Structural characterization of arachidonyl radicals formed by aspirin-treated prostaglandin H synthase-2.

Peroxide-generated tyrosyl radicals in both prostaglandin H synthase (PGHS) isozymes have been demonstrated to couple the peroxidase and cyclooxygenase activities by serving as the immediate oxidant for arachidonic acid (AA) in cyclooxygenase catalysis. Acetylation of Ser-530 of PGHS-1 by aspirin abolishes all oxygenase activity and transforms the peroxide-induced tyrosyl radical from a functional 33-35-gauss (G) wide doublet/wide singlet to a 26-G narrow singlet unable to oxidize AA. In contrast, aspirin-treated PGHS-2 (ASA-PGHS-2) no longer forms prostaglandins but retains oxygenase activity forming 11(R)- and 15(R)-hydroperoxyeicosatetraenoic acid and also retains the EPR line-shape of the native peroxide-induced 29-30-G wide singlet radical. To evaluate the functional role of the wide singlet radical in ASA-PGHS-2, we have examined the ability of this radical to oxidize AA in single-turnover EPR studies. Anaerobic addition of AA to ASA-PGHS-2 immediately after formation of the wide singlet radical generated either a 7-line EPR signal similar to the pentadienyl AA radical obtained in native PGHS-2 or a 26-28-G singlet radical. These EPR signals could be accounted for by a pentadienyl radical and a strained allyl radical, respectively. Experiments using 11d-AA, 13(R)d-AA, 15d-AA, 13,15d(2)-AA, and octadeuterated AA (d(8)-AA) confirmed that the unpaired electron in the pentadienyl radical is delocalized over C11, C13, and C15. A 6-line EPR radical was observed when 16d(2)-AA was used, indicating only one strongly interacting C16 hydrogen. These results support a functional role for peroxide-generated tyrosyl radicals in lipoxygenase catalysis by ASA-PGHS-2 and also indicate that the AA radical in ASA-PGHS-2 is more constrained than the corresponding radical in native PGHS-2.

Substantial evidence supports the branched-chain radical mechanism for PGHS catalysis originally proposed by Ruf and co-workers (14,15). In this mechanism, a specific proteinlinked tyrosyl radical produced in the peroxidase cycle serves a catalytic role in cyclooxygenase activity. EPR observations in single turnover experiments provide convincing evidence that the peroxide-generated tyrosyl radicals in PGHS-1 and -2 can oxidize AA, the cyclooxygenase substrate, thus generating a pentadienyl AA radical characterized by a 7-line EPR spectrum (16 -18). The tyrosyl radical mechanism explains the observed heme dependence of both enzyme activities and the requirement of the cyclooxygenase for a hydroperoxide activator (3,19). The mechanism also is consistent with crystallographic data that show a tyrosine residue (Tyr-385 in PGHS-1 and Tyr-371 in PGHS-2) positioned between the heme and the AA binding site (20 -23).
In PGHS-1, aspirin treatment causes a change of the peroxide-induced tyrosyl radical EPR spectrum from a 34 -35-G doublet to a 25-26-G narrow singlet; this narrow singlet tyrosyl radical is unable to oxidize AA (16,24). In contrast, the peroxide-induced tyrosyl radical in ASA-PGHS-2 retains the EPR characteristics of the peroxide-induced tyrosyl radical in native PGHS-2, a 29 -30-G wide singlet (25). In the present study, we have examined whether the wide singlet tyrosyl radical of ASA-PGHS-2 is capable of AA oxidation to support the 11-and 15-lipoxygenase activity. The results indicate that the peroxide-induced tyrosyl radical in ASA-PGHS-2 does indeed react with AA to generate fatty acid radical intermediates. Interestingly, the AA radical species in ASA-PGHS-2 appears capable of adopting either a pentadienyl structure (as found with native PGHS-2) or a more constrained conformation (similar to that observed with native PGHS-1).
