Evidence for a Peroxide-initiated Free Radical Mechanism of Prostaglandin Biosynthesis *

High levels of NaCN (20 to 250 mM) were required to inhibit cyclooxygenase catalysis and cause extended lag periods (up to 1.6 min), whereas CO failed to inhibit catalysis. T h i s NaCN inhibition was easily overcome by endogenous or exogenous hydroperoxides. Added hydroperoxides acted to eliminate lag periods without undergoing net conversion to other chemical species. In addition, experiments with glutathione peroxidase inhibition showed that hydroperoxides were essential not only in the early phases, but throughout catalysis. In spectrophotometric experiments, NaCN formed a complex with ferriheme cyclooxygenase ( K d = 1.3 mM) and inhibited hydroperoxide interaction with t is form of the enzyme. Phenolic antioxidants, only slightly extended lag periods while inhibiting oxygenation rates more than 50%. Low levels of phenol (which is normally stimulatory) or a-naphthol when combined with NaCN or glutathione peroxidase (agents which interfere with peroxide activation) resulted in potent synergistic inhibition with long lag times. A mechanism consistent with all of the above properties of cyclooxygenase has been elucidated. Further mechanistic explanation was sought for reaction-catalyzed self-inactivation of cyclooxygenase. This phenomenon could not be explained simply by heme lability, or cyclooxygenase sensitivity to destruction by ambient hydroperoxides. Rather, it appears to involve a destructive reaction intermediate intrinsic to the cyclooxygenase mechanism.

the inhibitory action of glutathione peroxidase on crude (9)(10)(11) and purified (12) ovine cyclooxygenase. The removal of peroxide by glutathione peroxidase caused a lag period in reaching maximum catalytic activity in a manner similar to that caused by added cyanide (13)(14)(15) or diethyldithiocarbamate (15). The lag times correspond to the time needed for the hydroperoxide product activator to accumulate at levels sufficient to exert its effect. Those lags were shortened or eliminated by the addition of PGG, (16) and a variety of other hydroperoxides including contaminants in arachidonate stock solutions (15,17).
Cyclooxygenase catalysis appears to involve radical intermediates, since crude preparations of vesicular gland cyclooxygenase were strongly inhibited by a variety of antioxidants (18)(19)(20). Also, a pronounced EPR signal at g = 2 has been observed (19) that may reflect either cyclooxygenase intermediates or derivatives of products. Furthermore, nonenzymic model systems have yielded prostaglandin stereoisomers from free radical reactions (21)(22)(23).
In an attempt to subdivide the cyclooxygenase reaction mechanism, we examined 11,14-eicosadienoic acid, a cyclooxygenase "half-substrate" which only undergoes oxygenation at carbon 11 (12). The results showed that all major features of cyclooxygenase catalysis (heme requirement, peroxide activation, and self-destruction) were part of the oxygenation mechanism at carbon 11, and there was no evidence for a heme-free half-reaction resembling that of soy lipoxygenase.
During catalysis, cyclooxygenases from ovine (10,16,24,25) or bovine (1) vesicular glands become inactivated within a few minutes. More studies are needed to help explain or circumvent this "self-destruction" which causes an intrinsic limitation to the extent of reaction.
The current studies with purified enzyme have been designed to interpret in detail the role of heme, peroxide, and free radicals in cyclooxygenase-catalyzed product formation and self-destruction.2 velocity. However, if 8 p~ 15-hydroperoxy arachidonic acid  was present with the 250 mM NaCN, the lag phase and the decrease in optimal velocity were abolished (dashed line, Fig. 1A). Lower concentrations of 15-HPETE (0.3 to 0.9 p~) also effectively shortened the lag phase caused by 250 mM NaCN (see Fig. 2

below).
As in earlier studies (13), a second addition of enzyme (Fig.  lA, arrow), after self-destruction had halted the prior reaction, re-initiated catalysis without a lag phase. Thus, the accumulated activator peroxide from the first reaction was able to overcome the lagging effect of 250 mM NaCN. Cyanide also caused a significant increase in the total extent of reaction (up to 2.7-fold). The maximum number of turnovers achieved per enzyme prior to destruction increased from 75 to 200 (on the basis of 2 02/turnover). Extraction and quantitative analysis of the products (see "Materials and Methods") indicated that 1 mol of arachidonic acid was consumed/l.8 mol of 0, (average of five experiments), both in the presence and absence of NaCN (250 mM). Thin layer chromatography in ethyl acetate/isooctane/acetic acid (50:50:0.5) indicated that the major product isolated after a 2-to 8-min reaction in the presence or absence of NaCN (250 mM) was PGG, (Kf = 0.5).
