The Identification of Primary Sites of Superoxide and Hydrogen Peroxide Formation in the Aerobic Respiratory Chain and Sulfite Reductase Complex of Escherichia coli *

The fitness of organisms depends upon the rate at which they generate superoxide (O·̄2) and hydrogen peroxide (H2O2) as toxic by-products of aerobic metabolism. In Escherichia coli these oxidants arise primarily from the autoxidation of components of its respiratory chain. Inverted vesicles that were incubated with NADH generated O·̄2and H2O2 at accelerated rates either when treated with cyanide or when devoid of quinones, implicating an NADH dehydrogenase as their source. Null mutations in the gene encoding NADH dehydrogenase II averted autoxidation of vesicles, and its overproduction accelerated it. Thus NADH dehydrogenase II but not NADH dehydrogenase I, respiratory quinones, or cytochrome oxidases formed substantial O·̄2 and H2O2. NADH dehydrogenase II that was purified from both wild-type and quinone-deficient cells generated ∼130 H2O2and 15 O·̄2 min−1 by autoxidation of its reduced FAD cofactor. Sulfite reductase is a second autoxidizable electron transport chain of E. coli, containing FAD, FMN, [4Fe-4S], and siroheme moieties. Purified flavoprotein that contained only the FAD and FMN cofactors had about the same oxidation turnover number as did the holoenzyme, 7 min−1 FAD−1. Oxidase activity was largely lost upon FMN removal. Thus the autoxidation of sulfite reductase, like that of the respiratory chain, occurs primarily by autoxidation of an exposed flavin cofactor. Great variability in the oxidation turnover numbers of these and other flavoproteins suggests that endogenous oxidants will be predominantly formed by only a few oxidizable enzymes. Thus the degree of oxidative stress in a cell may depend upon the titer of such enzymes and accordingly may vary with growth conditions and among different cell types. Furthermore, the chemical nature of these reactions was manifested by their acceleration at high temperatures and oxygen concentrations. Thus these environmental parameters may also directly affect the O·̄2 and H2O2 loads that organisms must bear.

treated with cyanide or when devoid of quinones, implicating an NADH dehydrogenase as their source. Null mutations in the gene encoding NADH dehydrogenase II averted autoxidation of vesicles, and its overproduction accelerated it. Thus NADH dehydrogenase II but not NADH dehydrogenase I, respiratory quinones, or cytochrome oxidases formed substantial O 2 . and H 2 O 2 . NADH dehydrogenase II that was purified from both wild-type and quinone-deficient cells generated ϳ130 H 2 O 2 and 15 O 2 . min ؊1 by autoxidation of its reduced FAD cofactor.
Sulfite reductase is a second autoxidizable electron transport chain of E. coli, containing FAD, FMN, [4Fe-4S], and siroheme moieties. Purified flavoprotein that contained only the FAD and FMN cofactors had about the same oxidation turnover number as did the holoenzyme, 7 min ؊1 FAD ؊1 . Oxidase activity was largely lost upon FMN removal. Thus the autoxidation of sulfite reductase, like that of the respiratory chain, occurs primarily by autoxidation of an exposed flavin cofactor. Great variability in the oxidation turnover numbers of these and other flavoproteins suggests that endogenous oxidants will be predominantly formed by only a few oxidizable enzymes. Thus the degree of oxidative stress in a cell may depend upon the titer of such enzymes and accordingly may vary with growth conditions and among different cell types. Furthermore, the chemical nature of these reactions was manifested by their acceleration at high temperatures and oxygen concentrations. Thus these environmental parameters may also directly affect the O 2 . and H 2 O 2 loads that organisms must bear.
The discovery of superoxide dismutase (SOD) 1 in 1969 (1) was the first indication that aerobic organisms are threatened by superoxide (O 2 . ). SOD catalyzes the dismutation of superoxide (Reaction 1) and, in combination with catalase (Reaction 2), helps clear the cell of reactive oxygen species.
