Module 5.4.4

Oxidative Stress

JAMES A. IMLAY

[SECTION EDITOR: JOHN FOSTER]

Posted September 26, 2009

Department of Microbiology, University of Illinois, Urbana, IL 61801

Mailing address: Department of Microbiology, B103 CLSL, 601 S. Goodwin Ave., Urbana, IL 61801. Phone: (217) 333-5812; E-mail: This e-mail address is being protected from spambots. You need JavaScript enabled to view it .

INTRODUCTION

Life emerged 3.8 billion years (byr) ago in an anaerobic world. It was under these conditions that enzymes refined their catalytic mechanisms and that biochemical pathways coalesced and were integrated into metabolic networks. Absent any selective pressure to do otherwise, the emergent biological systems employed some mechanisms and structures that ultimately turned out to be incompatible with the presence of oxygen. Photosystem 2 appeared 2.7 byr ago, but molecular oxygen probably did not accumulate in quantity until it had titrated sulfide and ferrous iron from the seas, a period of another billion years. At that point the diverse life forms on the planet had to find ways to cope with oxygen, or else perish.

The ancestors of Escherichia coli and Salmonella ultimately evolved to thrive in air-saturated liquids, in which oxygen levels reach 210 μM at 37°C. However, in 1976 Brown and colleagues reported that some sensitivity persists: growth defects still become apparent when hyperoxia is imposed on cultures of E. coli (24). This residual vulnerability was important in that it raised the prospect that normal levels of oxygen might also injure bacteria, albeit at reduced rates that are not overtly toxic. The mechanism of such injuries was not self-evident, because molecular oxygen does not readily react with the basic structural molecules from which cells are made: amino acids, lipids, nucleic acids, and carbohydrates. Thirty years later we are still pursuing this problem, but a general picture has come into view. The intent of this article is both to describe the threat that molecular oxygen poses for bacteria and to detail what we currently understand about the strategies by which E. coli and Salmonella defend themselves against it.

COPING WITH MOLECULAR OXYGEN PER SE

The essential chemical activity of oxygen is the abstraction of electrons from reduced molecules. This behavior is thermodynamically favored and is exploited when respiring organisms oxidize foodstuffs, and yet the uncontrolled and undesirable oxidation of other biomolecules constitutes a threat. Fortunately, such reactions are kinetically limited by the unusual orbital structure of molecular oxygen. Molecular oxygen is a di-radical, with unpaired, spin-aligned electrons occupying two orbitals. A consequence of this arrangement is that molecular oxygen cannot simultaneously abstract two electrons from conventional, spin-paired organic molecules; instead, in adventitious reactions oxygen can only receive one electron at a time (197). This feature places an important restriction on the reactivity of oxygen, because the affinity of molecular oxygen for the first electron (−0.16 V) is quite low (Fig. 1). For example, guanine—the most oxidizable of the DNA bases—has a univalent reduction potential near +0.5 V, ensuring that oxygen cannot directly pull an electron away from it. Thus, the combination of its orbital arrangement and its modest univalent electron affinity prevents molecular oxygen from reacting at discernible rates with structural biomolecules.

Fig. 1The redox states of molecular oxygen. Left to right: molecular oxygen, superoxide, hydrogen peroxide, the hydroxyl radical, and water. Reduction potentials are shown; the reduction potential for molecular oxygen considers the standard state to be 1 M. At the bottom, the relative cytoplasmic concentrations of molecular oxygen, superoxide, and hydrogen peroxide in unstressed aerobic E. coli are shown.

Nevertheless, the radical structure of molecular oxygen predisposes it to bond with the unpaired electrons of organic radicals. Enteric bacteria employ enzymes that use organic radicals to catalyze a number of reactions that cannot easily be managed by nonradical mechanisms. Among these enzymes are ones that generate radicals from S-adenosylmethionine and adenosylcobalamin (B₁₂). These radicals are formed momentarily during the catalytic cycle, and either their transience or their occlusion by substrate largely protects them from reacting with dissolved oxygen. However, such is not true of glycyl-radical enzymes (232), which maintain a solvent-exposed radical even when the enzyme is in the resting state. This family arose among microbes in the ancient anaerobic world, and it includes key catabolic enzymes in contemporary anaerobes. Within the enterics it includes the anaerobic ribonucleotide reductase (NrdD), pyruvate:formate lyase (Pfl), and 2-ketobutyrate:formate lyase (TdcE). In these enzymes the glycyl radical initiates a chemically difficult reaction by abstracting a single electron from substrate. The subsequent mechanisms are still only speculative, but the cycle is resolved by the return of the electron to substrate with regeneration of the enzyme radical.

Radical-radical reactions are famously rapid, and so it is not surprising that a small molecule like molecular oxygen can enter the active site of the resting enzyme in vitro and combine with the glycyl radical. The resultant peroxyl species cleaves the polypeptide backbone within seconds, eradicating activity (269).

The oxygen sensitivity of these enzymes is tolerable for E. coli and Salmonella, because in aerobic habitats these enzymes are expendable. The anaerobic ribonucleotide reductase is supplanted by the aerobic (NrdAB) isozyme, an enzyme that not only tolerates oxygen but also requires it for activation of its catalytic tyrosyl residue. Meanwhile, the raison d'etre of Pfl is the anaerobic cleavage of pyruvate without the generation of NADH, thereby avoiding the wasteful use of acetyl-CoA as an electron acceptor rather than as a source of ATP. In aerobic habitats this economy is unnecessary, and pyruvate dehydrogenase assumes the task of pyruvate dissimilation, generating NADH that is simply reoxidized through respiration.

More intriguing is the question of how these isozymes are managed in intestinal environments in which oxygen might be present only briefly. Without further arrangements one might expect a bolus of oxygen to destroy the anaerobic enzymes, necessitating their expensive resynthesis when anaerobiosis is restored. For Pfl (and presumably the other glycyl-radical enzymes) an elegant answer has been proposed: when erstwhile anaerobic cells encounter oxygen, cellular machinery quickly reduces the glycyl radical to a conventional glycyl residue (232). In this form the enzyme would be inactive but incapable of being damaged by oxygen, and when anaerobiosis returned the enzyme would be reactivated by its dedicated activase system. Interestingly, the Pfl-deactivation process has been proposed to require the participation of alcohol dehydrogenase (150). The mechanism by which oxygen might trigger this activity of the dehydrogenase, and the mechanism by which the dehydrogenase would deactivate Pfl, remain to be determined. A more recent study, however, did not detect evidence of protection of Pfl by Adh (201).

In toto, the pairing of Pfl and Pdh, of aerobic and anaerobic ribonucleotide reductases isozymes, and of the putative glycyl-radical activation/deactivation systems comprise adaptations that allow E. coli and Salmonella to thrive during periods of oxygenation. The absence of such flexibility is, in part, what consigns committed anaerobes to oxygen-free niches.

SOURCES OF SUPEROXIDE AND HYDROGEN PEROXIDE

Formation of Cytoplasmic Superoxide and Hydrogen Peroxide by Endogenous Processes

E. coli mutants that lack either superoxide dismutases or catalases and peroxidases exhibit a variety of growth defects (34, 140). These phenotypes constitute the best evidence that aerobic cells continually generate intracellular superoxide (O₂⁻) and hydrogen peroxide (H₂O₂) at potentially lethal doses. Similar growth defects are created when wild-type cells are exposed to hyperoxia (26), which suggests that high oxygen levels accelerate the formation of O₂⁻ and H₂O₂ to such a degree that the scavenging enzymes are inadequate defenses.

Superoxide is created when molecular oxygen abstracts a single electron from a donor, and hydrogen peroxide is formed when two electrons are abstracted (Fig. 1). None of the enzymes in major metabolic pathways produces either of these species as deliberate, stoichiometric reaction products; instead, reactive oxygen species (ROS) are apparently formed through the adventitious oxidation of redox enzymes. A wide variety of reduced flavoenzymes inadvertently transfer electrons to oxygen when it collides with their flavin moieties (186). For example, while the proper function of NADH dehydrogenase II is to transfer electrons from NADH to the ubiquinone pool, about 0.5% of the electron flow is diverted when molecular oxygen steals electrons from the solvent-exposed flavin of the reduced enzyme (189). Similar behaviors have been documented for other flavoenzymes, including succinate dehydrogenase, fumarate reductase, NADH dehydrogenase I, glutamate synthase, lipoamide dehydrogenase, and sulfite reductase (89, 106, 162, 189, 190). The diversion of such a small fraction of electrons has little direct impact on metabolism, but the subsequent reactivity of superoxide and H₂O₂ is consequential. Reduced flavins are generally good low-potential univalent reductants, which explains their ability to transfer electrons even to a poor acceptor like oxygen (187). The vast majority of redox enzymes that are found in contemporary aerobes were inherited from anaerobic ancestors, and the tendency of these enzymes to autoxidize may reflect the original absence of selective pressure to avoid doing so. However, more recent adaptations may diminish this behavior—by restricting the solvent exposure of the flavin isoalloxazine ring, by raising its redox potential, or by limiting the electron residence time on it. For example, succinate dehydrogenase is far less prone to autoxidation than is fumarate reductase, its structural homologue and anaerobic ancestor (190). This difference has been ascribed to the evolutionary acquisition of a heme-binding site on succinate dehydrogenase; the heme pulls the electron density on the resting enzyme away from the flavin and may thereby make the enzyme more appropriate for use in an aerobic habitat (277).

The transfer of one electron to oxygen produces superoxide. Frequently, a second electron is transferred to the nascent superoxide before it leaves the flavoprotein active site, presumably after an electron-spin flip that accommodates the spin restriction (Fig. 2) (190). In this circumstance H₂O₂ is formed—and, in fact, most reduced flavoenzymes appear to release more H₂O₂ than O₂⁻ as a diffusible product. Once released into the cytoplasm, O₂⁻ is dismuted either enzymatically or spontaneously, generating H₂O₂. Thus, the autoxidation of flavoenzymes may be the predominant source of both O₂⁻ and H₂O₂.

Fig. 2The adventitious oxidation of enzymic dihydroflavin generates hydrogen peroxide and superoxide.

E. coli mutants that lack catalase and peroxidase release H₂O₂ into the medium, where it can be directly quantified (239). Such measurements indicated that 10 to 15 μM/s H₂O₂ is formed inside exponential-phase E. coli in air-saturated medium. Because O₂⁻ is generated as a lesser product during enzyme oxidation, its rate of formation in vivo has been estimated to be approximately 5 μM/s. Nevertheless, the steady-state concentrations of H₂O₂ and O₂⁻ probably rise no higher than 20 nM and 0.2 nM, respectively, inside wild-type cells, because of the abundance and high activity of scavenging enzymes (see below). The rates of enzyme oxidation, and therefore of ROS production, are proportionate to oxygen concentration; therefore, the concentrations of these species would be far lower in microaerobic habitats.

No single enzyme has yet been identified as a preponderant source of ROS in E. coli. Mutations that eliminated most respiratory functions did not diminish H₂O₂ release from catalase/peroxidase mutants, which raises the prospect that cytosolic enzymes might turn out to be the most substantial sources (240).

Formation of Superoxide in the Periplasm

E. coli synthesizes a periplasmic copper-zinc superoxide dismutase that is encoded by sodC (16). E. coli mutants that lack sodC exhibited phenotypic defects, which suggests that endogenous processes form potentially harmful doses of superoxide in the periplasm (96). Indeed, superoxide is released into the periplasm when reduced menaquinone reacts with molecular oxygen (156). Ubiquinone, the more abundant respiratory quinone in aerobic cells, has a higher electron affinity and appears not to generate superoxide.