Other products co-migrated with standard PGH, ( K F = 0.32), hydroxy acids ( R F = O X ) , and an unresolved mixture of PGEL, I'GFZ,,, and PGI):! (Rp = 0.16 to 0.30). The identity of PGG, was confirmed by demonstrating that it was completely reduced by triphenylphosphine (26) to a compound which comigrates with PGF2,,. Cyanide, up to 250 mM, failed to inhibit Mn"' protoporphyrin-cyclooxygenase (result not shown), thus affirming that the lagging effect of cyanide depends on a specific interaction with the heme iron.
Low levels of 15-HPEL'E (ROOH), which effectively shortened the lag caused by 250 mM NaCN, were recovered unchanged after catalysis (Fig. 2). Itadioscans of ether extractable material (pH 3.0) on thin layer chromatograms showed no evidence for oxygen-oxygen bond cleavage, since no discernible peaks of radioactivity appeared in the hydroxy acid region, or any other region except that corresponding to lipid hydroperoxide.
Considering the level of enzyme (0.12 p~) and the fact that 0.3 /*M 15-HPETE only partially shortened the lag time, any obligatory conversion of 15-HPETE (if it had occurred) might be expected to be significant.
Antioxidant Inhibition of Cyclooxygenase-In contrast to the effect of NaCN, increasing amounts of the antioxidant, a-naphthol (Fig. 1B) only slightly increased the lag phase (lag 5 2 0 s), while causing almost complete inhibition of the optimal velocity. A reaction catalyzed in the same chamber by a second addition of enzyme (after 1.4 min) was also inhibited by 72 p~ u-naphthol (not shown). Furthermore, the addition of 13 p~ hydroperoxy arachidonic acid (15-HPETE) failed to abolish the effects of 72 PM a-naphthol (not shown).
At concentrations above 1 to 2 mM, phenol, like a-naphthol, also inhibited cyclooxygenase (Fig. 3 at the indicated concentrations. The effect of phenol on Mn"' protoporphyrin cyclooxygenase (0.041 ,UM) prepared as previously described (16) was also determined (0---0). agents (tryptophan, methional, and tyrosine) resemble phenol in their ability to stimulate cyclooxygenase and act as prostaglandin hydroperoxidase co-substrates (1,25,27), they did not inhibit the cyclooxygenase reaction at the concentrations tested. Mn"' protoporphyrin-cyclooxygenase, which has no prostaglandin hydroperoxidase activity (6,16), was inhibited by phenol without exhibiting the stimulation at low concentrations.
Synergistic Inhibition of Cyclooxygenase-When a-naph-tho1 and cyanide were included in the same reaction mixture, synergistic inhibition was observed (Fig. 4). Cyanide alone (10 mM) had little effect on optimal reaction rates or lag periods (as predicted from Fig. l), and a-naphthol alone needed to be present at 55 to 60 p~ to give 50% inhibition. However, 10 mM NaCN caused pronounced inhibition of the optimal velocity with increasing amounts of a-naphthol (50% inhibition with 20 p~ a-naphthol, Fig. 4A) and caused lengthy lag periods (up to 143 s, Fig. 7B).
Another antioxidant, phenol, also gave highly synergistic inhibition with NaCN (Fig. 5). Phenol alone (0.67 mM) or NaCN alone (0 to 10 m) caused minimal inhibition of optimal velocity (0.67 mM phenol stimulates) and only slightly extended the lag periods. However, the combined agents inhibited the optimal velocity ( Fig. 5A) and extended the lag time from 9 s to 110 s (Fig. 5B). This lag was prevented when 9 p~ 15-HPETE was included in the reaction mixture (lag time remained at 9 s; data not shown). The Mn"'-protoporphyrin cyclooxygenase was not inhibited by 10 II~M NaCN in the presence of 0.67 m phenol, again a f f i i n g the specificity of NaCN inhibition for the heme form.