PGHS-1 was purified from sheep seminal vesicles (26), and human PGHS-2 was purified from recombinant material expressed in a baculovirus system (27). The holoenzymes were prepared by replenishing with heme; excess heme was removed by passage over DEAE-cellulose (28). Cyclooxygenase activity was determined by the rate of oxygen consumption (26). The batches of PGHS-1 used in this study had specific activities of 100 -120 mol of O 2 /min/mg; the PGHS-2 batches had specific activities of 20 -42 mol O 2 /min/mg. ASA-PGHS-2 was prepared as described previously (25). Briefly, the PGHS-2 holoenzyme was treated with aspirin, and the residual oxygenase activity was monitored periodically as the incubation proceeded at room temperature. When the oxygenase activity dropped to half the phenol, and 10% glycerol was incubated on ice under anaerobic conditions with 5 eq of EtOOH for 17 s before the addition of d 8 -AA (1.5 eq). EPR spectra were obtained for samples taken 9 s after the addition of EtOOH (a) and 7 s (b) or 36 s (c) after the subsequent d 8 -AA addition. Right panel, a second set of EPR spectra (aЈ-cЈ) was obtained after the three samples in the left panel were thawed and incubated aerobically on ice for 30 s. The numbers at the left side of each spectrum are the signal intensities obtained from double integration (in spins/heme). The dashed line is the simulation using parameters for PGHS-2 and d 8 -AA given in Table I, third column. EPR conditions for the Varian E6 spectrometer were: modulation amplitude 2 G, microwave frequency 9.2 GHz, 1 mW power, and temperature 96 K. SCHEME 1. Aspirin pretreatment effects on eicosanoid products produced by PGHS-1 and -2. PGH 2 , prostaglandin H 2 .
original value (reflecting the decreased oxygen consumption stoichiometry upon transition from cyclooxygenase to lipoxygenase catalysis), the reaction mixture was passed over a desalting column (10DG, Bio-Rad) to remove salicylate and unreacted aspirin. The use of a standardized oxygenase activity criterion for ending aspirin treatment was prompted by the observation of considerable variation in residual oxygenase activity after extended aspirin treatment among individual batches of PGHS-2 (data not shown). When necessary, the ASA-PGHS-2 was concentrated by ultrafiltration on a YM30 membrane (Amicon) before the addition of 15-20% glycerol and storage at Ϫ70°C.
Single turnover experiments using an anaerobic titrator were performed as previously described (16,17). EPR spectra were recorded at liquid nitrogen temperatures on a Varian E-6 or a Bruker EMX spectrometer (16). The EPR conditions were: modulation amplitude, 2 or 3.2 G; time constant, 1 s; power, 1 mW; temperature, 96 K for the Varian E-6 and 110 K for the Bruker EMX. Radical concentrations were determined by double integration of the EPR signals, with reference to a copper standard (16). Computer simulations of EPR spectra were performed on a PC using a modified version of the POWFUN program (29), kindly provided by Drs. Gerald T. Babcock and Curt Hoganson, Michigan State University or by the Simfonia package included in the Bruker system. The two simulation programs gave similar results.
Spin density calculations were performed using Gaussian 98W (Gaussian, Inc.). The B3LYP density functional theory was employed and applied to the 6 -31G ϩ basis set. 2-Pentene was built using Gauss-View 2.1 for Windows (Gaussian, Inc.) and converted into the corresponding neutral radical by deleting a hydrogen atom on carbon 4. The dihedral angle subtended by carbons 2-5 was varied using GaussView.

Sequential Reaction of ASA-PGHS-2 with Hydroperoxide
and Arachidonate-When ASA-PGHS-2 was reacted anaerobically with 5 eq of EtOOH, the EPR spectrum revealed a 29-G wide singlet (denoted WS2) (19) amounting to about 0.2 spin/ heme (Fig. 1, spectrum a). A new radical was formed upon subsequent anaerobic addition of 1.5 eq of d 8 -AA (Fig. 1, spectra  b and c). This radical displayed a 5-line EPR centered at g ϭ 2.0027, with hyperfine splittings of 13.8 -14.5 G. These EPR characteristics are very similar to those of the d 8 -AA radical observed earlier in both PGHS-1 and -2 (16,17). The samples trapped at 7 and 36 s after the addition of d 8 -AA showed a similar radical species, with intensities of 0.2-0.3 spin/heme. Thawing and aerobic incubation of the samples with the d 8 -AA-derived radical regenerated the WS2 tyrosyl radical (Fig. 1, spectra bЈ and cЈ), consistent with formation of PGG 2 by reaction of the d 8 -AA fatty acid radical with oxygen and subsequent reformation of the tyrosyl radical.