Direct measurements indicate that the rate of periplasmic superoxide formation in exponentially growing cells (3 μM/s) is similar to the estimated rate of cytoplasmic superoxide formation (ca. 5 μM/s). Even greater periplasmic O₂⁻ stress may occur when pathogenic bacteria are exposed to fluxes of O₂⁻ from the phagocytic NADPH oxidase (198). For this reason Salmonella species express additional SodC isozymes whose function is necessary for full virulence (33, 71, 76) (see below). Workers have not yet identified the periplasmic biomolecules that these enzymes serve to protect.

External Sources of Oxidative Stress

Their basal defensive enzymes allow E. coli and Salmonella to thrive in environments that are saturated with air. Nevertheless, these bacteria induce scavenging enzymes to even higher titers when H₂O₂ and redox-active chemicals are added to laboratory cultures (111, 223). This observation poses the question: What are the natural sources of oxidative stress that require these extra defenses?

Superoxide (pK_a = 4.8) is a deprotonated anion at neutral pH, and as a charged species it cannot cross cell membranes (157, 178). Therefore intracellular superoxide stress can only be created by mechanisms that accelerate its formation within the cell. Experimenters have imposed superoxide stress by adding redox-cycling compounds such as paraquat and menadione to cultures (112). These agents can diffuse into cells, where they generate superoxide by chemically oxidizing flavoenzymes and then transferring the electron to molecular oxygen. When high doses of these agents are added to well-fed cells, superoxide can be formed intracellularly at a rate of several mM/s, which is three orders of magnitude above the endogenous rate. Redox-cycling antibiotics might constitute important sources of intracellular superoxide and hydrogen peroxide in some natural habitats. A variety of plants release soluble quinones such as plumbagin and juglone from their leaves and seeds, presumably as herbicides that suppress the competitive growth of other plants in their environment. When swept into surface waters these compounds would have a toxic effect on microbes. Further, some microbes follow a similar strategy in poisoning their competitors. Pseudomonas aeruginosa, for example, excretes the phenazine pyocyanin as a redox-cycling antibiotic that penetrates nearby microbes (114). The pertinence of such antibiotics is implied by the fact that the SoxRS system of E. coli and Salmonella induces not only superoxide dismutase but also enzymes that enhance both drug exclusion and export (55, 68, 282).

Hydrogen peroxide differs from superoxide in that it can penetrate cells fairly quickly (239). Therefore, intracellular H₂O₂ stress arises whenever cells encounter H₂O₂ in their extracellular environment. Environmental H₂O₂ can be formed in significant quantities when metals catalyze the chemical oxidation of organic or sulfur compounds. Extracellular superoxide and H₂O₂ are also produced whenever riboflavin (or a similar chromophore) is illuminated. (Such processes produce substantial H₂O₂ in complex media under room lighting, which can affect laboratory experiments.) Other potential sources of H₂O₂ in surface waters include catechol compounds that are released when plants rot, as well as reduced sulfur species that seep from anaerobic sediments.

Hydrogen peroxide is also introduced into biological habitats by the pyruvate and lactate oxidases of lactic acid bacteria (97, 242, 248). Laboratory cultures of some lactic acid bacteria accumulate millimolar H₂O₂, which far exceeds the doses that block the growth of most other microbes. During H₂O₂ production the lactic acid organisms themselves cannot operate pathways that contain H₂O₂-sensitive enzymes. Lactobacillus acidophilus (44) lacks iron-sulfur dehydratases entirely, while these enzymes are present but apparently nonfunctional during H₂O₂ production in Streptococcus pneumoniae (248). The latter bacterium therefore requires exogenous branched-chain and α-ketoglutarate-derived amino acids.

In sum, environmental sources of oxidative stress include both adventitious chemical oxidation processes and the influx of redox-cycling antibiotics or of hydrogen peroxide itself. In these cases the cell must cope with doses of oxidants that far exceed what is routinely generated by its own metabolism. The inducible OxyR and SoxRS systems are not activated by mere aerobiosis, and so presumably they exist to protect cells from these external sources of stress.

HOW DOES SUPEROXIDE DAMAGE CELLS?

Superoxide has reduction potentials that allow it to serve in vitro as either a weak univalent reductant or a stronger univalent oxidant (Fig. 1). The latter activity, however, is substantially impeded because, as an anion, superoxide will be repelled from many potential electron donors and protonation must precede electron transfer. Early attempts to identify biomolecules that O₂⁻ could damage were not successful (20, 74, 77, 233). Superoxide does not react in vitro at substantial rates with nucleic acids. It may show low activity with cysteine (274) but not with other amino acids. Results were also negative with the carbohydrates that were tested, although more recent data suggest that it may oxidize short-chain sugars (204).

For a while these results provoked doubt as to the biological significance of superoxide and thus about the true physiological role of superoxide dismutase. The latter question was resolved, however, when Carlioz and Touati generated mutants of E. coli that lacked both its cytoplasmic manganese- and iron-cofactored superoxide dismutases (MnSOD and FeSOD) (34). These mutants grow well in the absence of oxygen, but they exhibit marked defects in aerobic media. These include an inability to grow without supplementation with branched-chain, aromatic, and sulfurous amino acids; poor growth on nonfermentable carbon substrates, such as acetate and fumarate; and a high rate of mutagenesis (24, 34, 72). These phenotypes were complemented by plasmids that encoded the structurally distinct mammalian copper-zinc superoxide dismutase (199), confirming that the defects arise from the lack of superoxide dismutation.

Damage to Iron-Sulfur Clusters

Investigations of these phenotypes have identified specific biomolecules that superoxide can damage. SOD-deficient mutants cannot synthesize leucine, isoleucine, and valine because superoxide inactivates dihydroxy acid dehydratase, an enzyme in the common pathway of branched-chain synthesis (78, 161). They cannot catabolize substrates that feed into the citric acid cycle because aconitases A and B and fumarases A and B are similarly damaged (86, 173). These enzymes all belong to a family of dehydratases that employ a (4Fe-4S) cluster to dehydrate and/or hydrate substrates (79). Serine dehydratase, threonine deaminase, isopropylmalate isomerase, and 6-phosphogluconate dehydratase are other family members. In each enzyme a solvent-exposed iron atom within the cluster binds substrate and then serves as a Lewis acid that abstracts a hydroxide anion from it (164). The problem is that, in the absence of substrate, the catalytic iron atom can be complexed by anionic superoxide. After protonation, perhaps by bulk solvent, electron transfer can occur from the cluster to the superoxide, generating hydrogen peroxide and an oxidized (4Fe-4S)³⁺ cluster (Fig. 3) (79). The cluster is unstable in this valence and hydrolyzes, releasing the substrate-coordinating iron atom as Fe²⁺ and leaving behind a (3Fe-4S)⁺ cluster that is catalytically inactive. Polypeptide is unaffected by the reaction.

Fig. 3Oxidative inactivation of a dehydratase [4Fe-4S] cluster by superoxide.

The rate constant for this oxidation is exceptionally high for an adventitious reaction: 10⁶ to 10⁷ M⁻¹ s⁻¹. The superoxide concentration in aerobic E. coli has been estimated to be about 10⁻¹⁰ M, which implies that the half-life of a dehydratase would be about 30 min (95). Notably, cells continuously reactivate damaged clusters, so that probably >95% of the enzyme pool is active at any time in aerobic wild-type cells (87, 95).

Iron-sulfur clusters that are buried in polypeptide cannot coordinate superoxide, and the enzymes that contain them are not damaged by superoxide. Most electron-transfer enzymes, including respiratory dehydrogenases, fall into this category.

The destruction of dehydratase clusters is indirectly responsible for the high mutation rate of SOD mutants. Oxidative DNA damage occurs when DNA-bound iron reacts with hydrogen peroxide. Because SOD mutants continuously release iron from their clusters, the cellular content of “free” (unincorporated) iron is high. Some of this lost iron will bind by chance to nucleic acids, where it can catalyze their damage (151, 174).

Other Injuries from Superoxide Stress

Benov and Fridovich have proposed that the aromatic auxotrophy of SOD mutants results from the inactivation of transketolase (15). Transketolase employs a thiamine cofactor to transfer two-carbon units from one sugar to another, and its activity is needed for the formation of erythrose-4-phosphate, a precursor in the synthesis of aromatic compounds. Prior work with plant enzymes indicated that superoxide can oxidize the dihydroxyethyl thiamine intermediate of the transketolase catalytic cycle, releasing glycolic acid (250).

SOD-deficient mutants are also unable to use sulfate as a sole sulfur source (13, 34). The basis of this phenotype is obscure. Enzymes in the sulfate-assimilation pathway appear not to be inactivated by superoxide, and yet sulfur species are found in the medium, as if a pathway block results in excretion of an intermediate (14). Interestingly, a similar phenotype is observed in SOD1 mutants of yeast (40).

HOW DOES HYDROGEN PEROXIDE DAMAGE CELLS?

H₂O₂ is not a radical species, and in principle it can act as either a univalent or divalent oxidant (Fig. 1). In both cases, however, the dioxygen bond must be fractured; and despite the overall thermodynamic driving force, this bond-breaking step imposes a high energy of activation that makes H₂O₂ inert towards most biomolecules.

Nevertheless, the addition of micromolar H₂O₂ to lab media will immediately block the growth of most cells, and protracted exposure will result in the loss of viability (132). Most of the toxic actions of H₂O₂ can be ascribed to forms of the Fenton reaction, in which the univalent reduction of H₂O₂ by ferrous iron generates a hydroxyl radical:

Fe²⁺ + H₂O₂ → (FeO²⁺) → Fe³⁺ + HO.   (1)

This reaction is kinetically competent because iron coordinates and stabilizes the hydroxyl-like product in a metastable ferryl complex. Once released, the hydroxyl radical is an extremely powerful oxidant (Eo′ = +2.3 V; Fig. 1), and it reacts at nearly diffusion-limited rates (10⁸ to 10¹⁰ M⁻¹ s⁻¹) with most organic molecules (52). Therefore, it reacts near the site of its formation, which is to say the sites at which ferrous iron might be found inside cells.

Damage to DNA

After a brief exposure to H₂O₂, E. coli cells grow into long filaments, indicating that they have suffered DNA damage that blocks replication. The implication is that some unincorporated iron atoms bind adventitiously to nucleic acids and that on exposure to H₂O₂ these iron atoms generate hydroxyl radicals on the surface of the DNA (222). If a hydroxyl radical abstracts an electron from the ribose moiety, a ribosyl radical is formed, and the addition of molecular oxygen leads ultimately to strand cleavage (61, 128). When hydroxyl radicals abstract electrons from DNA bases, a wide variety of adducts can be formed. Interestingly, an electron can tunnel from low-potential guanine to a nearby base radical, reverting the original radical to a normal base but generating a guanyl radical, so that 8-hydroxyguanine is a disproportionate product of DNA oxidation (31).

The rate constant of the Fenton reaction depends on the ligands that bind the ferrous iron atom; measurements have ranged from ca. 5,000 M⁻¹ s⁻¹ for the average iron molecule bound to DNA to ca. 20,000 M⁻¹ s⁻¹ for hexaqueous and ATP-bound iron (210, 228). These values are sufficiently high that lethal levels of DNA damage can be generated when cells are exposed to low-micromolar concentrations of H₂O₂ for an extended period (210).

In wild-type cells the intracellular H₂O₂ level is predicted to be no higher than 20 nM (239). Nevertheless, even this dose creates enough damage that certain DNA repair mutants—recA xthA and polA recB strains—are viable only if they are cultured in anaerobic media (130, 192). Further, consistent with equation 1, DNA damage is proportionately accelerated by circumstances that elevate the intracellular level of unincorporated iron. For example, high rates of mutagenesis are observed in SOD⁻ mutants, which leak iron from damaged enzyme clusters, and in fur mutants, which lack homeostatic controls on iron content (151, 260).