In addition to acting synergistically with NaCN (Figs. 4 and 5) the antioxidants, phenol (0.67 mM and 1.33 mM) and anaphthol (20 PM), acted synergistically with glutathione peroxidase to cause extended lag periods (Fig. 6B). The glutathione peroxidase alone had minimal lagging or inhibitory effect, and phenol alone was stimulatory. Control experiments showed that added glutathione peroxidase was ineffective if its essential co-substrate, GSH, was either omitted or blocked with N-ethylmaleimide. Additional control assays demonstrated that glutathione peroxidase activity was not influenced by added phenol.
Inhibition of Cyclooxygenase Reactions Already in Progress-Reactions which were partially inhibited by 150 mM NaCN were completely inhibited by glutathione peroxidase plus GSH, whether the glutathione peroxidase was added after 6, 26, or 35 s (Fig. 7). The same amount of glutathione peroxidase (12.5 K units/&) was noninhibitory when added The reaction rates (loluer curves; dO,/dt) and oxygen consumption (upper curues) were determined as described under "Materials and Methods." without cyanide. Glutathione (GSH) had no effect by itself. Reactions catalyzed by heme-cyclooxygenase or manganese protoporphyrin-cyclooxygenase, each in the presence of 0.67 mM phenol, were also inhibited by glutathione peroxidase together with GSH added after ?h to 2.0 min (not shown).
Furthermore, the addition of NaCN (10 m) in the presence of 0.67 mM phenol to a cyclooxygenase reaction which had already reached optimal velocity, caused a rapid 82% decrease in velocity (not shown).
The Effect of Carbon Monoxide-The addition of carbon monoxide (displacing either 35% or 69% of the air in airsaturated buffers) to cyclooxygenase in the presence of phenol (0.67 mM) and glutathione (0.67 mM) resulted in no inhibition. When glutathione peroxidase (10 K units/&) was also present, enzyme activity was inhibited 41% (relative to no glutathione peroxidase). However, 16% and 48% CO caused not more than 2% further inhibition relative to the glutathione Spectral Properties of Cyclooxygenase-For native cyclooxygenase, a single broad Soret peak is the only feature characteristic of the visible spectrum. Dithionite caused a shift of the Soret peak (409 nm) to a longer wavelength (427 nm) and produced additional spectral features in the 520 to 580 nm region (peak a t 560 nm, shoulder at -540 nm). Since dithionite reduces oxygen to hydrogen peroxide, and cyclooxygenase is extremely labile in the presence of peroxide, it was essential to perform dithionite reductions anaerobically.
Titration of native cyclooxygenase with cyanide caused the heme Soret peak to shift from 409 toward 423 nm (Fig. 8) while a trough appeared at 510 nm and a peak at 540 nm. NaCN, perhaps due to a slight chaotropic effect after extended incubation. The spectrum of neither the native cyclooxygenase nor a preparation partially formed in a complex with cyanide (2.5 mM NaCN) was perturbed by phenol (1.0 mM), GSH (0.5 mM) or glutathione peroxidase (44 K units/ml). The ability of NaCN to prevent cyclooxygenase-hydroperoxide interaction (suspected from the results in Fig. lA) was examined further spectrally. As shown in Fig. 9, 17 PM 15-HPETE caused a rapid loss of absorbance in the Soret region (monitored at 415 nm), which was slowed significantly in the presence of 20 mM NaCN.
T o gain insight into the role of heme during catalysis, the Soret peak of cyclooxygenase was monitored after the addition of substrate (Fig. 10). The Soret peak disappeared within 2 to 5 min and no appreciable amount of reduced enzyme (ferroheme form) accumulated since no significant shift to a longer wavelength was observed during that time.

DISCUSSION
Hydroperoxide Interaction with Ferriherne-The spectral properties of cyclooxygenase heme, its reduced form, and its unreduced cyanide complex are typical of ferriheme proteins. Similar spectra were found for ovine cyclooxygenase by Van der Ouderaa et al. (8).
Cyanide apparently can compete with hydroperoxide at the ferriheme site, since cyanide protected the ferriheme enzyme against destruction by hydroperoxide (Fig. 9). The mechanistic importance of this competition is discussed below.