SOD was found to be virtually ubiquitous among aerobic organisms, suggesting that O 2 . might be an unavoidable by-product of metabolism in air. This idea was extended to develop the hypothesis that oxygen toxicity might be generally mediated by intracellular O 2 . (2). The molecular details that underpin this idea (the intracellular sources and targets of O 2 . ) were unresolved.
Within the past 10 years many details of oxygen toxicity have been revealed. In 1986 Carlioz and Touati (3) reported the properties of a mutant strain of Escherichia coli that lacked both of its cytosolic isozymes of SOD. The mutant grew normally in the absence of oxygen; however, in aerobic medium it exhibited requirements for branched chain, aromatic, and sulfurous amino acids, an inability to grow on non-fermentable carbon sources, and a high rate of spontaneous mutagenesis (4). These same traits were elicited when SOD-proficient wildtype strains were exposed to hyperbaric oxygen (5), suggesting that in these conditions O 2 . formation must be accelerated enough to overwhelm the cellular defenses. Therefore these observations supported the original model of oxygen toxicity. The root cause of the branched chain auxotrophy was tracked to the ability of O 2 . to inactivate dihydroxy-acid dehydratase, an enzyme midway in this pathway (6). O 2 . does so by oxidizing and destabilizing the [4Fe-4S] cluster that acts as a Lewis acid during catalysis (7,8). Iron dissociates from the oxidized cluster, causing a complete loss of activity. The requirement for fermentable carbon sources apparently stems from similar damage to aconitase and fumarase, which also belong to the [4Fe-4S] dehydratase class (9,10). The high rate of mutagenesis is linked to the same damage: the iron that is lost from destabilized clusters floats freely into the cytosol where it catalyzes DNA oxidation by H 2 O 2 (11)(12)(13). Thus to date all the well understood deficits of SOD mutants arise from this single type of lesion. In higher organisms O 2 . toxicity is linked to similar damage to mitochondrial aconitase (14). Less clear is the mechanism by which O 2 . is generated in cells in the first place. Molecular oxygen is actually a poor chemical oxidant, because its triplet state constrains it to accept one electron at a time from potential donors. Because biological electron carriers such as NAD(P)H resist the loss of a single electron, and the oxidizing potential of the O 2 (21). Although fumarate reductase contains three iron-sulfur clusters, its autoxidation, like that of xanthine oxidase, occurred exclusively from the flavin. One purpose of the present study was to determine whether that is a general rule. Sulfite reductase comprises a second, albeit self-contained, electron transport chain in E. coli. Like the respiratory chain, sulfite reductase transfers electrons through FAD, FMN, [4Fe-4S], and siroheme, which are all good univalent redox enzymes and could plausibly transfer an electron to oxygen. In fact its reactivity with oxygen has been well documented (22)(23)(24). By comparing the sites, rates, and valencies of electron transfers from these carriers to oxygen, we wished to identify characteristics that may predispose redox enzymes to generate O 2 . and/or H 2 O 2 . If these oxidants are preponderantly generated by only a few enzymic sources, then variations in the titer of such enzymes (from one growth condition to another or one organism to another) might substantially change the degree of endogenous oxidative stress.
Strains-Strains used in this study are listed in Table I. Mutant strains were constructed by P1 transduction. All comparisons in this work were between congenic strains. Mutant fnr alleles were co-transduced with a linked tetracycline-resistant Tn10; inheritance of the fnr allele was demonstrated by the inability of mutants to grow anaerobically on glycerol/nitrate medium (25). In a ubi men (quinoneless) background this screen cannot work, so the allele from putative transductants was transduced back out into a quinone-proficient background and screened. The frd deletion was co-transduced with a Tn10 and screened for growth on glycerol/fumarate. ndh::cam alleles were selected by chloramphenicol. The absence of NADH dehydrogenase II was screened by enzymatic assay for NADH:plumbagin oxidoreductase activity after vesicles were incubated at pH 7.8 to inactivate the NADH dehydrogenase I complex (26). The nuo mutant allele was selected by the linked Tn10. Potential NADH dehydrogenase I-deficient transductants were screened for deamino-NADH:plumbagin oxidoreductase activity (27).