Kinetic experiments indicate that during a period of H₂O₂ exposure the unincorporated iron atoms turn over repeatedly, generating a hydroxyl radical each time. This behavior requires that the oxidized ferric form be re-reduced by a univalent electron donor. Indeed, the relative slowness of this reaction (equation 2) limits the rate at which high concentrations of H₂O₂ damage DNA in vivo. The biological reductants are likely to be either cysteine or free reduced flavins, since the damage rate inside cells is elevated by manipulations that increase the concentrations of either (209, 276). In fact, these experiments suggest that homeostatic controls on cysteine levels (158) may exist, in part, to minimize the rate of Fenton chemistry.

Does Copper Mediate Oxidative Damage?

In the test tube, copper shares with iron the ability to transfer electrons to hydrogen peroxide, thereby generating hydroxyl radicals (108). However, most studies indicate that iron is the exclusive source of hydroxyl radicals in vivo, even when cells are overloaded with copper (132, 181). A possible explanation is that cytoplasmic copper is avidly bound by the thiolates of proteins, metabolites, and glutathione. Such thiol compounds likely reduce nascent (CuO)⁺ species before they dissociate into copper (II) and hydroxyl radicals. This scheme implies that the thiols are oxidized in the process, which has led some workers to posit that copper may debilitate cells by creating disulfide bonds in proteins (120). It is important to note that copper inhibits the growth of E. coli under anaerobic conditions, too, so that a nonoxidative pathway of toxicity must also exist (18, 181a, 206).

Damage to Iron-Sulfur Clusters

It is unlikely that E. coli and Salmonella experience more than micromolar doses of H₂O₂ in natural habitats. When these concentrations of H₂O₂ are added to laboratory cultures, growth is transiently inhibited, and then scavenging enzymes reduce the ambient H₂O₂ concentration to subtoxic levels. Therefore, it has proven useful to study growth inhibition in mutant (ahpCF katG katE) strains that lack the catalase and peroxidase activities (238). In such cultures low (micromolar) concentrations of H₂O₂ can be maintained long enough for investigators to track the growth problems and to biochemically characterize the responsible damage.

These strains cannot grow in aerobic medium without aromatic amino acid supplements (J. M. Sobota and J. A. Imlay, unpublished data). As little as 0.5 μM intracellular H₂O₂ is sufficient to create this phenotype. Catabolism of gluconate is also defective, suggesting that the specific problem is a lack of pentose-phosphate-pathway function. While this phenotype mirrors the behavior of superoxide-stressed cells, the root cause seems unlikely to be the mechanism proposed for superoxide, since H₂O₂ is not believed to be capable of metal-free univalent oxidations.

Low-micromolar doses of H₂O₂ also block the tricarboxylic acid (TCA) cycle and leucine biosynthesis (140). As with superoxide toxicity, this problem stems from the ability of H₂O₂ to oxidize the (4Fe-4S) clusters of dehydratases: isopropylmalate isomerase in the leucine pathway, and aconitase and fumarase in the citric acid cycle. These reactions are essentially Fenton reactions and exhibit rate constants of 10³ to 10⁴ M⁻¹ s⁻¹. This oxidation probably occurs in two steps: the initial formation of a hydroxyl-like ferryl radical, and its subsequent abstraction of a second electron from the cluster. If so, the outcome would resemble the oxidation by H₂O₂ of the ferrous iron atom of PerR, in which an initial radical abstracts a second electron from the residues that coordinate the iron atom (168). With enzymic iron-sulfur clusters the second electron is withdrawn from the residual cluster, thereby avoiding covalent damage to the polypeptide chain. The absence of such damage leaves the enzyme in a form that the cell can reactivate.

Damage to Polypeptides

Neither hydrogen peroxide nor superoxide directly oxidizes most amino acids quickly enough for these reactions to be of physiological significance. The conditional exception is cysteine, which is discussed below. However, if Fenton reactions occur on the surface of proteins, the hydroxyl radicals they form can oxidize amino acid side chains and the α-carbon of the polypeptide backbone at rates (10⁸ to 10¹⁰ M⁻¹ s⁻¹) that approach the diffusion limit (52). Thus, a variety of enzymes that use iron as a prosthetic metal can be inactivated by hydrogen peroxide in vitro, including alcohol dehydrogenase (253) and iron-containing superoxide dismutase (19). Moreover, other enzymes that nonspecifically bind divalent cations can be metallated by iron in vitro, and in this form they too can be inactivated through Fenton chemistry (Sobota and Imlay, unpublished) (194).

The immediate product of amino acid oxidation by a hydroxyl radical is a radical species, whether it is generated through hydrogen atom abstraction (as with aliphatic side chains) or hydroxyl addition (to sulfur atoms or unsaturated amino acids). In aerobic environments molecular oxygen is likely to add to the amino acid radical, forming a peroxyl species; in proteins this radical can in turn abstract a hydrogen atom from nearby residues and thereby propagate a chain reaction (53). Peroxide adducts are unstable and disintegrate into a large number variety of products. Many of these are difficult to quantify directly, and so investigators commonly assay specific species as proxies of protein oxidation. Common choices include disulfide bonds, methionine sulfoxide, 2-oxohistidine, or 3,4-dihydroxyphenylalanine (DOPA)—the predominant products of the oxidation of cysteine, methionine, histidine, and tyrosine, respectively. Several side-chain radicals can ultimately decompose to carbonyl products, which are easily detected through Schiff's base reagents. The abstraction by hydroxyl radical of an electron from polypeptide α-carbon atoms can generate an imine intermediate, whose subsequent hydrolysis cleaves the polypeptide chain.

This chemistry has been solved in vitro. The key question is whether any E. coli enzymes commonly use solvent-exposed ferrous iron atoms as prosthetic cofactors in vivo. To date no metabolic failures in oxidatively stressed cells have been conclusively linked to protein-localized Fenton reactions. However, carbonyl residues have been detected in extracts from aerobic cells, and their levels were elevated when the cells were exposed to hydrogen peroxide or redox-cycling drugs (64). In that study some of the most abundantly oxidized proteins were iron- or divalent cation-binding proteins, including glutamine synthetase and isocitrate dehydrogenase. Further, the protective effect of manganese on H₂O₂-stressed cells can be rationalized by the displacement of iron from mononuclear metalloenzymes (below). Still, further work is needed to pin down the rates and impact of these injuries.

Do Oxidants Oxidize Protein Thiols?

Because protein thiols gradually oxidize in vitro, it was easy to believe that they would oxidize even more rapidly inside cells that are stressed with hydrogen peroxide. The likely mechanism was through formation of a sulfenic acid species that might then condense with a second cysteine residue, forming an intra- or intermolecular disulfide bond:

Cys-SH + H₂O₂ → Cys-SOH + H₂O   (2)

Cys-SOH + Cys-SH → Cys-SS-Cys + H₂O   (3)

However, the evidence to support this idea is scanty. None of the phenotypes of peroxidase/catalase mutants have yet been traced back to protein sulfhydryl oxidation. In general, investigators have resorted to near-millimolar doses of H₂O₂ to generate disulfide bonds that could be detected in proteomic analyses of E. coli and Saccharomyces (165, 169, 252).

The key fact is that H₂O₂ simply does not react quickly with “typical” thiols. The rate constant for its reaction with free cysteine is only about 2 M⁻¹ s⁻¹ at physiological pH (274); even in basic solutions, which fully deprotonate cysteine to its oxidizable thiolate form, it is only 20 M⁻¹ s⁻¹. For comparison, this value is more than two orders of magnitude below those for Fenton reactions. The upshot is that the submicromolar H₂O₂ doses that damage DNA and inactivate dehydratases within minutes would require days to oxidize a standard cysteine residue.

What about cysteine residues that are specially activated by their polypeptide context? OxyR, AhpC, and OhrR each feature thiols that react with H₂O₂ with rate constants in excess of 10⁷ M⁻¹ s⁻¹ (8, 216). Their high activity probably stems from two features: a cationic residue adjacent to the reactive cysteine, which favors its deprotonation, and a nearby proton-donating residue that presumably polarizes the H₂O₂ dioxygen bond and facilitates its cleavage by protonating the hydroxyl leaving group (Fig. 4) (121, 154, 247). The steric arrangements of these residues must be important, because glyceraldehyde-3-phosphate dehydrogenase, which has a generally similar environment surrounding its catalytic cysteine residue, is not easily inactivated by H₂O₂ (k ~ 60 M⁻¹ s⁻¹ [279]).

Fig. 4Likely mechanism for the activation of hydrogen peroxide to enable oxidation of a cysteinyl residue to a sulfenic acid. Based on Kona and Brinck (154).

E. coli contains glutathione, two thioredoxins, and three glutaredoxins, and they all have the capacity to reduce sulfenic acids and disulfide bonds (224). Under routine growth conditions they reduce disulfide bonds that are generated during the normal catalytic cycles of ribonucleotide reductase, PAPS reductase, arsenate reductase, etc. Their additional involvement in defending the cell against oxidative stress seemed likely when it was discovered that thioredoxin 2 (encoded by trxC) and glutaredoxin 1 (grxA) are induced by the OxyR system during exposure to H₂O₂. Nevertheless, various redoxin mutants do not exhibit aerobic growth defects aside from the failure of the redoxin-driven enzymes listed above (224), and catalase/peroxidase mutants are not further sensitized to H₂O₂ by trxC and grxA deletions (J. E. Martin and J. A. Imlay, unpublished results).

These results suggest that thiol oxidation by H₂O₂ may be rare. Oxidation by Fenton chemistry is likely, but it should target cysteine only marginally more than other amino acids. Oxidation by other species—including molecular oxygen itself or disulfide compounds such as cystine—is not excluded. Indeed, methionine synthase is minimally sensitive to H₂O₂ but is a particular target of the disulfide stressor diamide (125), and the chaperone Hsp33 is activated by disulfide formation but is resistant to physiological doses of H₂O₂ (138, 273).

Do Oxidants Peroxidize the Membranes of E. coli and Salmonella?

Oxidative stress triggers the peroxidation of lipids in mammalian membranes, and authors have speculated that a similar event may debilitate bacteria. However, bacterial membranes lack polyunsaturated lipids, and polyunsaturation is regarded as a key element in the ability of lipids to propagate the peroxidative chain reaction. Figure 5 depicts the generally accepted model of peroxidation. The relevant point is that the peroxyl radical species can abstract hydrogen atoms from fatty acids only if, by doing so, it produces a methylene radical that is conjugated to—and thus stabilized by—two double bonds. The experimental evidence that supports this idea is that in vitro model systems failed to peroxidize membranes that lacked polyunsaturated fatty acids (21).

Fig. 5The standard model for lipid peroxidation. The identity of the initiating oxidant in reaction 1 is unknown. Propagation reaction 3 is possible only because of the stabilizing effect of resonance upon the radical product. Unlike the lipids in the figure, bacterial lipids are not polyunsaturated and therefore resist peroxidation.

Nevertheless, workers have detected some thiobarbituric acid-reactive species (TBARS) when bacteria have been exposed to oxidative stress (92, 243), and the aldehyde reductase YqhD has been proposed to be a scavenger of peroxidation-derived by-products (215). The TBAR assay is convenient and has been used frequently as an indicator of lipid peroxidation. However, the method does not directly quantify lipid peroxides—it detects some of their breakdown products—and so it is potentially misleading. An alternative role of YqhD might be to eliminate active aldehydes that form on oxidation of nonlipid biomolecules (109). More work is needed to resolve these uncertainties.