In general, the appearance of a lag during cyclooxygenase catalysis is indicative of a deficiency in the level of hydroperoxide which is slowly overcome as product hydroperoxide accumulates (11,13,15,16). Added hydroperoxides at 0.2 to 1.0 PM or less, effectively eliminated lag phases (15-17 and Fig. 2). Accumulation of such stimulatory levels of endogenous hydroperoxide is probably not hindered by the prostaglandin hydroperoxidase activity of the enzyme (see below) since the apparent K,,, for hydroperoxides in that process (for the bovine enzyme) is 12 to 30 PM (27), 2 orders of magnitude above that for activation of cyclooxygenase.
The inhibition caused by cyanide (Figs. lA and 5) has the characteristics of peroxide deficiency, since the lags were shortened by the accumulated product peroxide from a previous reaction, or by the addition of exogenous hydroperoxide. Thus, as peroxide product accumulates, the enzyme becomes less sensitive to cyanide and even 250 mM NaCN caused less than 50% inhibition of the optimal reaction rate.
Sodium cyanide might have been expected to inhibit at lower concentrations based on its K,, (1.3 mM total cyanide). However, that determination was not made in the presence of high affinity peroxide which apparently competes with cyanide during catalysis. In contrast to the cyanide insensitivity of cyclooxygenase, other known heme-containing dioxygenases are readily inhibited by cyanide; tryptophan dioxygenase (58%, 0.4 mM (29)) and indoleamine dioxygenase (loo%, 0.1 mM (30)). An important overall conclusion from the spectral and kinetic studies is that peroxide activation occurs at a cyanide-sensitive site which is most likely ferriheme.
The inability of cyanide to inhibit manganese protoporphyrin cyclooxygenase is in agreement with the general failure of cyanide to form complexes with manganese protoporph.yrin forms of cytochrome c peroxidase, horseradish peroxidase, and myoglobin (31). Despite the lack of inhibition by NaCN, activation by hydroperoxide is part of the manganese protoporphyrin cyclooxygenase mechanism since the manganese form of the enzyme exhibits both inhibition by glutathione peroxidase (see "Results") and peroxide-reversible lags (16,17). The manganese protoporphyrin enzyme is a useful model system for cyclooxygenase studies, since it is free of the prostaglandin hydroperoxidase activity (6, 16) normally associated with native cyclooxygenase (1,2,27). Thus, the requirement for hydroperoxide activator can be regarded as an intrinsic aspect of the cyclooxygenase mechanism independent of the hydroperoxidase activity.
Tryptophan oxygenase (29) exhibits a lag phase that is prevented by an initial reductive activation with peroxide. However, once activated, this enzyme has no further need for peroxide. The essential difference for cyclooxygenase is that peroxide is needed continuously throughout catalysis, since at any time, the reaction may be stopped, or slowed by the addition of glutathione peroxidase or NaCN. Equation 1 summarizes the role of hydroperoxide (ROOH) and the effect of cyanide on catalysis. This equation demonstrates how either the removal of ROOH (e.g by glutathione peroxidase) or the removal of free E(Fe"') (e.g. as the cyanide complex) could reversibly limit the activation of cyclooxygenase.   (18-20). Since the inhibition by antioxidants at higher concentrations did not cause significant lags and was not reversed by added peroxide, it presumably occurred by a mechanism different from that for either cyanide or glutathione peroxidase. Nonetheless, it was important to consider the possibility that the antioxidants were effective inhibitors because they removed essential hydroperoxides by acting as co-substrates for prostaglandin hydroperoxidase. This possibility was ruled out with the finding that many other hydroperoxidase cosubstrates were not inhibitory to cyclooxygenase (Fig. 3). Also, the manganese protoporphyrin form of cyclooxygenase (which has no hydroperoxidase activity) was inhibited by phenol. It thus appears most likely that the antioxidants inhibit by removing intermediate radicals essential to the cyclooxygenase mechanism. The synergistic increase in sensitivity to antioxidants caused by glutathione peroxidase or NaCN (agents which interfere with peroxide activation) indicates that the production of antioxidant sensitive radicals depends indirectly on peroxide levels. Since antioxidants such as phenol and hydroquinone have often been included in cyclooxygenase reaction mixtures their influence must be considered when evaluating the previously described sensitivity of cyclooxygenase to low levels of cyanide (8,(13)(14)(15)32) or glutathione peroxidase (10, 12, 14). In this regard, the prior distinction between E , (glutathione peroxidase-sensitive, phenol-activated) and E,, (glutathione peroxidase-insensitive, without phenol) forms of cyclooxygenase (10,33) can now be reinterpreted in terms of phenol-glutathione peroxidase synergism accounting for the high sensitivity of "E," to glutathione peroxidase that is not found with the basal "E;' form.
