬C huPGHS-2.
We interpret the results with R120Q P127C A541C ⌬C huPGHS-2 as indicating that the Gln-120-containing mutant cannot bind endogenous ligands efficiently (9) and thus that during its preparation the R120Q P127C A541C ⌬C huPGHS-2 variant is not in a ligand-bound or closed conformation in which cross-linking is prevented. We interpret the results of cross-linking studies with S121P P127C A541C ⌬C huPGHS-2 to indicate that the S121P substitution leads to the allosteric loop, in this case with the P127C replacement, to be more flexible than in the native protein allowing it to neighbor and cross-link to the adjoining loop containing residue Cys-541.
Effects of S121P Substitutions in E allo Versus E cat of huPGHS-2-To determine how the S121P substitution enables the large increase in V max of huPGHS-2, we characterized mutants having alterations of Ser-121 and other amino acids in either E allo or E cat . As illustrated in Fig. 7, substitutions were introduced into a huPGHS-2 heterodimer variant denoted as Y385F/Native huPGHS-2. This latter variant has a catalytically inactive monomer E allo with a Y385F substitution and a cat- C. FIGURE 6. Cross-linking between monomers of P127C S541C ⌬C-huPGHS-2, R120Q P127C S541C ⌬C-huPGHS-2, and S121P P127C S541C ⌬C-huPGHS-2. The mutant proteins indicated in panels A, B, and C are P127C S541C ⌬C-huPGHS-2, R120Q P127C S541C ⌬C-huPGHS-2, and S121P P127C S541C ⌬C-huPGHS-2, respectively. These recombinant proteins were expressed in a baculovirus system, purified, and subjected to cross-linking by treatment with Cu 2ϩ /o-phenanthroline for the indicated times in the presence or absence of 100 M FBP as described under "Experimental Procedures." Samples were subjected to SDS-PAGE in the absence of a disulfide reducing agent (i.e. dithiothreitol), and Western blotting was performed using an anti-PGHS-2 antibody. FIGURE 7. Schematic of heterodimer mutants used to determine the effects of S121P and R120A substitutions in E allo versus E cat of huPGHS-2 conformational heterodimer. Heme has a marked preference for binding the native subunit versus a subunit having a Y385F substitution (9). An intact Tyr-385 and bound heme are necessary for COX activity.
alytically active monomer E cat that is the native subunit (9). We established that 90% of E cat is the native subunit and remains in this form during the life of the protein (9). S121P and other amino acid substitutions were made in either E allo , the subunit with the Y385F substitution, or the native E cat subunit.
Results comparing Y385F/Native huPGHS-2 with S121P mutants in this background are presented in Table 5. At least two-thirds of the increase in the V max of the S121P/S121P huPGHS-2 is attributable to the S121P substitution in E allo ; however, the S121P substitution in E cat also contributes at least in some way to the increase in V max (Table 5). S121P Y385F/ Native huPGHS-2 has 85% more activity than Y385F/Native huPGHS-2. This is an increase almost identical to the relative increase observed in comparing S121P/S121P huPGHS-2 to native huPGHS-2. Y385F/S121P huPGHS-2 also exhibits a 35% increase in activity over Y385F/Native huPGHS-2 suggesting that a S121P mutation in E cat also contributes to the enhanced activity observed when this substitution is in both monomers (i.e. with S121P/S121P huPGHS-2).
As might be expected were the S121P substitution partially replacing the allosteric stimulation of huPGHS-2 by FAs, Y385F S121P/Native huPGHS-2 was less sensitive to allosteric stimulation by PA than was the Y385F/S121P mutant (Table 5). In contrast, the relative increases in activity with PA were the same for the Y385F/Native huPGHS-2 control and Y385F/S121P huPGHS-2 where the S121P substitution is in E cat .