Growth and Media-Vesicles were prepared from cells grown in either LB medium supplemented with 0.2% glucose, minimal A glucose medium (28), or minimal A supplemented with 1% casamino acids and 0.5 mM tryptophan. Standard antibiotic concentrations were used for plasmid maintenance and P1 transductions (29). Media for anaerobic cultures were degassed by autoclaving and equilibrated in an anaerobic chamber (10% H 2 , 5% CO 2 , 85% N 2 ) for at least 24 h before inoculation.  (31). The difference in protocols reflects our evolving strategies to induce maximal expression of the sulfite-reductase operon. Enzyme Purification-Respiratory vesicles were customarily prepared as described previously (19). When we sought to avoid the inactivation of NADH dehydrogenase I, cells were lysed, and vesicles were prepared in 50 mM MES buffer, pH 6.0, containing 10% glycerol (26). These conditions were not necessary for activity, however, so reactions with respiratory vesicles and enzymes were conducted in 50 mM KP i , pH 7.8.
NADH dehydrogenase II was purified from both quinone-proficient and -deficient strains. The wild-type enzyme was purified from KM50, an ndh-overexpressing strain which is quinone-proficient but lacks cytochrome oxidases o and d. Preliminary preparations had revealed that trace amounts of cytochrome oxidase activity interfered slightly with assays of NADH oxidase activity during purification. Because cytochrome oxidase mutants grow poorly in air, KM50 was grown in 10 liters of anaerobic LB, 0.2% glucose to an A 600 of 0.3 and then shifted into air and grown to a final A 600 of 0.45. The anaerobic culture permitted a high enzyme titer to be achieved, particularly because of the loss of suppression (32) by Fnr protein. The final growth period in air ensured ubiquinone sufficiency. 6.7 g of cell pellet was suspended in 20 ml of 50 mM KP i , pH 7.8, and lysed by French press. Debris was removed by centrifugation at 15,000 rpm ϫ 20 min, and vesicles were pelleted at 100,000 ϫ g for 3 h. NADH dehydrogenase was extracted from resuspended membranes with deoxycholate and purified over hydroxyapatite as described (33). Active fractions were verified by both NADH:ferricyanide and NADH:ubiquinone-1 oxidoreductase assays. Purified fractions produced a single band on SDS gels (not shown). Samples were dialyzed against 90 volumes of 5 mM KP i , pH 7.5 and 0.1% cholate with two changes of buffer, in order to remove exogenous FAD.
Ubiquinone-deficient NADH dehydrogenase II was purified from an anaerobic culture of KM44, which lacks both menaquinone and ubiquinone. The deletion of fnr and presence of a multicopy ndh-expressing plasmid again ensured abundant synthesis of the enzyme even during anaerobiosis. Six liters of 0.2 A 600 cultures were harvested. The lack of respiratory activity in the vesicles verified that quinones were absent. NADH dehydrogenase II was extracted and purified as above.
Sulfite reductase holoenzyme was purified from JI132/pJRS102. This plasmid (34) includes cysG, which encodes an enzyme needed to generate mature siroheme for sulfite reductase and nitrite reductase, and causes abundant overproduction of sulfite reductase in sulfur-limited cells. A 2.6-g cell pellet was recovered from 4.5 liters of 0.4 A 600 culture. Cells were lysed by French press. The purification scheme of Ref. 24 was followed, including streptomycin sulfate and ammonium sulfate precipitations, Superose sizing column, and hydroxyapatite column. The enzyme activity was monitored by assay for sulfite reductase activity. Purified enzyme migrated as 56-and 61-kDa bands. The spectra of pure fractions matched that observed by other investigators (22).
Sulfite reductase flavoprotein (subunit B) was purified from DH5␣/ pcysJ. Six liters of culture were harvested at 0.25 A 600 , and the purification scheme described above was followed. Activity was followed by NADPH:cytochrome c reductase activity. Active fractions were bright yellow throughout the purification, and the spectrum of the pure enzyme reproduced that reported earlier for the flavoprotein (30).