Intermediate Forms Reacting with Oxygen-The failure of carbon monoxide to inhibit significantly is in agreement with a similar lack of inhibition reported for the bovine vesicular gland cyclooxygenase (27). This result indicates that a ferroheme form of cyclooxygenase, if it occurs, is not available for binding ligands such as CO or O2 during catalysis. Furthermore, no evidence for liganded oxygen intermediates has yet been found since the heme spectrum did not appear to undergo major shifts during catalysis (other than the disappearance due to heme loss). Tryptophan dioxygenase is inhibited by CO (34) and forms a complex with O2 (35) as also does indoleamine dioxygenase (36). Thus, cyclooxygenase again differs from the other known heme-containing dioxygenases.
As an alternative t,o enzyme activation of oxygen, it appears more feasible to propose that the fatty acid is activated to form a radical which then reacts with oxygen. This would be analogous to the widely accepted mechanism (37) for the auto-oxidation of organic compounds ( R H ) to produce hydroperoxide (ROOH) as shown by Equations 3 to 5.
(- Possible Mechanism for Cyclooxygenase-Equations 1 to 5 can be combined into an overall mechanism (Scheme 1) which we propose to account for peroxide triggering of ferriheme catalysis; the formation of essential radical intermediates in a manner dependent on peroxide levels; and the activation of fatty acid for reaction with oxygen in a way not inhibitable by CO. The diagram illustrates hydroperoxide and ferriheme interaction to form a hydroperoxy radical (which probably is enzyme-bound) in an initiation step for the overall reaction. This radical, with the help of the orienting enzyme then propagates the reaction by abstracting the 13-S-hydrogen (38) from arachidonic acid ( R H ) . Then, following the mechanism outlined by Samuelsson (39), oxygen reacts with the alkyl radical at carbon 11, cyclization occurs, and another mole of O2 reacts at carbon 15. The resulting 15-hydroperoxy radical may abstract hydrogen from ROOH to form the end product, PGGn, and start another cycle. The hydroperoxy radical, rather than the alkoxy radical, is proposed to propagate the reaction since no evidence was found for oxygenoxygen bond cleavage during peroxide activation and there is much precedent for involvement of the hydroperoxy radical in auto-oxidation mechanisms (see below).
The continuous need for hydroperoxide is consistent with an unfavorable equilibrium for the initiation (E"' HOOR s E".OOR). The reduction potential for ROO'/ROOH is probably around 0.7 V at pH 8.5, assuming that it is analogous to that for HOO'/HOO-(40). Hemoproteins (ferric/ferrous) generally have reduction potentials of 0.05 to 0.3 V. Thus, the initiation of cyclooxygenase catalysis may be unfavorable by 0.5 to 0.65 V (AG = +9 to 15 kcal). However, the subsequent steps, such as the binding of fatty acid may yield 10 to 20 kcal/mol (41) and ''pull" the reaction toward completion. Termination may occur by a reverse of the initiation reaction or by the removal of essential radicals by antioxidants ( A H ) .
If glutathione peroxidase (which removes peroxide) or NaCN (which inhibits peroxide-heme interaction) are present simultaneously with antioxidant, then synergistic inhibition results as discussed above. The mechanism in Scheme 1 leads to better understanding of the possible regulatory role of peroxides previously outlined (15). When there is limited accumulation of hydroperoxide in the environment of the cyclooxygenase (as might be expected for in vivo conditions), then phenolic agents are expected to be far more potent as cyclooxygenase inhibitors than many in vitro assays would suggest. This reasoning may explain the anti-inflammatory effects of 2-aminomethyl-4-t-butyl-6-iodophenol (MK-447) and phenol despite their having stimulated cyclooxygenase in vitro (42). Variability in local peroxide levels also may account for the variety of effects of acetamidophenol (Tylenol) on cyclooxygenase that is encountered in different tissues and with different assay systems (43-45).