REGULATORY RESPONSES TO OXIDATIVE STRESS

Bacteria cannot control their environments, and so they have evolved robust transcriptional responses to defend themselves against external stressors. The need for inducible antioxidant systems seems especially obvious for enteric bacteria, which move quickly from the anaerobic gut to fully aerobic surface waters or even to ROS-perfused phagolysosomes. E. coli and Salmonella have provided two paradigmatic models of oxidative-stress responses: the SoxRS and OxyR systems.

The SoxRS System

Hassan and Fridovich discovered that the manganese-containing superoxide dismutase (MnSOD) is strongly induced when E. coli is treated with redox-cycling antibiotics that generate intracellular superoxide (111). The Demple and Weiss laboratories subsequently demonstrated that this response is controlled by two proteins: SoxR, a sensor protein that detects the redox stress, and SoxS, a response regulator that transcriptionally activates about two dozen genes that are scattered around the chromosome (Fig. 6) (103, 261). SoxR is a homodimer that binds near the soxS promoter. It contains one (2Fe-2S) cluster per subunit, and in unstressed cells these clusters maintain a +1 valence (59, 88). However, electron paramagnetic resonance (EPR) studies demonstrated that the cluster was oxidized to a +2 state when redox-cycling agents were added to cultures. This redox change provokes a conformational shift in SoxR, perhaps because of electrostatic repulsion between the two oxidized clusters. This conformational change contorts the bound promoter region, thereby improving an RNA polymerase binding site that is otherwise poorly functional because of an unusually short distance between its −35 and −10 promoter elements (117).

Fig. 6The SoxRS response to oxidative stress. Neither the identity of the oxidant nor that of the reductant is certain.

This process increases transcription of soxS by more than 20-fold. The newly synthesized SoxS protein then enhances expression of the genes listed in Table 1 (218).

The physiological oxidant of SoxR is not known. Initially, it seemed likely that SoxR was activated by superoxide itself, since O₂⁻ is produced by redox-cycling agents and since the induction of SOD is a prominent part of the response. Indeed, SOD mutants, which differ from wild-type cells only in their steady-state concentration of superoxide, exhibit elevated expression of soxS and of members of the regulon (95, 175). However, in these mutants the extent of induction is only a fraction of what can be achieved with redox-cycling drugs, even though the level of superoxide inside them is sufficient to inactivate several metabolic pathways. Conversely, SOD overproduction does not block SoxRS induction by the redox-cycling drug paraquat (95, 172). Further, in other bacteria the SoxR response can be triggered by redox-cycling antibiotics under anaerobic conditions, in which superoxide formation is absolutely precluded (58). Thus it seems that superoxide is neither necessary nor sufficient to activate SoxR. In vitro biochemical studies do not clarify the situation, since, like many redox-active proteins, SoxR can be oxidized by a variety of electron acceptors, including molecular oxygen and even the redox-cycling drugs themselves (202).

The redox state of the SoxR protein represents the balance between its oxidation and reduction, which raises the possibility that redox-cycling drugs might be potent activators because they interfere with the reduction step. Using a screen for constitutive soxS::lacZ expression, Koo, Roe and colleagues determined that the RsxABCDGE and RseC proteins are involved in the reduction of SoxR in vivo (155). They may directly reduce SoxR, since Rsx proteins are homologues of the Rnf proteins that reduce nitrogenase in Rhodobacter capsulatus, while RsxC can catalyze electron transfer from NADPH to model metalloproteins such as cytochrome c. If so, perhaps redox-cycling drugs inhibit SoxR reduction either by directly oxidizing these protein complexes or by depleting the cell of the NADPH that provides electrons to them.

In many other bacteria the Sox response is driven by a single protein, SoxR, which directly stimulates transcription by binding upstream of every gene within the regulon (65, 153, 208, 211). The two-component signal transduction arrangement in E. coli and Salmonella may allow responses to be amplified more strongly, but it also creates a control problem: How is the response turned off? The eventual reduction of SoxR will end the synthesis of SoxS; but, without any active process to inactivate SoxS, the induced synthesis of other proteins in the regulon would continue until SoxS protein were diluted out by cell division—a process that is slow in most natural habitats. This puzzle has been solved by Griffith, Shah, and Wolf (104, 245, 246). They determined that the expression of SoxS-controlled genes ceases soon after the oxidative stress is removed. The key is the exceptionally short lifetime (t_0.5 = 2′) of SoxS in vivo. By dissecting SoxS they discovered that its N-terminal 17 amino acids target it for rapid degradation by the Lon protease. Thus, when oxidative stress subsides and the SoxR-driven synthesis of SoxS stops, the intracellular concentration of SoxS quickly drops, and the now-unnecessary expression of the regulon ends.

SoxS and SoxR each bind upstream of their structural genes and repress expression, thereby moderating both their basal and induced levels of expression. Hidalgo et al. (118, 203) showed that this arrangement has the net effect of enhancing the degree of SoxS induction when the system is activated.

A DNA consensus sequence for SoxS binding has been deduced, but inspection of the genome indicates that these sites are far too abundant to serve as the sole determinant of SoxS recognition. Instead, SoxS first binds to the α-subunit of RNA polymerase, and this complex then scans the chromosome for tandem Sox-box/promoter binding sites, thereby excluding DNA sequences that bind only one or the other protein (182, 244). The genes that comprise the SoxRS regulon have been identified both by classic methods and by microarray strategies (23, 183, 217, 218). They are listed in Table 1, and their roles are discussed below.

The OxyR System

In 1985 Christman et al. (43) reported the isolation of Salmonella mutants that were hyperresistant to hydrogen peroxide. It was quickly determined that the mutation constitutively activated a transcription factor, dubbed OxyR. In wild-type strains OxyR is activated during H₂O₂ exposure, and it stimulates the transcription of at least 20 genes (Table 2 and Fig. 7) (286). Among these is catalase.

Fig. 7Cellular damage from hydrogen peroxide, and OxyR-mediated adaptations. H₂O₂ inactivates the Isc cluster assembly system, degrades exposed iron-sulfur clusters, damages DNA, inactivates the Fur repressor, and oxidizes polypeptides of ferroproteins. Induced defenses (bracketed, dashed lines) include H₂O₂ decomposition by Ahp and catalase, cluster assembly by the Suf system, iron sequestration by Dps, superinduction of Fur, and import of manganese by MntH. Disulfide bond formation, and re-reduction by glutaredoxin and/or thioredoxin, have been proposed but not yet demonstrated to occur at physiological doses of H₂O₂.

Cys-199 of OxyR normally sits in a hydrophobic pocket. However, it is rapidly oxidized by H₂O₂ to a sulfenic acid form, and the increased polarity and size evidently trigger dissociation and substantial conformational movements. These changes place the sulfenic derivative of Cys-199 near Cys-208, with which it condenses to form a disulfide bond (41, 166, 284). Significantly, substitutions for Cys-208 do not eliminate transcriptional responsiveness, indicating that the disulfide bond merely stabilizes a structure to which the sulfenic form of the protein is naturally inclined.

While OxyR binds only in its oxidized form to most members of the regulon, it binds to its own promoter whether it is reduced or oxidized (258). Binding occludes the RNA polymerase binding site, and thus this arrangement allows feedback regulation that establishes the titer of OxyR in the cell. OxyR protein synthesis is not activated during oxidative stress (249).

Purified OxyR can be oxidized by molecular oxygen to its active form, but that reaction is relatively slow, and the regulon is not activated merely by aerobiosis. Instead, hydrogen peroxide per se is the direct physiological inducer. The reaction is very rapid: 100 nM H₂O₂ creates a disulfide bond in OxyR with a half-time of 30 s (8). A similar response was observed in vivo: 30 s after the addition of 200 nM H₂O₂ to culture medium, OxyR protein was predominantly oxidized, and transcription of the regulon had begun. This exquisite sensitivity is a physiological necessity, since submicromolar levels of intracellular H₂O₂ cause substantial damage to DNA and enzymes (140, 210).

After the H₂O₂ stress subsides, glutaredoxin 1 deactivates OxyR by reducing its Cys-199/Cys-208 disulfide bond (284). Interestingly, both glutaredoxin 1 and glutathione reductase are members of the OxyR regulon; thus, the OxyR response is moderated by a negative feedback cycle that presumably allows induction to occur rapidly but to be limited in extent. It is not yet clear whether the redox status of glutaredoxin 1 is perturbed by H₂O₂—if true, this arrangement would provide another layer of control of OxyR activity.

DEFENSIVE STRATEGIES

Scavenging Systems

Cytoplasmic superoxide dismutases.

E. coli and Salmonella each synthesize two cytoplasmic SOD isozymes, one each of the manganese- and iron-cofactored types (MnSOD and FeSOD). Laboratory strains of E. coli additionally express a single copper, zinc-cofactored enzyme (CuZnSOD, also called SodC) in the periplasm (12, 16), while Salmonella strains may have up to three periplasmic CuZnSOD isozymes (SodC1, SodC2, and SodC3) (33, 71, 76). Superoxide is anionic at physiological pH and therefore does not cross membranes (157, 178); thus, the cytosolic and periplasmic enzymes scavenge superoxide that is formed by different sources and that threatens different biomolecules.

The dismutation of superoxide can occur spontaneously through reaction between protonated and anionic superoxide, with a second-order rate constant of 8 × 10⁷ M⁻¹ s⁻¹ (221):

HO₂ + O₂⁻ + H⁺ → H₂O₂ + O₂   (4)

This reaction is not sufficient to establish low intracellular concentrations, however, both because it requires protonation of superoxide (pK_a = 4.8) and because the second-order reaction slows at physiological O₂⁻ levels. In contrast, the enzyme-catalyzed reaction is first-order in superoxide and exhibits a rate constant that approaches catalytic perfection: 10⁹ M⁻¹ s⁻¹ (29, 30).

Given its catalytic efficiency, the titer of SOD inside enteric bacteria is astounding: exponentially growing E. coli contain about 20 μM active cytoplasmic SOD (MnSOD plus FeSOD) (calculated from reference 133). In the face of endogenous superoxide formation (ca. 5 μM/s), the high titers of SOD keep steady-state superoxide levels at about 10⁻¹⁰ M—an average of less than 1 molecule per cell. Yet even at this minute concentration, labile (4Fe-4S) enzymes are still inactivated every 30 min or so, a consequence of the extreme reactivity of their clusters with superoxide (k = 10⁶ to 10⁷ M⁻¹ s⁻¹) (79). Substantial deficiencies in enzymatic activities and in growth rate become apparent if the SOD titer is reduced by more than 50% (95). The implication is that E. coli needs to synthesize so much SOD to protect vulnerable enzymes from endogenous superoxide.

The MnSOD and FeSOD enzymes (encoded by sodA and sodB, respectively) are structurally and kinetically similar. Each is active only with its cognate metal, however, and their roles are to ensure that SOD activity is present under a wide range of metal bioavailability. When iron is abundant, FeSOD is synthesized and is activated by iron, while the iron-charged Fur repressor inhibits transcription of sodA (45). Conversely, when iron is scarce, sodA expression is derepressed and the manganese transporter MntH is induced, leading to high titers of active MnSOD; simultaneously, sodB mRNA is degraded through the action of the RyhB small RNA (185).

MnSOD synthesis is also blocked in anaerobic conditions by the actions of Fnr and ArcA (45, 113). FeSOD synthesis persists. At first blush the latter behavior might seem surprising, since superoxide is not formed inside anaerobic cells; however, anaerobic FeSOD synthesis is necessary to prepare E. coli for subsequent aeration (148). If SOD is not synthesized preemptively, then the superoxide that forms on aeration will immediately inactivate biosynthetic enzymes and block new protein synthesis. In general, FeSOD is the enzyme that evolution favors in anaerobic habitats, probably because in those environments iron is predominantly in the ferrous form and is biologically available. In fact, when experimenters have artificially forced the transcription of sodA in anaerobic cells, very little MnSOD activity resulted, due to the mis-metallation of MnSOD protein by iron (259).