The FAD and FMN content of purified proteins was measured by extraction and fluorescence detection (22), with authentic FMN and FAD as standards. Specific removal of FMN from the sulfite reductase flavoprotein subunit was achieved by 18 h incubation with 1 mM chloromercuriphenylsulfonic acid at 4°C (31); the FMN content of treated enzyme was below the detection limit, and ϳ90% of FAD was retained.
Unless noted otherwise, sulfite reductase reactions were conducted in 50 mM Tris, pH 7.5, containing 50 M EDTA.
Assay compared with enzyme turnover. The oxidation rates of some enzymes were affected by the presence of ferric-ADP. Some of this effect was due to direct reduction of the ferric chelate, and this flux was subtracted in calculations of the oxidase activity of the enzymes.
Assay of H 2 O 2 Formation-H 2 O 2 production was measured using horseradish peroxidase. H 2 O 2 was generated in high yield by quinoneless respiratory chains and could be conveniently assayed using odianisidine as the dye (36). Because NADH is not consumed by quinoneless membranes and interferes with horseradish peroxidase-based assays by competing with the dye, respiration-proficient vesicles were added to scavenge the residual NADH before detection by the horseradish peroxidase system. The H 2 O 2 yielded by the oxidizing vesicles was scant compared with that from the quinoneless membranes, which were present in higher concentration. The initial reaction contained in a total volume of 4.2 ml: 30 M NADH, quinoneless vesicles containing 0.2 units of NADH dehydrogenase, and 30 units of SOD. At time points over 8 min, 800-ml aliquots were removed, and 0.02 units of NADH oxidase activity were added as UM1 membranes to scavenge rapidly the residual NADH. Then to the sample was added a 100 M mixture consisting of 150 M o-dianisidine, 0.06 mg/ml horseradish peroxidase, 25 M EDTA, and 50 mM KP i , pH 7.8. Absorbance was determined within 5 min at 460 nm. This protocol provided a time course of H 2 O 2 production by quinoneless vesicles; since the scant H 2 O 2 that was generated by UM1 vesicles was common to all time points, it was subtracted. We noted that when vesicles were formed by French pressing, some catalase was typically trapped inside the vesicles. Because this catalase interfered with H 2 O 2 detection, scavenging vesicles were prepared from the catalase mutant UM1, and the measurements of H 2 O 2 formation by respiratory chains were typically conducted on vesicles derived from catalase-free mutants.
The yield of H 2 O 2 from actively respiring vesicles was too small to be quantitated accurately by the dianisidine assay. Therefore the fluorescent dyes scopoletin, diacetyldichlorodihydrofluorescein, and amplex red were used as horseradish peroxidase substrates. Diacetyldichlorodihydrofluorescein was synthesized by standard methods (37). Vesicles were incubated with different amounts of NADH up to 300 M, in each case until the NADH was exhausted, and the H 2 O 2 that was generated was quantitated by published methods (38 -40). By so doing we determined the H 2 O 2 yield as a function of the amount of NADH that was oxidized. With all three fluorescent substrates, we observed a high yield of H 2 O 2 when as little as 10 M NADH was added; above 25 M NADH, however, the ratio of H 2 O 2 produced per NADH oxidized was consistent, and that is the value cited in this work. Whereas it is formally possible that H 2 O 2 formation is unusually rapid at low NADH concentrations, we suspect an horseradish peroxidase-based artifact, since the H 2 O 2 yield at these doses was dependent upon the period of dye development. The basis of the artifact remains unclear.
Enzyme Assays-NAD(P)H oxidase activities of vesicles and of purified enzymes were determined spectrophotometrically at 340 nm. NADH dehydrogenase activity was quantitated by monitoring NADH oxidation in the presence of 100 M plumbagin and 3 mM cyanide (21). Transhydrogenase, ferricyanide reductase, hydroxylamine reductase, O 2 . and H 2 O 2 Formation by Enzyme Autoxidation and sulfite reductase activities were measured by standard methods (23). Cytochrome c reduction was assayed with 0.1 mM cytochrome c rather than 1 mM. Unless otherwise indicated, all assays were conducted in air-saturated buffers at room temperature. For studies of enzyme oxidation at different oxygen concentrations, reaction mixtures were assembled in an anaerobic chamber. Anaerobic buffer was added to cuvettes, and the cuvettes were then sealed with rubber septa. The desired amount of aerobic buffer was then added by injection through the septa from an air-tight syringe. The cuvettes were filled to the top throughout this process to avoid any head space and prevent oxygen exchange with a gas phase. The reactions in the sealed cuvettes were then monitored at 340 nm. The temperature dependence of reactions was determined by using a thermostatted spectrophotometer and equilibrating the buffer temperature before initiating the reaction.