The same procedure was used to trap AA-derived radicals during the reaction of 11d-AA, 15d-AA, 13(R)-d-AA, 13(R),15d 2 -AA, and 16d 2 -AA with the tyrosyl radical of ASA-PGHS-2. EPR spectra of the radical species obtained after anaerobic addition of each labeled AA to pre-formed tyrosyl radical are shown in Figs. 2 and 3. Substrate labeled with a single deuterium at positions 15,13(R) or 11 led to a sextet EPR signal (spectra B and C in Fig. 2 and spectrum B in Fig. 3, respectively), whereas the 13,15 double-labeled AA gave rise to a 5-line spectrum similar to that obtained for d 8 -AA (compare spectra D and E in Fig. 2). The double-labeled 16d 2 -AA yielded a 6-line radical EPR signal (Fig. 3, spectrum B), indicating that only one of the two ␤ protons interacts strongly with the unpaired electron.
The results obtained using unlabeled AA were more complicated; representative data are shown in Fig. 4. The anaerobic addition of 1.5 eq of AA to ASA-PGHS-2 containing the WS2 tyrosyl radical (19) led to the replacement of the WS2 signal EPR spectra of substrate-derived radicals generated by 9 -14 M ASA-PGHS-2 after sequential anaerobic mixing with 10 eq EtOOH and 1.5 eq of unlabeled AA (spectrum A), 15d-AA (spectrum B), 13(R)d-AA (spectrum C), 13,15d 2 -AA (spectrum D), and d 8 -AA (spectrum E). Experimental procedures were essentially the same as those described in Fig. 1. Dashed lines overlaid on the spectra represent computer simulations based on the parameters in Table I using 0.15 of the value of the corresponding proton coupling constant for each deuterium substitution. (Fig. 4, spectrum a) with a new signal very similar to that of WS2 (Fig. 4, spectrum b). This second radical EPR has an overall line width of 27 G centered at g ϭ 2.0022 with similar total spin concentration to that of WS2. A sample freezetrapped 24 s later exhibited a very similar EPR line shape but with substantially increased spin concentration (Fig. 4, spectrum c). These three samples subsequently were thawed and mixed with air; all three showed decreased radical intensity with little change in line-shape ( Fig. 4, spectra aЈ, bЈ, cЈ). Spectra b and c in Fig. 4 are very different from the corresponding 7-line EPR spectrum obtained for control PGHS-2 sample (top spectrum in Fig. 6) (17). The lack of significant changes in the EPR spectrum after the addition of AA to ASA-PGHS-2 containing the tyrosyl radical (spectra b and c in Fig. 4) is unlikely to be due to oxygen contamination, because parallel experiments conducted under identical conditions using d 8 -AA clearly show the generation of an AA-derived radical ( Fig. 1, spectra b and c). The experiment described in Fig. 4 was repeated several times with different batches of ASA-PGHS-2. A similar narrow 26 -27-G EPR spectrum was obtained for samples trapped after AA addition in four of these replicate experiments. In one experiment, however, the narrow 27-G spectrum appeared initially after AA addition, with a 7-line EPR spectrum appearing subsequently (data not shown). In two other experiments, the addition of AA produced what seems to be a 7-line spectrum with a substantial amount of unconverted tyrosyl radical signal and perhaps some amount of the new narrow signal (spectrum A in Fig. 2). The variability evident in these results suggests that the radical generated from unlabelled AA in ASA-PGHS-2 can assume two different conformations, one producing a 27-G narrow singlet and the other producing a 7-line pentadienyl radical.
Power saturation studies were conducted for the 27-G singlet and both the 7-and 5-line EPR radical species. All three species relax more efficiently than isolated organic radicals, showing half-saturation power values around the milliwatt level at liquid nitrogen temperature (panel A in Fig. 5). The power dependence for the 7-line AA radical and 5-line d 8 -AA radical were determined for both the center and the immediately adjacent wing signals, with the wing signals showing a higher P 1/2 value (panel B in Fig. 5). This is expected because the wing hyperfine lines have more contributions from nuclear spins to the relaxation (30). The saturation behavior of these threecarbon-centered radical species was very similar to that of the tyrosyl radical (panel A in Fig. 5).