As summarized in Table 5 the magnitudes of allosteric inhibition of Y385F S121P/Native huPGHS-2 by naproxen and FBP were both decreased in comparison with Y385F/Native and Y385F/S121P huPGHS-2. These two NSAIDs function as allosteric inhibitors of huPGHS-2. In contrast, inhibition by diclofenac, a COX inhibitor that functions by competing for E cat (14), was unaffected in a test of S121P Y385F/Native huPGHS-2. Results comparing dose responses of native huPGHS-2 and S121P Y385F/Native huPGHS-2 to these three inhibitors are presented in Fig. 8. Celecoxib inhibits huPGHS-2 by competing directly with AA for binding to E cat . As noted above, celecoxib inhibition does not involve its binding to Arg-120 in E cat . Thus, we were surprised to find that inhibition by celecoxib was attenuated in huPGHS-2 variants having S121P substitutions in E allo (i.e. S121P Y385F/Native and R120A S121P Y385F/Native huPGHS-2). Presumably, Ser-121 in E allo contributes in some way to celecoxib interactions with E cat so as to decrease the inhibitory effect of celecoxib on E cat .
We also compared R120A S121P Y385F/Native to R120A Y385F/Native huPGHS-2 (Table 5). R120A Y385F/Native huPGHS-2 is unable to bind and thus unable to respond to allosteric effectors, including nsFAs and naproxen (9,14). When introduced into this background, a S121P substitution in E allo increases the V max by more than 3-fold from 16 to 54 units/ mg. This is consistent with the idea that the S121P substitution in E allo acts largely by supplanting allosteric responses to AA itself, nsFAs, and allosteric COX inhibitors.
Effects of Substitutions of Asp-125 and Arg-44 on huPGHS-2-Residue 125 is the one residue in the allosteric loop that differs between PGHS-1 and PGHS-2. (i.e. Pro-125 in huPGHS-1 versus Asp-125 in huPGHS-2 (Fig. 1)). We substituted Asp-125 in both E allo and E cat of huPGHS-2 with proline or alanine to determine the effect of altering this isoform-specific residue in the allosteric loop. We made parallel substitutions of Arg-44 with glutamine or alanine because Arg-44 residue appears to form a hydrogen bond with Asp-125 in muPGHS-2 crystal structures (14,29). We examined the effect of the various Asp-125 and Arg-44 mutations on basic kinetic parameters and responses to various FAs and to naproxen. The data are presented in summary form in Table 6. Overall, the results of studies of the D125A and R44A mutants show reductions in V max values with little or no changes in either K m values or in the affinities of AA or PA for E allo . In several cases, mutations in either E allo or E cat substantially reduced the V max values but to levels significantly lower than those seen with the mutations in both E allo and E cat . Comparison of the data obtained with these mutants indicated that both Asp-125 and Arg-44 contribute to the functioning of both E allo and E cat . Thus, we could not make definitive conclusions about the role of these residues in the individual monomers. We did note that with either D125A or R44A substitutions in E allo but not E cat that there were decreases in the allosteric responses to both PA and naproxen. However, the patterns seen with D125A and R44A were different from the ones observed with D125P or R44Q substitutions. We observed that values for basic kinetic parameters with enzymes having D125A substitutions were not mark- Properties of huPGHS-2 variants having S121P substitutions in E allo versus E cat subunits huPGHS-2 variants were prepared, expressed, and assayed for COX activities with AA as substrate as described previously and under "Experimental Procedures" (9). Increases in COX activity due to PA were measured using 5 M AA and 25 M PA. Relative responses to inhibitors were determined after preincubating the huPGHS-2 variants for 30 min with naproxen (0, 1, 2.5, 5, and 25 M), flurbiprofen (0, 1, 2.5, 5.0, and 25 M), diclofenac (0, 1, 2, 3, 4, and 8 M), or celecoxib (0, 0.5, 1, 2, 5 and 15 M). ND means not determined.