When aerobic cells are treated with redox-cycling drugs, SoxS accelerates the transcription of sodA by an order of magnitude (111). This is a key element of the cellular response to superoxide stress, because sodA mutants are hypersensitive to these drugs (34). The choice of MnSOD rather than FeSOD as the inducible enzyme befits the efforts of the cell to limit intracellular iron levels during periods of oxidative stress (below).

Periplasmic superoxide dismutases.

E. coli K-12 synthesizes a single periplasmic SOD that is encoded by sodC (134). Salmonella express a homologous gene, denoted sodCII. In addition, virulent Salmonella enterica serovar Typhimurium carries a related gene, sodCI, on a pathogenicity island; the common laboratory strain LT2 encodes a third isozyme from sodCIII (76); and E. coli O157:H7 expresses two sodCI homologues from lysogenic lambdoid phages (50). All of these enzymes are CuZnSODs. The total SOD activity in the periplasm can be very high, comparable to that of the cytoplasm (159). None of these genes is responsive to either the OxyR or SoxRS oxidative stress systems.

The E. coli enzyme and its Salmonella SodCII homologue evidently defend unidentified periplasmic targets from superoxide that leaks from respiratory-chain components on the outer aspect of the cytoplasmic membrane (156). These genes are under the control of RpoS and are strongly induced when aerobic cells enter stationary phase (96). Interestingly, whereas expression of E. coli sodC is suppressed in copper-poor medium, the Salmonella enzymes continue to be made and accumulate as apo-enzymes in the periplasm (159). They are quite stable in this form and will spontaneously remetallate even hours later if copper is added to the medium. They can do so because their active site exhibits more flexibility than those of the related eukaryotic CuZnSODs, which require dedicated chaperones for copper loading (47). The fact that these periplasmic SODs obtain their prosthetic metal directly from the medium may explain why these bacteria do not use a periplasmic iron isozyme: iron is scarce in aerobic surface waters, and the high-affinity siderophores release iron only on hydrolysis in the cytoplasm.

The SodCI serovar Typhimurium isozyme is critical for virulence and may help Salmonella withstand the oxidative burst of host macrophages (discussed in a subsequent section).

Peroxidases and catalases.

It is likely that all organisms employ peroxidases (equation 5) and/or catalases (equation 6) to eliminate endogenous H₂O₂.

RH₂ + H₂O₂ → R + 2H₂O   (5)

H₂O₂ + H₂O₂ → O₂ + 2H₂O   (6)

E. coli expresses two catalases, known as HPI and HPII, that are encoded by katG and katE, respectively. Mutants that lacked both enzymes did not exhibit growth defects, which initially called into question the idea that scavenging enzymes were necessary to protect cells against endogenous H₂O₂ (180, 234). However, while katG katE mutants could not degrade millimolar concentrations of H₂O₂, they were subsequently found to retain the ability to degrade H₂O₂ when it was present at low micromolar concentrations (238). This residual activity is due to an enzyme known as alkylhydroperoxide reductase (Ahp). This two-component enzyme had originally been identified as a scavenger of organic hydroperoxides (137). The original studies had not detected activity toward H₂O₂ itself; in retrospect, this result likely reflected the ability of high doses of H₂O₂ to inactivate the enzyme. Recent work has confirmed that Ahp is a two-component NADH peroxidase with a k_cat/K_m of 4 × 10⁷ M⁻¹ s⁻¹ (212).

Ahp consists of a peroxidatic component, AhpC, and an NADH-reducible flavin component, AhpF (219). H₂O₂ oxidizes Cys46 of AhpC to a sulfenic acid, which then condenses with Cys165 to form a disulfide bond. The disulfide bond is transferred through exchange reactions to other sulfhydryl residues on AhpC, and these are ultimately reduced when AhpC binds to reduced AhpF. In vitro the slow step in this process is likely to be the association and dissociation of AhpF and AhpC, rather than the initial reaction of AhpC with H₂O₂. However, the opposite is true under physiological conditions: because AhpC is abundant (ca. 5 μM [37, 170]) and it scavenges ~10 μM/s endogenous H₂O₂, its normal turnover frequency is low (ca. 2/s), and AhpC is predominantly in its reduced form. Thus a premium exists on the rapidity of the first half of the AhpC reaction cycle, which determines the lifetime of intracellular H₂O₂, rather than on the rate of its re-reduction by AhpF.

However, when E. coli enters an environment in which the concentration of H₂O₂ exceeds 20 μM, Ahp activity hits a limit (239), presumably because the re-reduction of AhpC cannot keep up with its oxidation. In addition, it is possible that the sulfenic acid intermediate of AhpC is overoxidized to a catalytically inactive sulfinic acid, as with peroxiredoxins found in other bacterial and mammalian systems (219). This arrangement is probably beneficial, since it ensures that the cell does not expend its entire NADH pool in a futile effort to degrade overwhelming doses of H₂O₂.

Catalases have millimolar (or higher) K_m values for H₂O₂ (119), so their turnover numbers are highest at the doses of H₂O₂ that saturate Ahp (239). Because they dismutate H₂O₂, they do not expend reducing equivalents. For the same reason, catalases continue to work even when metabolism has stalled, and so catalase constitutes the primary scavenging system of stationary-phase cells. Indeed, HPII (encoded by katE) is induced by RpoS to provide this activity (136). To date it is unclear why enteric bacteria employ one catalase in log-phase cells and another in stationary phase.

Three additional peroxidatic activities have been described for E. coli: a periplasmic cytochrome c peroxidase and two thiol-dependent peroxidases. The cytochrome c peroxidase was identified from the resemblance of the yhjA sequence to those of other bacterial cytochrome c peroxidases (213). Interestingly, in both E. coli and these other organisms, the protein is specifically synthesized under anaerobic conditions, due to its positive regulation by Fnr. The investigators have suggested that the enzyme might represent an effort by the cell to use H₂O₂ as an anaerobic electron acceptor.

Thiol peroxidase uses reducing equivalents provided by thioredoxin to scavenge peroxide species (36). Bacterioferritin comigratory protein (BCP) exhibits a similar substrate profile (141). Its catalytic activity is greatest with organic hydroperoxides, and bcp mutants appear to be more susceptible to these species than to H₂O₂ itself. Workers have speculated that Bcp may act on lipid hydroperoxides in vivo. No regulation of these enzymes has been established, and it remains a challenge to understand why bacteria would require enzymes to supplement the activity of Ahp.

Salmonella Typhimurium LT2 strains have an auxiliary manganese-cofactored catalase, encoded by katN, that is induced by the RpoS system (225). Its role is not clear.

Is hydrogen peroxide stress compartmentalized?

Hydrogen peroxide passively diffuses across bacterial membranes at a rate that is substantial but not unlimited (239). Indeed, the robust activity of scavenging enzymes can establish a transmembrane gradient: if the environmental level of H₂O₂ is less than 10 μM, Ahp can scavenge H₂O₂ as quickly as it diffuses into the cell, thereby keeping the intracellular concentration up to an order of magnitude lower than the extracellular concentration. At higher doses of H₂O₂, Ahp is saturated, but catalase can have the same effect if it is first induced to high titers. Conversely, these enzymes scavenge endogenous H₂O₂ so quickly that virtually none of it diffuses out of the cell. Thus, the role of these scavenging enzymes is to protect the individual cell from H₂O₂ stress, whether the H₂O₂ is generated endogenously or flows into the cell from external sources. This analysis contradicts the conclusions that were drawn from earlier experiments that were conducted with millimolar levels of H₂O₂ and noninduced levels of catalase (180).

Exclusion of Redox-Cycling Antibiotics

Redox-cycling antibiotics are secreted by bacteria that seek to suppress the growth of their competitors and by plants that wish to block the growth of invasive microbes. These antibiotics are soluble quinones, viologens, or phenazines that bind nonspecifically to many redox enzymes, abstract single electrons from their solvent-exposed flavins, and then transfer the electron to molecular oxygen. The superoxide that is formed can damage enzymes directly; it also dismutates, enzymatically or chemically, and generates H₂O₂ that itself damages enzymes and DNA. The SoxRS system minimizes the accumulation of redox-cycling drugs in the periplasm by at least two mechanisms. First, the SoxRS system induces synthesis of MicF (42), a small RNA that blocks the translation of the message that encodes the large OmpF porin (55). The net effect is to reduce the mean porin size and to slow the entry of antibiotics (282).

Second, the SoxRS response activates the synthesis of the AcrAB drug-export system (68, 179, 272). AcrAB collaborates with TolC in using the membrane potential to export a wide variety of antibiotics. Transcription of acrAB is also activated by the regulatory proteins Rob and MarA, which respond independently to antibiotic stress. Importantly, the inclusion of micF and acrAB in the SoxRS regulon strongly implies that the environmental circumstance that typically activates SoxRS is the exposure of bacteria to redox-cycling compounds. This conclusion conforms with the observation that SoxRS responds more effectively to antibiotics than to superoxide per se (95).

Protection of Iron-Sulfur Clusters

Repair of clusters.

Iron-sulfur enzymes of the aconitase class are primary targets of both superoxide and hydrogen peroxide (78, 85, 86, 122, 140, 161, 171). The oxidation of the solvent-exposed [4Fe-4S]²⁺ clusters destabilizes them, and they release an iron atom and thereby generate a [3Fe-4S]⁺ cluster that is catalytically inactive (79). Protracted exposure to oxidative stress leads to the loss of the [3Fe-4S]⁺ EPR signal and may indicate that the cluster has decayed to a smaller partial cluster or else disintegrated entirely (63). Enzymes of this family belong to fermentative (6-phosphogluconate dehydratase, serine dehydratase) and oxidative (aconitase, fumarase) catabolic pathways and to the branched-chain biosynthetic pathway; therefore, their inactivation will impede growth under most conditions.

The primary action that E. coli takes against cluster damage is to repair it. In laboratory experiments dehydratase activities returned to their initial values within 5 to 10 minutes of the cessation of superoxide or H₂O₂ stress, even when new protein synthesis was blocked (87, 95). The [3Fe-4S]⁺ EPR signal disappeared simultaneously, suggesting that the clusters were repaired in a single, concerted process (63). A plausible repair process would involve the one-electron reduction of the cluster (to the [3Fe-4S]^o state) and its subsequent remetallation by ferrous iron. Indeed, full activity returns when oxidized dehydratases are incubated in vitro with dithiothreitol and ferrous iron. Strains that lack iron-storage proteins cannot repair clusters if chelators block iron uptake, which indicates that the repair process involves new iron atoms rather than the same ones that were lost during cluster damage (152). This repair process is evidently distinct from the de novo cluster-assembly process, however, since both isc and suf mutants efficiently repair damaged [3Fe-4S]⁺ enzymes (63).

Genetic evidence suggests that YtfE, NfuA, and/or YggX proteins are involved in repair. YtfE was identified as a protein that is induced when the NsrR repressor is inactivated by nitric oxide stress (147). Mutants were constitutively defective in the activities of enzymes that contain iron-sulfur clusters, and this deficiency particularly affected their ability to use anaerobic respiratory substrates. Thus, YtfE may have a role in cluster assembly aside from a protective function during oxidative stress. Yet ytfE mutants also recovered poorly from aerobic H₂O₂ exposure, and their extracts did not repair the clusters of oxidized dehydratases. YtfE purified as a di-iron protein that is probably in the Fe(II)-Fe(II) resting state in vivo. Therefore, it conceivably might serve as a donor of either iron or electrons in the repair process. More work is required to resolve this.