Characterization of Carbon-centered Arachidonate Radicals and Analysis of EPR Spectra-The 7-line AA radical EPR spectra obtained with PGHS-2 or ASA-PGHS-2 are essentially identical and can be simulated using the parameter set published earlier (Table I) (17). Representative EPR data and the corresponding pentadienyl radical simulation are shown in Fig.  6 (top two traces). The parameter values producing optimal simulations are close to those of a prototypical pentadienyl radical (38) (fourth column, Table I). Our previous structural interpretation for this pentadienyl radical involves six distinct protons interacting strongly with the unpaired electron generated by initial hydrogen abstraction at the 13-pro-S position of AA (17). As illustrated at the top of Fig. 7, these six protons are located at C10, C11, C13, C15, and C16. We favored an assignment with two strongly interacting C10 ␤ protons at dihedral angles of 32°/152°and one strongly interacting C16 ␤ proton at a dihedral angle of 51°. This arrangement places C9 closer to C11 for endoperoxide formation and also brings C8 closer to C12 for subsequent ring closure. These assignments fix the conformation of the fatty acid skeleton from C9 to C17 of the carbon-centered substrate radical formed in the first step of the cyclooxygenase reaction with unlabeled AA in PGHS-2. When d 8 -AA was used, instead of obtaining a 7-line EPR, we observed  Table I using 0.15 of the value of the corresponding proton coupling constant for each deuterium substitution. A coupling constant of 13 G was used for both C10 ␤-protons for simulation of the 16d 2 -AA radical. EPR conditions for the Bruker EMX spectrometer were: 2 G modulation amplitude; microwave frequency, 9.29 GHz; 1 mW power; time constant, 0.327 s; temperature, 110 K. a 5-line EPR, characteristic of the pentadienyl d 8 -AA radical (Fig. 1, spectra b and c, and Fig. 2, spectrum E). This 5-line EPR is also optimally simulated using our previous parameters for the d 8 -AA radical intermediate found in native PGHS-2 (Table  I, third column) (17). The loss of one multiplicity in the spectra generated with the singly deuterated substrates are attributed to the loss of one hyperfine interaction with the protons at C11, C13(R), or C15. The same set of parameters used to fit the EPR data of unlabeled AA also reasonably simulated the EPR spectra obtained with the singly and doubly deuterated AAs when the value of the coupling constant for the deuterated position(s) was multiplied by 0.15 to account for the smaller nuclear gyromagnetic ratio of deuterium relative to hydrogen (Fig. 2,

spectra B-E).
A pentadienyl radical is expected to have a large overall width (ϳ80 G) and, thus, cannot adequately explain the narrow EPR spectrum of the unlabeled AA radical found with ASA-PGHS-2 (Fig. 4, spectra b and c). Attempts to deconvolute these narrow spectra as an arithmetic mixture of a pentadienyl radical and the WS2 tyrosyl radical were not successful due to the lack of wing features inherent in the EPR of the radical. The most likely alternative structures are conformations that limit electron delocalization and minimize the number of strongly interacting protons, thus decreasing the width of the signal. A good candidate is an allyl radical, with the unpaired electron delocalized over either C11-C13 or C13-C15. Such allyl radicals could arise if ASA-PGHS-2 bound the fatty acid in a conformation in which the C11-C12 and C14-C15 double bonds were not coplanar. With this model in mind and using the coupling constant values of typical allyl radical models as guidance (43), we performed computer simulations to obtain an optimal set of parameters to fit the data (Table I); the simulated allyl radical spectrum is shown in Fig. 6. Initial simulations using equal hyperfine interactions of the two endo protons (either the C11/ C13 pair or the C13/C15 pair) and isotropic A tensor values failed to fit the data well. Changing from isotropic to anisotropic A tensors with a 3:1:2 ratio, as reported for other allyl radicals (31)(32)(33), led to significant improvement in the fit of the line shape but not the line width. Final adjustment to give unequal hyperfine interactions with the endo protons, with the proton at position 13 having a larger hyperfine value (A iso ϭ 15.3 G) than the protons at position 11 or 15 (A iso ϭ 8.0 G), led to the simulation shown in Fig. 6. Such an uneven spin-density distribution suggests a twisted allyl radical conformation. Interestingly, this same set of parameter values also closely simulates the narrow EPR spectrum found in PGHS-1 (Fig. 6, bottom spectrum), suggesting that this previously unexplained EPR signal (16) may also arise from a twisted allyl radical species.
The optimal values for the A tensors of the ␤ protons derived from these simulations allowed calculation of the dihedral angle, , between the p z orbital at C11 or C15 with the C10-H or C16-H bond, using the conventional McConnell relationship (33,46), where A H is the coupling constant due to hyperconjugation, is the unpaired electron density at the terminal carbon of the pentadienyl radical, and the A tensor value measured for the endo proton of the putative allyl radical (8.0 G) relative to that of a methyl proton in the ethyl radical (23 G) can be used to estimate a value of 0.35 (33). The dihedral angles associated with the strongly interacting ␤-protons at C10 or C16 can then be calculated to be 52°a nd Ϫ68°, respectively. The proposed conformations of the two possible allyl radicals are shown at the bottom of Fig. 7. Direct reaction of these intermediates with molecular oxygen would be expected to produce 11-or 15-HPETE.