Discussion
The rate-limiting step in COX catalysis is incompletely resolved but involves control over the distance between and/or relative orientation of the 13-pro-S-hydrogen of AA and the Tyr-385 radical or in some cases the stability of the Tyr-385 radical in E cat (1,3,31). The E cat subunit of PGHSs provides a readout of the allosteric effects of ligand binding to E allo (10,12). In this study, we have investigated the mechanistic basis for COX-2 allosteric regulation. Most importantly, we have provided new evidence that Arg-120 and the region immediately downstream of Arg-120 in E allo involving residues 121-129 of huPGHS-2 mediates allosteric responses to FAs and allosteric COX inhibitors. Table 5, huPGHS-2 is tonically inhibited when no ligands are bound to E allo . Specifically, the V max of purified recombinant huPGHS-2 without FA bound to Arg-120 of E allo (i.e. with R120A Y385F/Native huPGHS-2) is 16 units/mg with AA as substrate. The V max , again with AA as substrate, is increased to 37 units/mg with AA bound to E allo (i.e. with Y385F/Native huPGHS-2). Thus, when bound to E allo , AA itself allosterically activates E cat toward the parent substrate AA.

Allosteric Regulation and Tonic Inhibition of PGHS-2 Involving E allo -As illustrated by the data in
Allosteric effects on huPGHS-2 that involve ligand binding and/or mutations to E allo can modulate the V max for human COX-2 activity over at least a 20-fold range from ϳ5-85 units/ mg. At the lower end, the allosteric inhibitor FBP bound to E allo causes more than 90% (albeit incomplete) inhibition. This is equivalent to an enzyme having a V max of Ͻ5 units/mg (9,14). At the other extreme one can estimate a V max of 86 units/mg for S121P Y385F/Native huPGHS-2 when tested with 25 M PA . Inhibition by naproxen, flurbiprofen, and diclofenac of Native/Native huPGHS-2 compared with the Y385F S121P/Native huPGHS-2 heterodimer. In the Y385F S121P/Native huPGHS-2 heterodimer, the S121P is in the E allo subunit as illustrated in Fig. 7  plus 5 M AA. This latter value underestimates the maximal activation by PA because E allo is less than 50% occupied by PA (versus AA) under these conditions. Potencies and Efficacies of Allosteric FA Activators and COX-2 Inhibitors-Many but not all FAs are capable of binding E allo . This includes most saturated and monounsaturated FAs, which all bind with similar but relatively low affinities (i.e. K d values of 5-10 M) but exhibit different efficacies (9,14). PA, the most efficacious common FA activator, causes a maximal allosteric activation that is 2-3 times greater than observed with AA alone, whereas maximal activation by stearate is 1.2fold (11). Dihomo-␥-linolenic, AA, and EPA bind with relatively high and comparable affinities to E allo (i.e. K d values of ϳ0.25 M (11)). Other 18, 20, and 22 carbon polyunsaturated FAs (11), 2-arachidonylglycerol, and 2-O-arachidonylglycerol ether bind poorly to E allo (10,12).
Allosteric COX-2 inhibitors also differ in their allosteric potencies and efficacies. IBP (functioning allosterically (12)), naproxen, and FBP cause about 50, 70, and 90% maximal inhibition, respectively, of AA oxygenation by native huPGHS-2 (and Y385F/Native huPGHS-2) (9, 12, 14). As noted above, when AA is unable to bind to E allo , the enzyme has ϳ55% less activity than when AA is bound to E allo (9). This implies that IBP, which causes ϳ50% inhibition via E allo (12), simply prevents AA binding to E allo without exerting an independent allosteric inhibitory effect. In contrast, the Ͼ90% inhibition seen with FBP must involve both prevention of AA binding and a significant additional allosteric inhibitory component. Naproxen, which causes about 70% maximal inhibition, must prevent AA binding to E allo and also have a modest additional allosteric inhibitory component.