NADH dehydrogenase II Is the Primary Site of O 2 . and H 2 O 2
Formation in the Aerobic Respiratory Chain-The first goal of this study was to identify the autoxidizable members of the aerobic electron transport chain. In a previous investigation fumarate reductase was identified to be an autoxidizable component of the anaerobic chain, but that enzyme is not normally synthesized in air and was shown not to be responsible for the O 2 . produced by most aerobically grown cells (21).
Membrane vesicles were prepared from cells that had been grown both aerobically and anaerobically on minimal glucose medium. These vesicles contained intact respiratory chains but were prepared at neutral pH, which inactivates NADH dehydrogenase I, one of two respiratory NADH dehydrogenases in E. coli. That was useful, because NADH dehydrogenase I has a cytochrome c reductase activity that interferes with the standard O 2 . assay (see below). In these vesicles NADH dehydrogenase II was intact, and the chain oxidized NADH efficiently. production and respiratory capacity were inversely related (Fig. 2). E. coli contains two abundant NADH dehydrogenases: NADH dehydrogenase I, a large proton-translocating complex analogous to the mammalian enzyme, and NADH dehydrogenase II, a non-proton-translocating single subunit enzyme important in maintaining a redox-balanced dinucleotide pool.  Wild-type ϩ ␣-glyc-P 1.0 3.9 5.
Frd Ϫ ϩ ␣-glyc-P 0. Oxidation of Its Flavin-The autoxidation of NADH dehydrogenase II was more closely examined after purification. The enzyme contains an FAD cofactor, which receives electrons from NADH, and a tightly bound ubiquinone cofactor that co-purifies with the enzyme and apparently mediates electron transfer from the FAD to the diffusible quinone pool (33). No metal centers exist in the enzyme. NADH dehydrogenase II was overexpressed from a plasmid in both wild-type and quinone-deficient backgrounds. Membrane vesicles derived from overexpressing cells generated proportionately more O 2 . than did wild-type cells (Table II, lines 9 and 10). The enzyme was purified to homogeneity from both the wild-type and quinone-deficient membranes using an established protocol (33). The turnover numbers for O 2 . and H 2 O 2 production were similar for both forms of the enzyme (Table  IV). Thus it is the flavin moiety that directly transfers electrons to molecular oxygen. Electron transfer to oxygen was much slower than electron transfer to good univalent oxidants such as ferricyanide or plumbagin, indicating that the rate-limiting step in O 2 . formation is the transfer of electrons from the reduced enzyme to dissolved molecular oxygen. This transfer was not mediated by adventitious metals, since metal chelators such as EDTA and DETAPAC had a slight (ϳ40% maximum) accelerating effect on autoxidation. Similar stimulations have been described for xanthine oxidase and fumarate reductase, which also autoxidize at reduced flavins (21), although the basis of the effect is not known. The reactivity of NADH dehydrogenase II with oxygen is presumably an accidental consequence of the exposure of its reduced flavin to dissolved oxygen. Accordingly, the reaction followed chemical kinetics and slowed proportionately at low

O 2 . and H 2 O 2 Formation by Enzyme Autoxidation
concentrations of oxygen (Fig. 3A). The second-order rate constant for enzyme oxidation was 1.9 ϫ 10 4 M Ϫ1 s Ϫ1 at 37°C (with saturating NADH). An apparent energy of activation of 26 kJ was calculated from data above 15°C. At lower temperatures autoxidation was slower than expected from Arrhenius behavior, possibly because of changes in the enzyme structure (Fig. 3B). Sulfite Reductase Is an Autoxidizable Soluble Protein-The conclusion that isolated respiratory proteins autoxidize primarily or exclusively from flavin moieties raised the question of whether there is any abundant protein in E. coli that spuriously reacts with oxygen at a non-flavin site. This question cannot be answered comprehensively without examination of every redox protein. However, most of the known non-respiratory redox enzymes contain only flavins as redox cofactors. Among the exceptions, sulfite reductase may be the most abundant. It contains FAD, FMN, [4Fe-4S] clusters, and siroheme, and electrons flow in that order from the NAD(P)H donor to sulfite. Sulfite reductase was known to be capable of transferring electrons to a broad range of chemical oxidants, including oxygen (23). We therefore sought to determine which of its moieties were involved in autoxidation.