Differences in the Chemical and Physical Properties of AA and d 8 -AA-The question arises why unlabeled AA and deuterated AA would give rise to different fatty acid radical signals. One possibility involves altered interactions between the fatty acids and the protein. The plausibility of such non-covalent isotope effects was evaluated by HPLC (Fig. 8). The retention time of d 8 -AA was several minutes shorter than AA on a C 18 reversed-phase column, indicating that deuterated AA had weaker interactions with the hydrophobic column matrix.  (53) to the equation log(S/P 1 ⁄2) ϭ Ϫb/2 log(P 1 ⁄2 ϩ P) ϩ b/2 log(P 1 ⁄2) ϩ log K, where P 1 ⁄2 is the power at half-saturation of the signal, and b and K are floating parameters. The optimal values obtained for b are shown in the figure.
The P 1/2 values obtained from nonlinear regression were 1.90 (center) and 3.54 (wing) mW for the PGHS-2 AA radical, 0.46 mW for the ASA-PGHS-2 AA radical, 1.24 (center) and 1.38 (wing) mW for the ASA-PGHS-2 d 8 -AA radical, and 0.44 mW for the tyrosyl radical.  (38). The endo protons in the pentadienyl radical are not equivalent to the ␤ protons in the AA radical. b Data are from Bascetta et al. (43). The endo protons in the AA allyl radical are either the H13/H15 or H11/H13 pairs. The exo protons are either the ␤ protons at C10 or C16.
c The A tensors of endo protons in the allyl radical are assumed to be anisotropic at a ratio of 3:1:2.
The possibility of biochemical differences between AA and d 8 -AA was further explored by measurement of their steadystate kinetic parameters for PGHS-1 cyclooxygenase activity. V max and K m values were determined by oxygen consumption assays using 20 nM PGHS-1 and either AA or d 8 -AA as substrate at 30°C (Table II). Values of 0.54 Ϯ 0.03 M O 2 /s and 4.6 Ϯ 0.7 M were obtained for V max and K m when AA was the substrate, whereas 0.50 Ϯ 0.03 M O 2 /s and 5.0 Ϯ 0.7 M were obtained as d 8 -AA was the substrate. Similar parameter values were also obtained for 13(R)d-AA and 15d-AA (Table II). Isotope replacement, thus, had little effect on the steady-state cyclooxygenase kinetics.

Oxidation of AA by Peroxide-induced ASA-PGHS-2 Radical(s)-
Oxidation of AA to a fatty acid radical is a key step in the proposed branched-chain mechanism for PGHS cyclooxygenase catalysis (14,15). It has been previously demonstrated that the peroxide-induced tyrosyl radicals characterized by a wide doublet (33-35 G) or wide singlet (33-35 G) in PGHS-1 and a wide singlet (29 -30 G) in PGHS-2 can oxidize AA to produce a carbon-centered AA radical (16 -18). The alternative tyrosyl radical characterized by a narrow singlet (25-26 G) formed in PGHS-1 treated with cyclooxygenase inhibitors or in PGHS-1 with the Y385F mutation was not able to oxidize AA to form the AA derived radical (16,17,34).
With a link between the wide singlet tyrosyl radical and oxygenase capacity established in PGHS-2 and ASA-PGHS-2, it was important to test the ability of the wide singlet radical in ASA-PGHS-2 to actually carry out the oxidation of AA required in lipoxygenase catalysis. The present results show that the 29 -30 G wide singlet in ASA-PGHS-2 is indeed able to oxidize AA to generate an AA-derived radical (Figs. 1-3). The EPR line-shape changes observed with ASA-PGHS-2 in the AA radical upon deuterium substitution are almost identical to those found for the same substrates with native PGHS-2 ( Fig. 1 in Peng et al. (18)). A smaller coupling constant for the ␤-protons at C10 was needed to achieve adequate simulation of the signal produced by 16d 2 -AA, indicating that the spin-density at C11 and C15 may not be equal. The 6-line signal observed upon incubation of ASA-PGHS-2 with 16-d 2 -AA indicates that there is only one strongly interacting ␤-proton at C16, similar to that observed with the AA radical in PGHS-2 (52).