NfuA is a homodimeric protein that coordinates a [4Fe-4S] cluster, which it can transfer to apoaconitase (6). Paraquat inhibits the growth of mutants that lack this protein (encoded by the yhgI), and so it has been suggested that NfuA assists in the reactivation of oxidized (4Fe-4S) enzymes.

YggX belongs to the SoxRS regulon, and mutants that cannot induce it are hypersensitive to paraquat and rapidly lose dehydratase activity (98). The yggX mutants still repair clusters, but they do so more slowly than do wild-type cells. Workers speculated that YggX might be the iron donor for the repair process, and a variety of in vivo data are consistent with that idea. However, biochemical studies of the protein have detected only weak interaction with coincubated iron (46, 205). It remains possible that the latter experiments lacked an iron-binding metabolite that is a cosubstrate for YggX. In summary, cluster repair is essential for tolerance of H₂O₂ and superoxide stress, but the process has not yet been solved on the molecular level.

Induction of the Suf system.

The assembly of [2Fe-2S], [3Fe-4S], and [4Fe-4S] iron-sulfur clusters inside bacteria can be catalyzed by the Isc and Suf protein complexes (10, 144). Recent work suggests that the CsdA-CsdE proteins may have this capacity, too, although no phenotype has yet been observed in mutants (176). The Isc system is the primary source of iron-sulfur clusters under routine growth conditions: mutations in any of these genes cause the diminution of activities of cellular enzymes that contain Fe-S clusters (236, 257). The proteins that make up this system are encoded on adjacent iscRSUA and hscB-hscA-fdx-iscX operons. The former operon is negatively regulated by IscR, a repressor that itself is activated by a coordinated [2Fe-2S] cluster; in this way feedback repression controls the synthesis of cluster-generating enzymes (237).

The sufABCDSE operon encodes a second assembly system (251). Expression is very low during routine growth, due in part to its repression by Fur protein. Expression increases when cells are starved for iron, both because Fur is deactivated and because the operon is induced by apo-IscR (207, 278). The suf genes are very strongly induced when cells are exposed to H₂O₂, and in this case the deactivation of Fur, binding of apo-IscR, and stimulation by OxyR all contribute to its expression (167, 207, 286). Phenotypic analysis and enzyme assays show that suf mutants are defective in the activities of Fe/S enzymes in the presence of iron chelators (207) or of H₂O₂ (195) (S. Jang and J. A. Imlay, unpublished data). The implication is that the Suf machinery can function during iron starvation and/or H₂O₂ stress, whereas the Isc machinery cannot (Fig. 7). It is not clear how H₂O₂ disrupts the Isc system, although it is plausible that H₂O₂ oxidizes nascent clusters as they are assembled on the IscU scaffold.

By restoring de novo cluster assembly during periods of H₂O₂ stress, Suf preserves the function of all classes of Fe/S enzymes. In addition, because dehydratase clusters can degrade beyond the [3Fe-4S] state during protracted stress, it is likely that the Suf system repairs apoproteins that are formed in this way, too.

Induction of oxidant-resistant isozymes.

When repair processes cannot keep up with the pace of cluster damage, a strategy of last resort is the induction of oxidant-resistant isozymes. The first example was fumarase C, a cluster-free homologue of the mitochondrial fumarase. The fumC gene lays immediately upstream of fumA, encoding the primary aerobic fumarase, but fumC has a SoxS binding site that triggers its strong induction when cells are exposed to redox-cycling antibiotics (171).

Aconitase A is a minor isozyme that is also induced to high levels during SoxRS induction (107). Purified aconitase A, like aconitase B, has an iron-sulfur cluster that is rapidly destroyed by superoxide and H₂O₂ (146); however, the enzyme appears to be protected in vivo and in concentrated cell extracts (265). The mechanism of protection is unknown.

Together the induction of fumarase C and aconitase A can allow the TCA cycle to function during periods of oxidative stress. This elegant strategy raises an interesting question: Why doesn't E. coli constitutively replace its unstable fumarase A and aconitase B with their resistant isozymes? The obvious hypothesis was that the resistant enzymes are poorer catalysts, but kinetic studies indicate that this is untrue (80, 146), and, in fact, mammalian mitochondria operate with only a fumarase C homologue. Another intriguing possibility is that cluster instability serves an important purpose. In mammals and some bacteria, the aconitase cluster disintegrates when intracellular iron levels fall, and the apoenzyme that is formed is an RNA-binding protein that coordinates the cellular response to iron starvation. Guest and colleagues have suggested that a similar system may operate in E. coli (254). In addition, the loss of aconitase activity should cause the accumulation of citrate—an excellent iron chelator that can be exported (265), where one imagines it may function as an iron siderophore, using the dedicated Fec import machinery. In this light the vulnerability of aconitase to oxidants may be the collateral cost for the use of its labile cluster as an iron sensor. However, an analogous rationale for the retention of fumarase A is not yet apparent.

Protection of DNA

Hydroxyl radicals react with virtually all organic biomolecules at nearly diffusion-limited rates, and for this reason it is likely that they react very near the site at which they are generated. For this reason it is not possible for organisms to develop a mechanism that scavenges them: an impossibly high concentration of the scavenger would be necessary for it to contact the radical before the radical collides with a biomolecule. Therefore organisms try to minimize the frequency at which hydroxyl radicals are generated, and they employ repair mechanisms to reverse any damage that does occur.

Control of unincorporated ferrous iron.

The formation of hydroxyl radicals can be minimized by controls on the level of intracellular iron. This is the primary function of Fur protein, the transcriptional repressor that is activated by the binding of ferrous iron (200). When metallated, Fur blocks synthesis of the three complete iron-import systems of E. coli: the Feo complex that imports ferrous iron, the Fec system that imports citrate-chelated iron, and the high-affinity Ent system that imports iron chelated to the siderophore enterochelin. In natural habitats E. coli can also use transporters that recognize siderophores that are excreted by other microorganisms, and the synthesis of these transporters is also controlled by Fur.

Both the complex and the defined media that are used by most laboratories provide enough iron to metallate Fur and thereby substantially repress synthesis of iron-import systems. Cells grown in defined medium contain about 20 μM unincorporated iron, while those grown in iron-rich complex medium contain up to 100 μM (151, 210). Deletion of the fur gene elevated intracellular iron levels 5- to 10-fold. As a consequence, fur mutants suffer proportionately more oxidative DNA damage (151, 260).

Expression of the fur gene is induced by both the OxyR and SoxS proteins (285). Positive control by OxyR is evidently required because H₂O₂ tends to inactivate the ferrous-Fur complex, perhaps by oxidizing the ferrous iron cofactor (Fig. 7) (264). In this way the presence of H₂O₂ might be mistaken for the absence of iron; if not corrected, the natural result would be up-regulation of iron-import systems—which might be a catastrophically inappropriate response during periods of H₂O₂ stress. The OxyR-mediated induction of high levels of Fur protein is apparently an effective compensatory response. Indeed, when fur induction was blocked by mutations that interrupted its OxyR binding site, micromolar H₂O₂ triggered the induction of iron-import systems, the cell became overloaded with free iron, and the toxic amounts of DNA damage were generated. The significance of the regulation of fur by SoxRS has not yet been tested.

When iron is pumped into the cytoplasm by its import proteins, it is likely (although not yet shown) that these proteins transfer the iron atoms directly to dedicated trafficking proteins or metabolites. A failure to do so would result in unproductive associations between released iron and the charged surfaces of lipids, nucleic acids, and proteins, which would both be wasteful and make these biomolecules vulnerable to the localized production of hydroxyl radicals. IscA, SufA, YggX, and CyaY are among the proteins that may deliver iron to enzymes that require it (11, 60, 99, 177, 268).

If more iron has been imported than is necessary for enzyme metallation, then the excess iron can be stored in ferritins. E. coli and Salmonella synthesize three iron-storage proteins during routine growth: ferritin A, ferritin B, and bacterioferritin. Both ferritin A and bacterioferritin store iron as a hydrated ferric oxide crystal; 24 subunits of the ferritin protein delineate an 8-nm cavity that can accommodate up to 4,500 iron atoms. In this way the iron can be saved for later use in iron-poor conditions, and yet it is sequestered in a form that cannot react with hydrogen peroxide. Workers have not yet distinguished the roles of ferritin A and bacterioferritin. Interestingly, the toxicity that E. coli fur mutants experience from excessive iron import can be compensated for by the engineered oversynthesis of ferritin A, while Salmonella fur mutants are protected by overproduced bacterioferritin (260, 267). Meanwhile, the function of ferritin B is even less understood. This protein lacks the ferrooxidase centers that the other storage proteins use in converting ferrous iron to the stored ferric form, which has led to the suggestion the ferritin B may maintain iron in a reduced form that is ready to metallate proteins (267).

Dps, a fourth iron-storage protein (100, 129), is strongly induced by RpoS during cell starvation and by OxyR during H₂O₂ stress (3). In the latter situation iron sequestration has an obvious purpose: to suppress hydroxyl-radical formation from Fenton chemistry (Fig. 7). Dps-iron complexes differ from ferritin A complexes in that they feature 12 rather than 24 subunits at their vertices. Also, Dps appears to use H₂O₂ instead of molecular oxygen as the electron acceptor during iron oxidation (283). The latter feature may be significant not because it consumes H₂O₂, but because it offers a mechanism whereby the storage function may be deactivated when H₂O₂ stress has ended. Dps mutants are hypersensitive to H₂O₂ (184); in fact, catalase/peroxidase mutants are unable to grow in aerobic environments if the dps gene is inactivated, because the endogenous H₂O₂ creates overwhelming amounts of DNA damage (210).

Interestingly, Dps has a second biochemical activity that appears to have an important role: purified Dps can bind the chromosome into a tight, crystalline array (2). This lattice has been visualized by microscopy inside stationary-phase E. coli, which indicates that the activity is not artifactual (83). In a general way this process inside starved cells resembles the condensation of chromosomes in sporulating bacteria, and it has been suggested that this action may similarly shield the DNA of nongrowing cells from chemical damage—including DNA oxidation. However, this DNA-binding function of Dps may be wholly independent of its iron-sequestration function, and the H₂O₂ sensitivity of log-phase dps mutants can be complemented by mutant Dps or heterologous Dpr proteins that can sequester iron but lack DNA-binding activity (35, 210). Further, the action of Dps inside H₂O₂-stressed cells does not block general transcription, which it certainly would do if Dps were to crystallize DNA so tightly as to exclude chemical agents. At this point the indication is that Dps has acquired two DNA-protecting roles that are mechanistically independent.

Repair of oxidative DNA damage.

Because mutagenesis has an outsized impact on human health, the mechanisms of DNA repair became the subject of intense study as early as the 1960s. At that time E. coli was a model system that was uniquely accessible to both genetic and biochemical investigation, and so the DNA-repair field has a deep history in this organism. Initial experiments commonly employed far-UV radiation as a damaging agent, because it is specific and easily manipulated. By the 1980s, however, attention shifted toward stressors that are more likely to be encountered in natural habitats, including oxidants.

DNA has a substantial affinity for ferrous iron, and so it is a locus for Fenton chemistry (116, 132). The hydroxyl radicals that are formed are powerful oxidants (Em = +2.3 V) that abstract electrons from methyl, methylene, or alcoholic carbon atoms or that add to the unsaturated bonds of aromatic bases. The immediate products are DNA radicals. In some cases the radical resolves directly; in others, the addition of molecular oxygen creates a peroxyl radical adduct that undergoes further reaction. Scores of distinct base and ribose oxidation products have been detected (61, 128). Many are inadequate templates for the replicative DNA polymerase III and therefore constitute replication blocks; others base-pair with noncomplementary nucleotides and thus are mutagenic.