When aerobic cells were grown on the poor sulfur source djenkolate, which induced sulfite reductase synthesis about 100-fold above that of cystine-supplemented cells, no reliable difference in the total NADPH oxidase activity of the soluble extract could be determined. However, the overproduction of sulfite reductase from a plasmid caused a substantial increase in the NADPH oxidase activity of the whole-cell extract (not shown). Superoxide formation was difficult to measure by established assays, since sulfite reductase directly reduces both naive and acetylated cytochrome c (see below).
Holoenzyme was purified by a standard protocol, from SODdeficient cells in which sulfite reductase had been overproduced. The enzyme spectrum matched that of previous reports. Turnover numbers were 1450 min Ϫ1 (2160 min Ϫ1 ), 4420 min Ϫ1 (6800 min Ϫ1 ), and 18,700 min Ϫ1 (31100 min Ϫ1 ) for sulfite, hydoxylamine, and cytochrome c reduction, respectively. The parenthetical rates were obtained with added FMN and ranged from 93 to 118% of published values (23). The purified enzyme was somewhat FMN-deficient, containing 0.51 FMN/FAD.
Oxidase activity was indeed present; the turnover number was 63 NADPH min Ϫ1 per holoenzyme, similar to the value 75 min Ϫ1 measured by Siegel et al. (23). The turnover number was increased somewhat by iron chelators, as seen for other enzymes. Because FMN slowly dissociates from sulfite reductase (44), it was important to verify that the turnover to oxygen was not mediated by the enzymic reduction of free flavins. The sulfite reductase transfers electrons from its FAD moiety to free flavins (31), and the subsequent oxidation of those flavins by molecular oxygen could provide a spurious oxidase activity. To test whether the apparent oxidase activity arose from such a reaction, the NADPH oxidase activity was measured as a function of FMN concentration when free FMN was provided. In fact, the provision of FMN in high concentrations stimulated electron transfer both to oxygen and to other acceptors. However, the stimulation of the oxidase activity was biphasic. Extrapolation of the second phase to low FMN concentrations indicated that the initial phase approximately doubled the activity (Fig. 4). This was presumably achieved by filling the empty FMN sites. The residual effect of higher FMN was less pronounced, exceeded the capacity of the enzyme to stably bind additional FMN at unoccupied sites, and was not saturated at 1 M. We concluded that the excess FMN stimulated the oxidase activity by acting as an artificial electron acceptor rather than by occupying authentic FMN binding sites in the slightly de-flavinated enzyme. However, substantial basal oxidase activity existed even at very dilute enzymes concentrations (2 nM). In that circumstance the possible amount of free FMN could not have accounted for the oxidase activity, if one assumes a linear relationship between free FMN and spurious oxidase turnover.
Therefore the oxidase activity of the isolated, unsupplemented enzyme reflected bona fide electron transfer directly from the holoenzyme to oxygen, without a free-FMN intermediate. In subsequent oxidase experiments free flavin was not   (23), cyanide blocked electron transfer to sulfite but did not inhibit the oxidase activity. That result suggested that the oxidase reaction did not occur at the siroheme site, although electron transfer from an alternative face of the heme was not formally ruled out. The latter possibility was refuted by the observation that oxidase activity persisted in enzyme that was synthesized by a cysG mutant, which generates a siroheme-deficient enzyme (data not shown).