The oxygenase activities in PGHS-2 and ASA-PGHS-2 give rise to very different products. Given the similarity in the EPR signals of the radicals produced with deuterated AA substrates, the mechanistic divergence in the two enzymes must occur at some point after AA radical formation. This is consistent with analyses of the stereochemistry of hydrogen atom abstraction at C13 and oxygen addition at C15 and with the K m,AA values for ASA-PGHS-2 and PGHS-2, all of which indicate that the methyl terminus of AA adopts an altered conformation when Ser-516 is acetylated (25,35,36).
Structure of the Arachidonic Acid Carbon-centered Radical in ASA-PGHS-2-The finding of two distinct types of EPR signal for the carbon-centered radical in ASA-PGHS-2 with FIG. 6. EPR spectra of AA radicals generated during single turnover reactions of peroxide-induced tyrosyl radicals in PGHS-2, ASA-PGHS-2, and PGHS-1 with AA. Individual protein samples were reacted sequentially with EtOOH (1.5 eq for PGHS-1 and 10 eq for others) and AA (1.5 eq) under very similar conditions to those described in Figs. 1 and 2. The thick traces are the EPR spectra, and the thin traces are the simulations generated using the parameter values listed in Table I. unlabeled AA but only one signal in experiments with labeled AAs complicates data interpretation. With d 8 -AA as substrate the AA radical has the same pentadienyl conformation in both PGHS-2 and ASA-PGHS-2. However, with unlabeled AA the pentadienyl radical was observed only in some of the samples, the conversion from tyrosyl radical to AA radical was often incomplete (Fig. 2, spectrum A), and a signal that can be simulated as a twisted allyl radical was frequently observed ( Fig.  4 and Table I).
Allyl radicals containing various electron-withdrawing groups at the exo position show minimal perturbation of the electron density distribution (37). The isotropic hyperfine value for the two endo-protons at C11/C15 in the putative ASA-PGHS-2 AA allyl radical is much smaller than expected from allyl radical models (Table I) (37,38). Nelson et al. (33) did observe an allyl radical intermediate in purple lipoxygenase with hyperfine coupling constants of the endo protons 2-3 G smaller than a typical allyl radical model. The structural implication of that observation was not entirely clear, although interaction of an adjacent double bond with the ferrous center or addition of oxygen to this double bond were suggested as possible explanations. Such mechanisms do not apply in our case because the heme center is distant from the site of AA radical formation, and the present experiments were conducted under anaerobic conditions. A recent interesting example of a twisted allyl radical was observed during the inhibition of lysine 2,3-aminomutase by trans-4,5-dehydrolysine. This radical showed a large difference of the coupling constant for the terminal protons (17.8 versus 11.1 G) (45). It was proposed that the protein environment induced the formation of the thermodynamically unfavored twisted allyl radical (45). However, the sum of the coupling constants of the terminal protons of the allyl radical (28.9 G) is noticeably larger than the correspond-ing value 23.3 G for the putative allyl radical species observed in this study. This difference is related to the variation of the McConnell's proportional constant Q that relates the coupling constant (A H ) and spin-density () (46), A H ϭ Q ϫ .
The value of Q varies from 20 to 30 G and is affected by the effective nuclear charge of the carbon involved in the C-H bond (47). The requirement of a low Q value to simulate our allyl radical EPR data implies a larger nuclear charge contribution compared with that found for lysine 2,3-aminomutase.
The best simulation of the putative allyl radical in ASA-PGHS-2 predicts a hyperfine coupling to the endo-proton at C13 that is much larger than would be expected for a planar allyl radical (Table I), suggesting a distortion away from planarity. Hyperconjugation of the ␤ protons at either C12 (for a C13-C15 allyl radical) or C14 (for a C11-C13 allyl radical) is expected to be minimal if the -system of the isolated double bond were nearly perpendicular to the -system of the allyl radical. This postulated distorted allyl AA radical presumably results from steric constraints exerted on the bound fatty acid by the protein structure.