The view that has emerged is that base excision repair (BER) constitutes the primary mechanism for the removal of oxidized bases (22, 62, 270). Glycosylases hydrolyze the amine linkage between base and ribose, generating a baseless (AP) site. Some of these enzymes also nick the DNA strand 5′ or 3′ to the sugar. An AP endonuclease then cleaves the chain 5′ to the linking phosphate, leaving a 3′-hydroxyl group that serves as a primer for repair synthesis 5′ to the lesion. The gap is then filled by the repair polymerase, DNA polymerase I, with any residual lesion fragment being removed either by the Pol I 5′-to-3′ exonuclease or by a ribosyl-removing activity of dedicated enzymes. The intact strand is then generated by ligation.

This scheme requires that the cell contain glycosylases that can address an enormous number of lesions that are chemically and structurally diverse. The economical solution to this problem is that these enzymes scan the DNA searching not for a particular lesion but for the loss of duplex structure. Endonuclease III and endonuclease VIII, for example, both recognize the disruption of the helix that results from saturation of pyrimidine residues. They hydrolyze overlapping subsets of the principle cytosine and thymine oxidation products; consequently, the nth nei double mutants are hypersensitive to H₂O₂ (142, 229).

Guanine residues are a particular target of DNA oxidation because of their ability to tunnel electrons to nearby base radicals (31). The primary product is 8-hydroxyguanine (8-oxoG). This lesion is highly mutagenic both because of its ability to mispair with adenine and because the 8-oxoG/A mispair presents to the polymerase proofreading system an O8 atom that mimics the O2 and N3 hydrogen-bond acceptors of cognate base pairs (123). G-to-T transversions result. To minimize this mutagenesis, E. coli elaborates three processes. Formamidopyrimidine DNA glycosylase (Fpg, also known as MutM) excises oxidized purines prior to replication (256). If lesions are misreplicated prior to excision, MutY protein excises adenine from these mispairs (191).

In addition, 8-oxoG can also arise in DNA by the misincorporation of 8-oxodGTP from a cellular pool of oxidized nucleoside triphosphates. To avoid this problem, MutT is an enzyme that cleanses the pool by hydrolyzing 8-oxodGTP to 8-oxodGMP (81). Mutations in mutT substantially elevate the rate of aerobic mutagenesis, suggesting that the ability of nucleotides to bind ferrous iron may make them significant targets of Fenton-generated hydroxyl radicals.

MutY is a monofunctional glycosylase that leaves a baseless site for further processing by other enzymes. In contrast, endonuclease III, endonuclease VIII, and Fpg additionally act as endonucleases that incise DNA adjacent to the oxidative lesion. Endonuclease VIII and Fpg both leave a 3′-phosphate at the terminus of the priming strand, while endonuclease III leaves an open ribose fragment. In all cases a 3′-hydroxyl end must be restored by either exonuclease III or endonuclease IV (57). These two enzymes also nick at urea residues formed by decomposition of oxidized thymines and at strand breaks that ensue from the direct oxidation of the ribose moiety. Their involvement in the repair of this wide variety of lesions is reflected in the extreme sensitivity to hydrogen peroxide of mutants that lack exonuclease III (56). Single mutants lacking endonuclease IV are not particularly sensitive, presumably because it contributes a minor fraction of the combined activity of these two enzymes (49). Indeed, the double mutants accumulate strand breaks, exhibit synergistic sensitivity to hydrogen peroxide, and grow poorly in aerobic medium.

Role of the SOS response.

The BER systems described above are inadequate when damage is profuse or when lesions appear that are not BER substrates, such as protein-DNA cross-links. Persistent lesions of these types trigger the SOS response. This back-up system includes the UvrABC excinuclease, which removes a wide variety of bulky lesions; SfiA, an inhibitor of septation that increases the time that is available for DNA repair; several “error-prone” polymerases that allow the replication fork to proceed through unrepaired lesions; and recombinational-repair proteins, such as RecA, that enable the repair of strand breaks that arise when the replication fork skips past lesions. Accordingly, hydrogen peroxide is highly toxic to lexA3 mutants that cannot activate the SOS system (131). The largest part of this sensitivity derives from the absence of recombinational repair.

Oxidative stress is mutagenic. Some lesions, such as 8-oxoG, stimulate mutagenesis by base pairing with noncognate bases; other lesions that strongly block Pol III progression in vitro are regarded as unlikely to be mutagenic. Accordingly, many of the latter lesions were lethal rather than mutagenic when they were incorporated into test plasmids and transformed into cells (270). However, it is conceivable that these experiments do not fully reflect the situation in oxidatively stressed cells, where DNA damage accumulates to a level that triggers the induction of three bypass polymerases: Pol II (encoded by polB), Pol IV (dinB), and Pol V (umuDC) (196). One experiment indicated that mutagenesis in peroxide-treated cells did not require Pol V (131), but further work is needed to clarify the situation. Interestingly, the OxyS small regulatory RNA, which is induced as part of the OxyR regulon, can suppress mutagenesis, but no mechanism has yet been identified (4, 93).

Is oxidation a primary cause of DNA damage?

Because hydrogen peroxide is continuously generated inside aerobic cells, oxidative damage has long been suspected to be the predominant source of DNA lesions in otherwise unstressed cultures. As mentioned, strains that are multiply deficient in general repair pathways—xth recA, recB polA, and even xth nfo mutants—grow well only in anaerobic habitats (49, 130, 192). Moreover, the mutation rates of wild-type strains are typically highest when cells are cultured aerobically, and the spectrum of mutations is dominated by those that are associated with oxidative stress (230).

These data indicate that in aerobic habitats oxidative damage is abundant enough to be a prime driver of mutagenesis and, were it not for the action of repair processes, to kill the cell. Perhaps for this reason most oxidative repair enzymes are constitutively expressed in aerobic cells, rather than being induced only under conditions of extreme oxidative stress. No DNA repair enzyme belongs to the OxyR system. In fact, Fpg, MutY, endonuclease III, and endonuclease VIII each belong to constitutive operons that contain genes of unrelated functions (90). Each is modestly induced when anaerobic cultures are aerated, but none is specifically responsive to hydrogen peroxide, gamma radiation, or paraquat.

In contrast, endonuclease IV is part of the SoxRS regulon, and exonuclease III is strongly upregulated by RpoS when cells enter stationary phase (38, 66, 103, 163, 261). Endonuclease IV is particularly adept at repairing the C4-keto-C1-aldehyde baseless sites that are generated by bleomycin (49), and one might speculate that its induction by SoxRS suppresses the toxicity of specific damage that is generated by such oxidative antibiotics. The induction of exonuclease III during entry into stationary phase—at the same time that the glycosylase titers diminish—suggests that the predominant repair mechanisms may shift when growth stops, but this idea has not been tested by experiment.

Protection of Polypeptides

Proteases and methionine sulfoxide reductase.

When amino acids are oxidized, both their size and hydrophobicity are changed, and so higher-order protein structure can be disrupted. Disordered proteins are potential foci for aggregation and can also become covalently cross-linked to one another; it is possible that these events, rather than the creation of metabolic bottlenecks, may be the primary hazard of protein oxidation. If so, proteases may be important defenses. Indeed, Fredriksson et al. (82) found that carbonylated proteins accumulate in mutants that are deficient in HslVU and Lon.

Methionine sulfoxide and cystine disulfide bonds represent the only two adducts that can be corrected without degradation of the oxidized protein. Two methionine sulfoxide reductases each reduce one of the two enantiomers of the oxidized residue (105):

MetSO + reduced thioredoxin → Met + thioredoxin disulfide + H₂O   (7)

In many bacteria the genes that encode the two enzymes are adjacent to one another or are even distinct domains in a common protein; however, in E. coli they are not genetically linked. Mutants that lack msrA were hypersensitive to H₂O₂ and paraquat (193). They also exhibited defects in protein secretion due to the oxidation of a methionine residue in the signal recognition particle complex (69). In the latter experiments exogenous oxidants were not provided, indicating that endogenous oxidants continually damage proteins.

Notably, the msr genes are not controlled by either OxyR or SoxRS. Perhaps the oxidation of methionine in vivo is rarely due to H₂O₂ or O₂⁻ stress. However, this inference might be misleading—these protein-repair enzymes may simply follow the pattern of DNA-repair systems, many of which are important in fending off oxidation but lie outside oxidant-inducible regulons. The reason may be that repair must continue beyond the period of damage, making it disadvantageous to limit the synthesis of repair enzymes to the time that the oxidants are actually present.

Role of manganese.

Studies have shown that a variety of microbes are most resistant to superoxide and/or H₂O₂ stress when the intracellular concentration of manganese is high (1, 7, 39, 51, 126, 135, 149, 220, 227, 231, 262). Therefore, it was not a surprise when OxyR was found to induce mntH, which encodes a pmf-driven manganese import system in both E. coli and Salmonella (149). Mutations in mntH block the aerobic growth of catalase/peroxidase mutants, confirming that intracellular manganese is essential during periods when the intracellular H₂O₂ concentration is as low as 0.5 μM (6a).

The role of the manganese is less clear. Workers have noted that manganese is not only a cofactor for MnSOD, but it retains its redox activity when it is bound to some metabolites. In this arrangement it can chemically scavenge both superoxide and H₂O₂ (17, 102, 127). However, the critical activity of manganese in the catalase/peroxidase mutant is evidently not to scavenge oxidants, since these cells are not under superoxide stress, and direct measurements show that the intracellular H₂O₂ concentration is not affected by their manganese content (6a). The catalase/peroxidase mutants suffer a high level of protein carbonylation if mntH is eliminated. This observation suggests that manganese may displace iron from enzymes that might otherwise be oxidized through local Fenton reactions (Fig. 7). Unlike iron, manganese cannot reduce H₂O₂ to a hydroxyl radical. A variety of E. coli enzymes nonspecifically employ divalent metals as cofactors; for example, isocitrate dehydrogenase and transketolase use metals to coordinate substrate and to bind thiamine pyrophosphate, respectively, and in vitro experiments show that either iron (II) or manganese (II) suffice (194; Sobota and Imlay, unpublished). During routine growth intracellular iron is abundant and manganese is scarce, which suggests that iron is likely to be the predominant cofactor for many such enzymes. However, in the presence of H₂O₂, iron-activated enzymes can quickly be damaged, whereas manganese-activated enzymes will retain activity. Thus, in the face of H₂O₂, the coordinate sequestration of iron by Dps and import of manganese by MntH is a plausible two-pronged strategy to ensure the continued function of these enzymes. This model needs further testing.

The RpoS Response

The RpoS response is activated under a wide variety of stress conditions that impede bacterial growth, including the exhaustion of nutrients; the signaling mechanisms are reviewed elsewhere in this volume. The response stimulates the expression of more than 100 genes. Prominent among these are genes whose products protect cells from oxidative stress, including katE, katG, ahp, sodC, fumC, acnA, dps, fur, gor, and xthA (48, 66, 214, 271). Nongrowing cells lack the energy and metabolic flux to induce robust responses; therefore, the RpoS system, which is activated before growth has fully stalled, may constitute a prudent effort by which bacteria preemptively set up defenses against any environmental stresses that might later come their way.

One might expect that oxidative stress would also activate the RpoS system in vegetative cells, by interfering with metabolism and blocking growth. However, it appears that the oxyS sRNA may suppress this response (281). Perhaps when a precise OxyR-based response is possible, a broad-based stress response is imprecise and unnecessary.

IS THE PHAGOLYSOSOME AN OXIDATIVELY STRESSFUL ENVIRONMENT?