In general, the cyclooxygenase mechanism outlined above parallels the widely accepted mechanism for auto-oxidation of organic compounds. In this regard, auto-oxidation of arachidonate has yielded stereoisomers of common prostaglandins and is postulated to proceed through endo-peroxide intermediates (21,23,46). Mechanisms proposed for another fatty acid dioxygenase, soybean lipoxygenase, have some notable similarities to the mechanism proposed here. Hydroperoxide participates in a reaction which forms fatty acid radicals (47) which then react with 0, (48). However, a continuous need for peroxide (10,49) has not yet been widely recognized, and several different mechanisms involving ferric and ferro forms of lipoxygenase (which contains non-heme iron) have been proposed (50,51). The mechanism for soybean lipoxygenase isoenzyme-2 may have even greater similarity to that for cyclooxygenase, since both enzymes form prostaglandins (52).
Self-destruction during Catalysis-There are four potential explanations that we considered for the mechanism of cyclooxygenase self-destruction. 1) Labile heme is irreversibly destroyed, since heme proteins often lose heme in the presence of peroxide (e.g. indoleamine dioxygenase (36). 2) Cyclooxygenase is inactivated due to the action of product peroxides, since the enzyme is unstable in the presence of peroxide (2,
Since we found that the rate of heme loss was significantly slower than self-destruction, Explanation 1 above is not adequate to explain the mechanism of self-destruction. Cyclooxygenase activity was lost in the presence of hydroperoxide, as predicted in Explanation 2 above. However, since phenol protected against this destruction, but not against the selfcatalyzed destruction, the two processes are not identical. The third hypothesis for self-destruction mentioned above appears insufficient, since manganese protoporphyrin cyclooxygenase also undergoes inactivation during catalysis despite the absence of hydroperoxidase activity (16). Finally, although avail-able peroxides can contribute to enzyme inactivation, the selfcatalyzed destruction is a separate feature intrinsic to cyclooxygenase catalysis and probably caused by reaction intermediates (Alternative 4 above). The increased extents of reaction seen with high levels of cyanide (Fig. 1A) probably reflect the ability of cyanide to protect against the peroxide-mediated destruction leaving more enzyme to proceed via the selfcatalyzed route of destruction.
Stimulation of Catalysis-Phenols, indoles, and methional (e.g. see Fig. 3), in addition to compounds such as lipoic acid (54, 55). diethyldithiocarbamate (12), and sodium iodide (55) share the ability to stimulate cyclooxygenase activity by 2-to 3-fold. Significantly, all such agents also promote hydroperoxidase activity as co-substrates (1,12,25,27,55). The correlation between stimulation of cyclooxygenase and activity with hydroperoxidase is more firmly established by the finding that the manganese protoporphyrin cyclooxygenase without hydroperoxidase activity was not stimulated by phenol (Fig. 3 ) .
Because phenol did not significantly alter the rate of selfdestruction (Figs. 11 and 12), the stimulation by phenol does not appear to be caused by prevention of destruction as suggested by others (2,25). A more likely explanation is that during catalysis, cyclooxygenase may occasionally develop nonproductive oxidized forms (Scheme 1) analogous to Compound I or I1 of horseradish peroxidase (56). The stimulating agents (which are all potential reducing agents) would then reduce the enzyme back to an active form of cyclooxygenase in a manner similar to that proposed for stimulation of peroxidase by phenolic agents (57). At high levels of peroxide, these nonproductive forms could be more favored, and a stimulating agent such as phenol would be less effective (as seen in Fig. 12). Not having hydroperoxidase activity, the manganese protoporphyrin cyclooxygenase does not develop these oxidized, nonproductive forms, and thus is not stimulated by agents that can remove them.
The present proposed mechanism now provides satisfactory explanation of all the major properties of cyclooxygenase activity. This unification of many diverse observations within a consistent, integrated paradigm will be helpful in interpreting in finer detail the nature of physiological and pharmacological regulation of prostaglandin biosynthesis.