In view of the unexpected adjustments of hyperfine coupling constants that were required to simulate the observed EPR spectrum, we have examined the consequences of structural distortions on the spin density distribution in the related 1,3dimethyl allyl radical using density functional theory calculations. As expected, when the allyl radical is planar, the spin densities on carbons 2 and 4 are identical (Fig. 9, dihedral angle of 0°); this planar conformation is the structure obtained after geometry optimization. When the radical is distorted away from planarity by rotation about the C3-C4 bond, the spin densities on carbons 2 and 4 vary systematically as the dihedral angle is increased to 90°, with the density on C2 decreasing and that on C4 increasing (Fig. 9). At 90°, there is almost no unpaired spin on C2 and a large, positive spin density on C4. A   FIG. 7. Proposed conformations for the pentadienyl and putative strained allyl AA radicals generated by abstraction of the 13-pro S hydrogen from AA in ASA-PGHS-2. The dihedral angles between the p z orbitals at C11 or C15 and the ␤ protons at C10 or C16 were estimated from the EPR data by computer simulations (Figs. 1 and 2) and calculated using Eq. 1 as described under "Results." The double arrow indicates a hypothetical interconversion between the pentadienyl and allyl radicals, and the single arrows denote the chemical steps in the conversion of AA to lipoxygenase products. similar result was obtained in a calculation using a C11-C13 arachidonic acid allyl radical. In this case, it is the spin density on C11 and C13 that is sensitive to the C11-12-13-14 dihedral angle. The density at C11 (C13) is almost at a maximum (minimum) in the conformation found in the crystallographic structure of the PGHS-1-AA complex (23) and essentially zero when the dihedral angle is rotated 90°away from this conformation. The spin densities on the other carbon atoms change only slightly over this range of rotation. A dihedral angle near 45°matches the spin density distribution needed to account for the putative allyl radical obtained with unlabeled AA (Table I and Fig. 7). The ratio of the spin-density at C2 and C4 computed for a 45°dihedral (0.53, Fig. 9) is almost identical to the ratio of the proton coupling constants between C11 and C13 (or C15 and C13) (0.52, Table I). This result provides theoretical support for the hypothesis of a strained allyl radical generated in ASA-PGHS-2 using unlabeled substrate.
Crystallographic data indicate that the volume of the cyclooxygenase substrate channel is substantially smaller in PGHS-1 than in PGHS-2 (4, 20 -23). Mutational analysis indicates this additional constriction in the PGHS-1 channel is the result of amino acid differences at positions 434, 513, and 523 (41). This difference in the size of the active site can account for the effects of acetylation of Ser-530 on oxygenase activity; activity is abolished in PGHS-1 but retained in PGHS-2. We propose that the observation of a putative allyl radical in native  PGHS-1 but not native PGHS-2 is another indication of the greater steric restrictions in the PGHS-1 cyclooxygenase site. A narrow EPR spectrum derived from an AA radical has been typically observed with PGHS-1 (Ref. 16 and the bottom spectrum in Fig. 6), and this narrow EPR signal can be simulated well using the same parameter set used to simulate the putative allyl AA radical EPR in ASA-PGHS-2 (Table I). In the latter case, it would be the presence of an acetyl group on Ser-516 that restricts the conformational space accessible to the radical intermediate, resulting in an allyl radical rather than the pentadienyl radical observed with unmodified PGHS-2. This perturbation of the conformation of bound AA in ASA-PGHS-2 is presumably also the reason for the generation of lipoxygenase products rather than prostanoids. Our EPR data suggest that three possible carbon-centered radical intermediates may be formed, as illustrated in Fig. 7; they are a pentadienyl radical and two twisted allyl radicals with electron delocalization occurring over either C11-C13 or C13-C15, leading to either 11-HPETE or 15-HPETE, respectively, in the presence of oxygen. Both products are in fact formed by ASA-PGHS-2, with 15-(R)-HPETE as the predominant product (25).
Formation of 15-(R)-HPETE by ASA-PGHS-2 is initiated by abstraction of the 13-pro S hydrogen atom, identical to the first step of the cyclooxygenase reaction in PGHS-1 and PGHS-2 (25,35). Rotation of the side arm of AA around the C13-C14 bond in response to acetylation of Ser-516 has been proposed to account for the reversed stereochemistry of the oxygenation at C15 (35). If side arm rotation occurs, then the pentadienyl radical we observe in ASA-PGHS would not have two exo protons at C11 and C15 as in PGHS-2 and PGHS-1. Instead it would have one exo (C11) and one endo proton (C15), and the hydrogen coupling constants for the entire conjugated system would be somewhat different. However, these changes are minor in the reported spectra of pentadienyl radicals in solution with the coupling constants of all protons in both isomers within one gauss (44). Such changes will not be easily distinguished from the pentadienyl radical EPR observed for untreated PGHS-2 due to the large line width in powder spectra such as those reported here. The changes in product profile seen with the PGHS-1 V439L mutant and other cyclooxygenase site mutants have also been attributed to perturbations of the conformation of bound AA (4,39). It is worth noting that most of these mutants yield products that lack the endoperoxide and the cyclopentane structural elements, suggesting that the AA conformation leading to PGG 2 formation is easily disrupted by active site modifications. More recent examination of the Ser-530 and Val-349 mutants in both COX-1 and COX-2 further substantiates the critical role of these residues in conferring the stereoselectivity of the oxygenation at C15 (40).