Phagocytes are a key component of the innate defense against bacterial pathogens. Endocytosis of bacteria is normally followed by the assembly in the phagosomal membrane of NADPH oxidase (Phox, or NOX2), which releases large fluxes of superoxide into the bacteria-containing vacuole. The dismutation of superoxide generates H₂O₂, and it has long been considered likely that the resultant stew of ROS plays a key role in the inactivation of the captive bacteria. The additional release of reactive nitrogen species and of antimicrobial peptides provide complementary mechanisms of toxification.

In neutrophils the oxidative burst is especially robust, and the additional production of hypochlorous acid confronts bacteria with a cell-permeable oxidant that can damage a broad range of biomolecules (115, 226). In macrophage phagosomes the release of ROS is more moderate but still substantial; nevertheless, Salmonella species are among a handful of pathogens that actually exploit this organelle for replication and dissemination. How they manage this feat in the presence of oxidants is currently a focus of spirited investigation.

Investigators do not yet have the ability to measure the concentrations of superoxide and hydrogen peroxide that are achieved in phagosomes. It is instructive, however, to calculate approximate values, using the assumption that the accumulation of superoxide and H₂O₂ is restricted only by superoxide dismutation and by diffusion of H₂O₂ across phagosomal and bacterial membranes, respectively. Superoxide can be formed in neutrophils at a rate of several millimolar per second (143, 275); the oxidative burst inside macrophages is less intense (198). If one assumes that the rate of superoxide formation in macrophage phagosomes is 10% that of neutrophils—that is, about 0.5 mM/s—then one calculates that the O₂⁻ concentration may rise to 50 μM in a phagosome at pH 7.4. At lower pH values the rate of spontaneous dismutation increases, so that in the acidified phagosome the calculated steady-state superoxide levels would be 10 μM at pH 6 and 2 μM at pH 4.5. Interestingly, the concentration of protonated superoxide (HO₂.)—which is likely to be a more potent oxidant of organic molecules—will be only 0.1 μM at pH 7.4 but 4 μM at pH 4.5. The impact of these extracellular O₂⁻ and HO₂· doses on bacteria is unknown, since most laboratory methods cannot create such high concentrations. Enzymatic systems that generate superoxide in vitro typically accumulate less than 1 μM steady-state O₂⁻ and 1 nM HO₂.

H₂O₂ is quickly cleared from phagosomes by diffusion across membranes, and so despite its rapid rate of formation its steady-state concentration should rise no higher than 1 to 4 micromolar. (This calculation presumes that the phagosomal volume is approximately twice that of a captive bacterium.) The concentration will be at least five-fold lower in the cytoplasms of bacteria that are exposed to these doses, due to scavenging by catalase and peroxidase (239). The resultant sub-micromolar concentrations approximate those that induce OxyR, stimulate DNA damage, and inactivate the aromatic biosynthetic pathway in E. coli, but they fall short of the concentrations that debilitate iron-sulfur enzyme activities. Recalling that the calculated oxidant concentrations are imprecise due to the substantial assumptions made in deriving them, it is fair only to conclude that the notion of H₂O₂ stress inside endocytosed bacteria is plausible but not a foregone conclusion.

The vitality of pathogenic Salmonella inside the macrophage phagosome depends on the SPI2-encoded type III secretion system, which injects multiple virulence proteins into the host cell. These proteins disrupt the maturation of the phagosome, including the delivery of the NADPH oxidase and nitric oxide synthase to the phagosomal compartment (266). The efficiency of this system is unlikely to be complete, however, because phox mutations enhance Salmonella virulence in both murine and human hosts (94, 188). Still, if Salmonella substantially abates ROS formation in this way, oxidative stress might be even less intense than calculated above.

Workers have suspected that bacterial defenses against ROS may provide additional mechanisms to cope with the potential oxidative stress in the phagosome. This hypothesis has been evaluated in three ways: by comparing the genetic repertoire of non-pathogenic enterics with those that inhabit macrophages; by evaluating whether bacteria inside macrophages induce antioxidant genes to high levels; and by appraising whether mutations that disrupt these defense systems diminish virulence. The interpretation of results is not straightforward, but the data suggest that oxidative stress may not be as severe as had been expected.

Stress from Hydrogen Peroxide

Virulent serovars of Salmonella generally possess the same consortium of antioxidant genes as do their well-studied nonpathogenic relatives, including E. coli K-12, suggesting that the phagosome has exerted little evolutionary pressure to amplify these defenses. (An exception, discussed below, is the appearance of a second isozyme of periplasmic superoxide dismutase.) Mutations that eliminate key basal defenses against oxidants—such as cytoplasmic superoxide dismutase activity (M. Craig and J. M. Slauch, unpublished results) (266), Dps protein (110), and recombinational repair (27, 32)—substantially reduce the ability of the bacteria to propagate in macrophages or in an infected mouse. Yet these mutations also diminish growth rates in standard aerobic laboratory media, meaning that those results per se do not prove that the phagosome constitutes an unusually stressful environment. The growth defects of recBC strains in macrophages are partially alleviated by phox mutations, consistent with the model that recombinational repair is specifically required to address lesions produced by NADPH-oxidase-derived H₂O₂; however, perplexingly, sbcB mutations that restore repair and H₂O₂ resistance in vitro (130) still did not permit growth in macrophages. Recent data indicate that virulence is not diminished by a lexA3 mutation; implicitly, oxidative DNA damage does not rise to the level that requires the SOS-encoded DNA-repair and lesion-bypass functions.

Other genes that are essential for H₂O₂ resistance in vitro apparently are dispensable during Salmonella infections. Most strikingly, this bacterium was not attenuated by the losses of oxyR (255), fur (84), ahpCF (255), or (in combination) katE and katG (28). In fact, microarray analysis of Salmonella gene expression inside macrophages indicates that catalases—and so presumably the OxyR regulon—are not induced (67). Further, while very low doses of H₂O₂ are sufficient to poison aromatic biosynthesis in laboratory cultures of E. coli, genetic studies show that this pathway remains functional during serovar Typhimurium infection (124).

Interestingly, pathogenic E. coli required oxyR for full virulence in a urinary tract infection model, but the requirement pertained to Phox⁻ as well as wild-type hosts (145). Implicitly, OxyR induction was probably important for entry into the habitat, rather than for specific resistance to the H₂O₂ that is generated by phagocytes.

Mutations that eliminate manganese import (in sit and mntH) (25, 139, 280) and base-excision repair enzymes (xth nfo and nth nei) (251, 252; M. Craig and J. M. Slauch, personal communication) diminish Salmonella survival in macrophages, and these mutants compete poorly with wild-type Salmonella during animal infections. The slight attenuation that the manganese-import defects confer is not conclusive evidence for oxidative stress, since these mutations can also diminish growth rates in iron-poor nonoxidative environments (101). The DNA repair mutants exhibited only minor defects in vitro, which makes a more compelling argument that their problems in macrophages stemmed from host-derived H₂O₂. Furthermore, those defects were substantially suppressed by a host phox mutation.

Thus, these studies have provided mixed evidence that Salmonella experiences significant H₂O₂ stress when it multiplies in the phagosome. Interpretation is complicated by the possibility that Salmonella might simply delay growth until the production of ROS wanes, since nongrowing bacteria are able to survive an extended period of H₂O₂ exposure (130). Indeed, the involvement of the RpoS system in virulence (70) suggests that there may be some period of bacteriostasis during infection.

Stress from Superoxide

Superoxide cannot cross membranes unless it is protonated, and so it is unlikely that phagosomal superoxide creates cytoplasmic superoxide stress in captive bacteria. Genetic studies support this inference (45a). However, high doses of extracellular superoxide might pose a chemical threat to extracytoplasmic biomolecules. Indeed, the sole apparent genetic accommodation to the potential oxidative stress of the phagosome is the presence in many pathogenic Salmonella serovars of an additional periplasmic superoxide dismutase, denoted SodCI. The enzyme is encoded by a gene carried on the lambdoid Gifsy-2 prophage (75). Like other genes important for virulence, sodCI is controlled by the PhoPQ two-component regulatory system and is strongly upregulated within macrophages (91, 263). In this environment the direct activator of PhoPQ may be a combination of low pH and antimicrobial peptides (9).

This regulatory arrangement suggests that sodCI might play an important role in pathogenesis, and indeed sodCI mutants of serovar Typhimurium are substantially attenuated (5, 54, 73, 159, 263). The sodCII gene, in contrast, is a homologue of the housekeeping sodC genes found in nonpathogenic organisms, and it is neither inducible by PhoPQ nor essential for virulence. Promoter-swapping experiments indicated that the structure of SodCI makes it more suitable for an infecting bacterium than the structure of SodCII is, but the key physical distinction has not yet been identified (160). Workers have inferred that superoxide released by the NADPH oxidase can injure the cell surface or periplasm of captive bacteria and that the housekeeping enzyme is an inadequate defense. The nature of the putative injury is not known. At present the sodC example presents perhaps the most compelling evidence that toxic doses of oxidants accumulate in the macrophage phagosome.

Finally, it is appropriate to place these Salmonella studies into the context of the larger question of how macrophages kill bacteria. Segal and colleagues have proposed that the oxidative burst is not aimed at poisoning target cells with ROS; instead, they suggest that the oxidase activity merely provides charge compensation that enables an influx of cations into the phagosome, where they activate nonoxidative killing processes (241). This hypothesis is controversial (e.g., 195); but even if it is correct, it still carries the implication that phagocytosed bacteria should be exposed to high doses of O₂⁻ and H₂O₂, which might be expected to have some toxic effect. The Salmonella studies may have limited utility in resolving this issue, since the virulence of Salmonella means, by definition, that it eludes the killing mechanisms that are successful with most bacteria. For example, Schlosser-Silverman et al. (235) reported that macrophages were mutagenic to endocytosed E. coli but not to Salmonella. Perhaps this problem will be resolved by cell biological and biochemical studies of the fate of phagocytosed nonpathogens.

PROSPECTS

Without any doubt, E. coli and Salmonella have served as the most profitable model systems in the exegesis of oxygen toxicity. Through them we have learned that intracellular superoxide and H₂O₂ are created at steady rates inside aerobic cells, that scavenging enzymes are necessary to avoid a consequent disruption of cell processes, that the destruction of iron-sulfur clusters and the oxidation of DNA are among the primary effects of these oxidants, and that a variety of secondary strategies are essential to cell health during periods of oxidative stress. What's next? Numerous details remain unsettled: how superoxide disrupts sulfur metabolism, the identity of the signal that SoxR senses, whether oxidants can directly damage lipids, and so forth. These issues have already been posed in the main text. However, some other large questions should not be overlooked.

What OxyR- and SoxRS-independent cell functions are essential for oxygen tolerance? To what extent does oxidative stress drive protein turnover and mutagenesis, relative to other forms of chemical damage? Are reactive oxygen species key constraints on growth rate and primary sources of cell death in natural environments? Do other oxidants—hypohalous acids, peroxynitrite, and singlet oxygen, for example—play important roles?

Finally, one must consider the role of E. coli as a model system. Moselio Schaechter and Fred Neidhardt introduced the first volume of this series by commenting, “All cell biologists have two cells of interest: the one they are studying, and Escherichia coli!” The complementary sentiment is that investigators of E. coli owe it to the larger community to define the extent to which what they learn is also true of other organisms. In the genomic age this task has become far more manageable; initial efforts in the field of oxidative stress have revealed much overlap but also intriguing differences. A challenge for the next generation of microbiologists is to understand the bases of those differences and the impact they have in shaping microbial lifestyles and communities. The larger biological community should also stay tuned: higher organisms are subject to the same fundamental oxidative chemistry that is being solved in microbes, and they have also inherited homologues or analogues of most of the defensive stratagems.