Both the siroheme and the iron-sulfur cluster are contained in the ␤ subunit, whereas FAD and FMN are bound by the ␣ subunit of the holoenzyme. The ␣ subunit (flavoprotein) can be purified in a stable form in the absence of the ␤ subunit, and this was achieved using an established protocol with cells that overexpressed only cysH, the flavoprotein structural gene. The purified flavoprotein exhibited a turnover number per FAD cofactor in excellent agreement with that of the holoenzyme (Table V). Siegel and Davis (45) had observed that flavoprotein that was dissociated from the iron protein by denaturants retained 10 -18% of the oxidase activity; their low yield was matched by poor recoveries of other reductase activities and presumably reflected enzyme damage during subunit dissociation. To identify more precisely the autoxidizing moiety, we removed FMN by treatment with p-chloromercuriphenylsulfonate. The treated enzyme retained FAD but was effectively devoid of FMN (Table V). Although the FMN-free flavoprotein retained transhydrogenase activity, the oxidase activity was reduced by 75%. This activity loss is quantitatively similar to the residual ferricyanide and menadione reductase activities that others (30) have reported. These experiments indicated either that FMN is the preponderant site of autoxidation or that it electronically interacts with FAD to facilitate the reaction of FAD with oxygen. FAD alone has slight reactivity with oxygen. The kinetics of sulfite reductase oxidation reflected its binding interactions with NADPH and chemical interactions with oxygen. The oxidase activity exhibited an apparent K m for NADPH of 5 M in air-saturated buffer, and high concentrations of either NADPH or NADP ϩ did not suppress oxidation (data not shown). The rate of autoxidation was proportionate to oxygen concentration (Fig. 5A). The rate constant for reaction with dissolved oxygen was 2.4 ϫ 10 3 M Ϫ1 s Ϫ1 per FMN at 37°C. The temperature dependence indicated an activation energy of 49.9 kJ/mol (Fig. 5B).
NADH-reduced Enzyme Reacts with Oxygen but Poorly with an NADP ϩ Analogue-Eschenbrenner et al. (31) reported the interesting observation that NADH can reduce the FMN moiety of sulfite reductase but that the reduced enzyme then reacts poorly with cytochrome c. This is in stark contrast to NADPHreduced enzyme. We repeated the observation, finding that NADH bleached the FMN absorbance peak. Interestingly, the NADH-reduced enzyme reacted well with oxygen, showing a turnover number only slightly lower than when NADPH was the reductant. The turnover number to cytochrome c was as low as that to oxygen, suggesting that the slow step might be the reduction of the enzyme. Surprisingly, acetylpyridine dinucleotide phosphate (AcPyNADP ϩ ), the NADP ϩ analogue that is reduced by reduced FAD, could only be reduced by the NADHtreated enzyme at a very slow rate, whereas acetylpyridine dinucleotide (AcPyNAD ϩ ), an NAD ϩ homologue, was as reducible as cytochrome c or oxygen (Table VI). This appears to confirm that the NADH-and NADPH-reduced enzymes are electronically dissimilar and that the electrons perhaps localize exclusively on the FMN in the NADH-reduced enzyme. Although the reason for this remains uncertain (perhaps a residual interaction with NAD ϩ affects the electron localization), it supports the conclusion that the preponderant site of electron transfer to oxygen is the FMN moiety rather than the FAD.  the immediate product. The univalent reduction potential of oxygen is low enough that it cannot pull electrons from unwilling donors; the donors must be proficient at univalent redox reactions. For this reason reduced dinucleotides and thiols are relatively stable in air. One anticipates that oxygen will react only with those electron carriers whose univalent oxidation states are stable: those carriers containing transition metals or with sufficient conjugation to stabilize the univalent oxidation products through resonance structures. In biological systems the obvious candidates are iron-sulfur clusters, hemes, quinones, and flavins.
In this study we have identified flavins as the primary sites of chemical oxidation of the respiratory chain and of sulfite reductase. In both cases the hemes, iron-sulfur clusters, and quinone pools were not substantial sources of either O 2 . or H 2 O 2 .