Repositioning of Residues 120 -129 in E allo Is Associated with COX-2 Allosterism-Previous studies have established the involvement of Arg-120 in E allo in the allosteric responses of PGHS-2 to saturated, monounsaturated, and certain polyunsaturated FAs as well as to the allosteric COX inhibitors IBP, FBP, and naproxen (9,11,12). It is also known that loop residues 123-127 can assume two alternative configurations in X-ray crystal structures of ovine PGHS-1 (7, 28). Additionally, cross-linking studies with huPGHS-2 demonstrate that changes in the relative positions of residues 123-127 and 541-543 occur in association with the binding of some ligands to the COX sites (5). Specifically, cross-linking is affected by certain 20-carbon FAs, including AA and EPA, and many COX inhibitors, including FBP and IBP, that can bind at low micromolar concentrations to both E allo and E cat (5,11). In contrast and importantly, 2-arachidonoylglycerol and FAs such as PA, oleic acid, linoleic acid, and docosahexaenoic acid that bind efficiently to only one of the two huPGHS-2 subunits do not affect cross-linking (5,11).
The combination of crystallographic and cross-linking studies reported here indicate that the S121P mutations in S121P/ S121P PGHS-2 cause the entire segment involving residues 119 -129 to become "unhinged" from the N-terminal end of helix D of the membrane binding domain. This occurs in parallel with a substantial increase in enzyme activity that can be chiefly ascribed to the S121P substitution in E allo . This S121P substitution largely compensates for ligand binding to E allo .
Thus, in comparing enzymes having E allo subunits unable to bind ligands, we find that incorporating the S121P substitution in E allo causes greater than a 3-fold increase in the V max (i.e. comparing R120A S121P Y385F/Native versus R120A Y385F/ Native huPGHS-2). We conclude that the S121P substitution leads to a large increase in the flexibility of the segment from Arg-120 to residue 129 of the adjoining allosteric loop in E allo and that movement in this region of E allo underlies the allosteric effects mediated by FAs and allosteric inhibitors that bind E allo .
There is also evidence that movement of residues between 120 and 129 in E cat occur upon ligand binding to this subunit and that these changes also contribute to COX allosterism. We infer that there is an E cat contribution based (a) on previous studies showing that only ligands that can bind to E allo and E cat simultaneously at micromolar concentrations affect cross-linking (5,11) and (b) on observations reported here that S121P and various Arg-44 and Asp-125 substitutions in E cat have significant effects on COX activity.
Browner and co-workers (37) published the crystal structures of huPGHS-2 in complex with two inhibitors, RS104897 and RS57067, derived from a scaffold of the nonselective NSAID zomepirac. RS104897 bound in the COX channel in a similar fashion to that observed for other NSAIDs, with Arg-120, Tyr-355, and Glu-524 forming interactions to close the aperture of the channel. However, the binding of RS57067 captured the COX channel entrance in an "open" conformation. In the open conformation, Arg-120 no longer interacts with the carboxylate group of Glu-524. Instead, the side chain of Glu-524 reorients, forming an ionic interaction with Arg-513 at the base of the side pocket. Arg-120 no longer participates in the interaction network because the pyridazinone ring of RS57067 displaces Arg-120 causing helix D to become detached from the catalytic domain. The ␣-carbon of Arg-120 moves 2.3 Å compared with its position in the closed conformation. To date, this is the only COX crystal structure that exhibits a change in the interaction network at the channel entrance. Unfortunately, the coordinates for the complex were never deposited in the PDB. Nevertheless, the report of the unwinding of helix D provoked by the binding of RS57067 suggests that the structural conformation of the main chain observed in the S121P/apocrystal structure represents a transient conformation that can be sampled by native enzyme.
Changes in Allosteric Responses with S121P Substitutions-Presumably, the degree and manner in which the allosteric loop interacts with the partner E cat subunit are what modulates E cat activity. The pattern of stimulatory effects of PA Ͼ oleic acid ϳ 11-cis-eicosaenoic acid are similar for native huPGHS-2, Y385F/Native huPGHS-2, and S121P/S121P huPGHS-2. The S121P mutation in E allo increases the flexibility of the allosteric loop while maintaining to some degree the overall specificity of nsFA activation. In contrast, oleic acid and 11-cis-eicosaenoic acid are equal to or better than PA in stimulating that activity of mutants with Arg-44, Asp-125, and Ala-543 mutations. These results reinforce the idea that each different nsFA when bound to E allo must cause a subtly different alteration in or repositioning of the allosteric loop.