Isotope Effects on AA Radical Conformation-In both the current study (Fig. 1) and our previous single-turnover EPR studies (16 -18), a 5-line pentadienyl radical EPR was always obtained when d 8 -AA was reacted with PGHS-1, PGHS-2, or ASA-PGHS-2 under the conditions used in the present study. In contrast, unlabeled AA leads to a putative distorted allyl AA radical in PGHS-1 and ASA-PGHS-2 and a pentadienyl radical in unmodified PGHS-2 (Fig. 6). If the hypothesis of an allyl radical is correct, then the greater tendency to form a constrained substrate radical with AA than with d 8 -AA suggests that the proteins have a different interaction with unlabeled substrate, stabilizing the thermodynamically disfavored twisted allyl radical. To place the required stabilization in a more quantitative context, the barrier of rotation to take one of the -systems out of conjugation has been reported to be 11.7 kcal/mol for the parent pentadienyl radical in solution (48). The initial penalty for rotation of a 1,5-disubstituted radical with both substituents in the endo position may be somewhat lower than for the parent pentadienyl radical because of relief of unfavorable steric interactions (1,3-strain), but it is likely to still be considerable. In addition, the twisting of the allyl radical will require an energetic penalty. In our theoretical 1,3dimethyl allyl radical model the energy of the conformer with a dihedral angle of 45°lies 11 kcal/mole above the planar conformation and at a 90°angle; the calculated energy was 22 kcal/ mole above the planar radical. This latter value is much larger than that reported for the rotational barrier in the parent allyl FIG. 9. Theoretical spin-density distribution diagram for a 1,3-dimethyl allyl radical. The spin density at each carbon atom was calculated as described under "Materials and Methods" with the indicated values of the dihedral angle subtended by C2-C3-C4-C5 (rotation around the C3-C4 bond). Each data set was fitted to a sine function, shown by the solid lines. radical (17.4 kcal/mol), obtained using B3LYP density functional theory calculations (49). The larger energy barrier may be a consequence of two methyl substituents on carbons 2 and 4 in the 1,3-dimethylallyl radical model. An increase in the stabilization energy of the allyl radical by methyl substitution has indeed been reported (50). Unfortunately, more accurate experimental measures of stabilization energy developed since this report (51) have not been applied to allyl radicals substituted with methyl groups to verify this conclusion.
Deuterium isotope effects in the absence of covalent bond breaking and formation have been extensively documented (42). Noncovalent deuterium isotope effects are primarily attributed to the shorter bond length of C-D versus C-H as well as to the changes in molecular volume and polarity or polarizability. Increasing deuterium substitution in hydrocarbons usually increases their polarity and, thus, hydrophilicity, and our HPLC results support such an effect in AA (Fig. 8). The difference in affinity for the C 18 stationary phase between AA and d 8 -AA may be relevant considering the hydrophobic nature of the cyclooxygenase substrate binding site, which mainly consists of nonpolar amino acid residues (20 -23). One interpretation is that the lower hydrophobicity of d 8 -AA compared with AA provides decreased interaction with the protein and, thus, increased flexibility once the fatty acid is bound. This decreased constraint favors formation of a pentadienyl d 8 -AA radical, whereas stronger binding interactions and higher constraint with unlabeled AA favor an allyl radical, especially when the binding site becomes more restricted, as in PGHS-1 or ASA-PGHS-2 (20 -23, 41). However, this interpretation finds little support in the steady-state kinetic data, where single or multiple deuteration showed little effect on the arachidonate K m values or the V max of the cyclooxygenase activity. Clearly, the lack of change of K m does not rule out binding affinity changes because the cyclooxygenase K m value may be dominated by factors other than binding; direct binding measurements are required to resolve this issue.
In summary, the present results show that the peroxideinduced tyrosyl radical in ASA-PGHS-2 is capable of oxidizing arachidonate to a carbon-centered fatty acid radical. This provides the first direct support for a tyrosyl radical mechanism of lipoxygenase catalysis in ASA-PGHS-2. Computer simulations of the EPR spectra suggest plausible structures for the arachidonyl radicals formed in ASA-PGHS-2 and in PGHS-1; these include a distorted allyl radical and a pentadienyl radical.