Why don't these other moieties react with oxygen? A number of answers are possible, but a likely one is that, unlike flavins, these moieties are sequestered in environments that are either sterically inaccessible to oxygen or are hydrophobic. The importance of burying clusters is underscored by those exceptional clusters whose catalytic function requires that they be solvent-exposed. These clusters (on nitrogenase, Fnr protein, and dehydratases) react rapidly with oxidants. In contrast, the clusters of sulfite reductase and NADH dehydrogenase I are buried in the protein, where they conduct internal electron transfers, and are stable in air. The clusters of succinate dehydrogenase and fumarate reductase are presumably near the enzyme surface, since their biochemical function is to reduce directly diffusible quinones, but electron transfer to oxygen may be blocked by the hydrophobicity of the membrane interior. . (47)). The rates at which flavoproteins react with oxygen are generally lower but range over orders of magnitude. Massey (51) has noted that members of the dehydrogenase class typically react slowly, in contrast with electron transferases, oxidases, and monooxygenases. We have seen substantial variation within nominal members of the dehydrogenase class, from undetectably low (NADH dehydrogenase I), to moderate (succinate dehydrogenase, sulfite reductase), to high (NADH dehydrogenase II, fumarate reductase). The different energies of activation for the oxidation of sulfite reductase (50 kJ/mol) and NADH dehydrogenase II (26 kJ/mol) suggest that electronic effects may control these rates; certainly local polypeptide context influences the relative stability of dihydroflavins and flavosemiquinones. However, several other factors are likely to be important, including the degree of flavin exposure and the local electron density. The failure of succinate dehydrogenase to autoxidize at the same efficiency as its homologue, fumarate reductase, may reflect the tendency of these enzymes to distribute their electrons differently; on reduced fumarate reductase electron density may be high on the flavin, in preparation for fumarate reduction; on succinate dehydrogenase the electron density is probably highest on its higher potential iron-sulfur clusters, in anticipation of ubiquinone reduction. It may be that NADH dehydrogenase I does not autoxidize because its electrons are primarily sequestered away from the flavin, on its iron-sulfur clusters. The rates at which flavoenzymes react with oxygen in vivo will also depend upon the overall enzyme redox status, which in turn depends upon the amount of oxidative substrate present. Thus the elimination by mutation of oxidized ubiquinone as a competitor for their electrons accelerated O 2 . formation  . The striking feature is that the univalent oxidation of reduced FMNH 2 will generate a semiquinone product that, with respect to oxygen, could be fairly stable. In fact this was observed as follows: while the reduced recombinant FMN fragment oxidized rapidly upon exposure to air, the product was a stable neutral flavosemiquinone whose own oxidation required hours. Since the enzyme lost only one electron, O 2 . must have been the exclusive product.
That observation contrasts with our finding that H 2 O 2 is almost the sole oxidation product of the complete flavoprotein; the difference is probably the influence of the adjacent FADH 2 .
As the FMNH 2 reacts with oxygen, the FADH 2 may be predisposed by its low potential to push an electron toward nascent FMNH and, from there, onto the nascent O 2 . , thus forming H 2 O 2 rather than O 2 . as the free product. In fact, Siegel et al. (23) originally speculated that FMN cycles between reduced and semiquinone states in bridging electron flow between FAD and the [4Fe-4S] cluster; in our analysis, oxygen has displaced the cluster as the electron acceptor. Strikingly analogous behavior was reported for xanthine oxidase; the two-electron-reduced enzyme generated only O 2 .
upon autoxidation. However, if the metal sites next to the autoxidizing flavin were also reduced, H 2 O 2 was generated as a near-stoichiometric product (16). and H 2 O 2 formation than will others. We are investigating whether this is the circumstance that forces some microaerophiles to avoid air-saturated environments. The first-order effect of oxygen concentration on enzyme oxidation makes plain the benefit of microaerobic habitats to such organisms. Finally, because O 2 . formation accelerates at high temperatures, oxidative stress is among those chemical problems to which aerobic thermophiles must have adapted. Since flavin autoxidizability depends upon polypeptide context, it will be interesting to determine whether homologous enzymes from thermophiles have evolved to be less reactive with oxygen than those of mesophiles.