The inhibitory responses to FBP and naproxen were blunted by having the S121P substitution in E allo but were unaffected when this substitution was present in E cat . The binding of the allosteric inhibitor FBP to S121P/S121P muPGHS-2 in crystal structures realigns residues 119 -122 into positions observed in native muPGHS-2. One would expect a similar realignment in solution, along with realignment of residues 123-129 of the allosteric loop. Thus, it was somewhat surprising to us that S121P Y385F/Native huPGHS-2 would exhibit significantly attenuated responses to the allosteric inhibitors FBP and naproxen. We expected, for example, that FBP and probably naproxen binding would realign the structure of the E allo subunit in S121P Y385F/Native huPGHS-2 to forms that would cause the same degrees of inhibition as observed with native enzyme (Fig. 8) or Y385F/Native huPGHS-2 (9). One possibility is that the X-ray structures are providing a view of what happens in E cat but not in E allo in solution. Perhaps only a partial realignment is occurring in E allo , one that yields only incomplete responses to both activating and inhibitory ligands. This would be consistent with the observation that the magnitude of the effect of the S121P mutation on COX activation is reduced but not eliminated in the presence of AA binding to E allo as discussed above.
Although it is clear from in vitro studies that COX-2 activity can be regulated allosterically by FAs and allosteric COX inhibitors, the physiologic significance of allosterism in COX-2 regulation remains to be explored. Changes in PG biosynthesis are observed in animals fed fish oil FAs (42), but these changes could arise from changes in substrate levels or other effects rather than allosterism. There is also a difference between naproxen and other nonspecific NSAIDS and COX-2 inhibitors in their cardiovascular profiles (43), but no data of which we are aware comparing naproxen and other inhibitors for their effects on PG formation in vivo.

Experimental Procedures
Materials-Arachidonic acid (AA) and oleic acid (18:19) were purchased from Cayman Chemical Co. Nickel-nitrilotriacetic acid Superflow resin and nickel-nitrilotriacetic acid were from Qiagen. Flurbiprofen, indomethacin, aspirin, naproxen, 11-cis-eicosaenoic acid (20:19), FLAG peptide, and FLAG affinity resin were from Sigma. Celebrex (celecoxib) was from physician samples. Ibuprofen was from Tocris Bioscience. Hemin was from Frontier Scientific, Logan, UT. Co 3ϩ -protoporphyrin IX was purchased from Frontier Scientific. BCA protein reagent was from Pierce. Polyacrylic acid (sodium salt) 5100 was from Hampton Research Corp. [1-14 C]AA (1.85 GBq/ mmol) was from American Radiolabeled Chemicals. C10E6 and decyl maltoside were purchased from Anatrace (Maumee, OH). N-Octyl ␤-D-glucopyranoside (␤OG) was purchased from Inalco (San Luis Obispo, CA). Hexane, isopropyl alcohol, and acetic acid were HPLC grade from Thermo Fisher Scientific, Inc. Complete protease inhibitor was from Roche Applied Science. Restriction enzymes were from New England Biolabs, Inc. All other materials were analytic grade from Sigma.
The QuikChange TM mutagenesis kit was purchased from Stratagene (La Jolla, CA). The Bac-to-Bac baculovirus expression kit and associated reagents, including fungizone and penicillin/streptomycin, were purchased from Invitrogen. Fetal bovine serum, Spodoptera frugiperda 21 (Sf21) insect cells, and ESF 921 protein-free insect cell media were purchased from Expression Systems (Davis, CA). HiTrap HP chelating and HiPrep Sephacryl 300-HR chromatography columns were purchased from GE Healthcare.
Preparation of Mutants-huPGHS-2 homodimers were prepared as detailed previously (9). A mutant protein variant denoted S121P/S121P huPGHS-2 indicates a variant in which both monomers of the huPGHS-2 dimer bear this mutation. Heterodimers were prepared as described in previous reports (4,44) or using the pFastBac dual vector (9). A PGHS-2 variant denoted S121P/Native huPGHS-2 indicates a variant in which one monomer of the huPGHS-2 heterodimer bears the S121P mutation, and the other monomer has the native sequence. A cDNA for huPGHS-2 containing an octahistidine (His 8 ) tag at the N terminus was subcloned into pFastBac plasmid (Invitrogen)). The QuikChange site-directed mutagenesis protocol (Stratagene) was used to construct the mutants. The presence of the various mutations was confirmed by DNA sequencing. Native huPGHS-2, mutant huPGHS-2 homodimers, and heterodimeric huPGHS-2 variants were expressed and purified using procedures described previously (9).
Cyclooxygenase Analyses-COX activities were determined using measurements of O 2 consumption using an O2 electrode essentially as detailed previously (9). In experiments involving COX inhibitors, the enzyme preparations were typically preincubated with the inhibitor, and an aliquot was added to the reaction chamber. The rates reported are maximal rates occurring after about a 10-s lag phase.
Difference absorption spectra of the S121P/S121P huPGHS-2 homodimer were obtained upon titration with heme, and K d values for heme and heme binding stoichiometry were calculated as described previously (9,14,15).
Quantitation of the binding of [1-14 C]AA to E allo of various homodimers and heterodimers at high enzyme to substrate ratios was as described in earlier reports (9,11,14,15).
Cyclooxygenase product analyses were conducted in 100-l reaction mixtures containing 1 M [1-14 C]AA, 0.10 -2.0 M huPGHS-2 or mutant variants, 5 M hematin and 1 mM phenol in 0.1 M Tris-HCl, pH 8.0, at 37°C. The reactions were quenched with 100 l of stop buffer consisting of ethyl acetate/ acetic acid (95:5; v/v). The resultant samples were centrifuged at room temperature for 5 min at 4000 ϫ g. An aliquot of the organic layer (100 l) was subjected to HPLC on a Luna C18 (2) column (5 m, 250 ϫ 4.6 mm, Phenomenex) mounted on a Shimadzu HPLC system equipped with radio-HPLC detector (IN/US system, ␤-RAM model 4). The bound products were eluted in a gradient elution mode eluting with Solution A (acetonitrile/water; 30:70) for 5 min and then with a linear gradient of Solutions A and B (acetonitrile/water; 90:10) for 15 min, then Solution B for 8 min, and finally Solution A for 4 min; Solutions A and B both contained 0.1% acetic acid (v/v). The flow rate was 1 ml/min. Representative elution times for PGH2, 17-hydroxyheptadecatrienoic acid, and AA were 14, 18, and 30 min, respectively. Prostaglandin product analysis was determined using the same methods as those used for measuring [ Oxidant-induced Disulfide Cross-linking between Monomers of huPGHS-2-Before cross-linking, the PGHS-2 variants containing cysteine substitutions (1 M monomer) (i.e. P127C S541C ⌬C-huPGHS-2, R120Q P127C S541C ⌬C-huPGHS-2, and S121P P127C S541C ⌬C-huPGHS-2) were reconstituted with a 1.5 M excess of Co 3ϩ -protoporphyrin IX. Cross-linking to form disulfide bonds was initiated by making the solution 250 M Cu(II)/(1,10-phenathroline) 3 (i.e. 250 M CuSO 4 plus 625 M 1,10-phenathroline) at room temperature for various times. To test the effect of FBP on cross-linking, the proteins were pre-incubated with FBP for 10 min before addition of the Cu(II)/o-phenanthroline oxidant. The reactions were stopped by adding SDS-loading buffer (Invitrogen) containing 20 mM N-ethylmaleimide and 20 mM EDTA (and no reductants). The reaction mixtures were then subjected to SDS-PAGE, and immunoblotting was performed with a rabbit polyclonal antibody reactive with against huPGHS-2 as detailed previously (45).
Crystallography-The S121P mutation in muPGHS-2 was generated using the QuikChange mutagenesis kit and the previously engineered His 6 -N580A muPGHS-2 in pFastBac1 (27) as the template. This protein variant is denoted S121P/S121P muPGHS-2 to indicate that both monomers of the dimer bear this mutation. Expression and purification of S121P/S121P muPGHS-2 for structural characterization was carried out as described previously (27). Prior to crystallization, purified S121P/S121P muPGHS-2 was concentrated to 3.3 mg/ml using a 50-kDa MWCO centrifugal concentrator (Millipore) and reconstituted with a 2-fold mole excess of Co 3ϩ -protoporphyrin IX to generate the holoenzyme. Crystals were grown at 23°C using the sitting drop vapor diffusion method by mixing 3 l of protein solution with 3 l of a solution consisting of 23-34% polyacrylic acid 5100, 100 mM HEPES, pH 7.5, 20 mM MgCl 2 , and 0.6% (w/v) ␤OG. Drops were then equilibrated over reservoir solutions containing 23-34% polyacrylic acid 5100, 100 mM HEPES, pH 7.5, and 20 mM MgCl 2 . For diffraction analysis, crystals were flash-frozen directly in the gaseous nitrogen stream after soaking in drop solution supplemented with 10% ethylene glycol. To generate inhibitor-bound structures, crystals were soaked in the above cryoprotectant that also contained 1 mM inhibitor for 20 min prior to flash freezing. X-ray diffraction data were collected on beamline 17-ID at the Advanced Photon Source (Argonne National Laboratory) using a Pilatus 6M detector and processed with HKL2000 (46). Data collection statistics are reported in Table 4.
Molecular replacement methods were used to determine the initial phases for each structure, with an emphasis on minimizing the introduction of model bias as described previously (27). A truncated search model of muCOX-2 derived from PDB entry 3HS5 was utilized in PHASER (47). Subsequent phases obtained from PHASER were input into ARP/wARP utilizing the "automated model building starting from experimental phases" option (48), which successfully built most of the deleted portions of the models. Iterative cycles of model building and refinement, using COOT (49) and PHENIX (50), were then implemented to fit the remaining residues and to add waters, inhibitors, and other ligand molecules. Translation-librationscrew (TLS) refinement (51), utilizing the TLSMD web server (52,53) was carried out during the final rounds of refinement. Final refinement statistics are summarized in Table 4. Model validation was carried out using MolProbity (54) and simulated annealing omit maps were calculated in PHENIX. The coordinates and structure factors for S121P/apo, S121P/cbx, and S121P/fbp have been deposited in the Protein Data Bank (PDB codes 5JVY, 5JW1, and 5JVZ, respectively).
Thermal Shift Assay-Thermal shift assays were performed using a Stratagene Mx3005P real time PCR instrument and the thiol reactive fluorescent dye 7-diethylamino-3-(4Ј-maleimidylphenyl)-4-methylcoumarin (CPM) (55). Native, R120A (29), and S121P/S121P muPGHS-2 were added to a final concentration of 1 M in a 30-l total reaction volume containing 25 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 0.53% (w/v) ␤OG. To analyze inhibitor binding, 50 M of either FBP, naproxen, or celecoxib, or 100 M IBP was added to the reaction, followed by incubation on ice for 30 min. CPM was then added to a final concentration of 25 M. The change in fluorescence was monitored using an Alexa filter with excitation and emission wavelengths of 350 and 440 nm, respectively. Temperature was raised from 25 to 98°C in 0.5°C increments over the course of 45 min, with fluorescence readings taken at each interval. The fluorescence data were plotted and normalized, and the first derivative of the curve was calculated to provide the T m . The change in melting temperature (⌬T m ) was calculated as the difference between the T m of the native or mutant construct in the presence of inhibitor compared with the native or mutant construct alone.
Author Contributions-L. D. and C. Y. prepared and characterized huPGHS-2 mutants and contributed to the data analyses and writing of the manuscript. B. J. O. and M. G. M. carried out X-ray crystallographic structure determinations and thermal melting temperature experiments. W. L. S. and the other authors were involved in experimental design, data analysis, and writing of the manuscript.