Transcription factors that defend bacteria against reactive oxygen species

The threat of H₂O₂

H₂O₂ arises in natural habitats through reactions between sulfur and oxygen at oxic/anoxic interfaces, the photochemical reduction of oxygen by chromophores, and the redox-cycling of pigments. More ominously, plants, animals, and certain bacteria excrete H₂O₂ to poison local microbes (5, 27, 77, 96). H₂O₂ is an ideal weapon. Because it is small and uncharged, it passively crosses membranes at rates similar to that of water, and so it cannot be excluded by the targeted cell (95, 113). Once inside it disrupts multiple aspects of iron metabolism, thereby attacking a feature of life that is virtually universal.

The ways in which H₂O₂ damages cells have been identified in the model bacterium Escherichia coli. Exogenous H₂O₂ damages DNA, so it is mutagenic. This effect arises from Fenton chemistry, in which H₂O₂ reacts with the intracellular pool of unincorporated iron (42):

The hydroxyl radical (HO·) reacts at diffusion-limited rates with most biomolecules. Some of the iron pool is loosely associated with DNA (92), and Fenton reactions on its surface produce the lesions that cause mutagenesis.

However, the most marked effect of H₂O₂ stress is to inactivate two families of iron-containing enzymes: non-redox mononuclear enzymes and [4Fe-4S]-containing dehydratases (2, 46, 101, 102). Both classes of enzymes employ a solvent-exposed reduced iron atom to bind and activate substrates, and in both cases incoming H₂O₂ can contact and oxidize the iron. The metal dissociates and activity is lost. Such enzymes sit within the TCA cycle, the pentose phosphate pathway, and key biosynthetic pathways, so growth stops. As little as 0.5 μM H₂O₂ is sufficient to poison cells in this way, so the titers of scavenging enzymes are calibrated to keep the steady-state H₂O₂ concentration below this level. This goal becomes much harder when bacteria enter a habitat containing extracellular H₂O₂. The H₂O₂ quickly flows across membranes and into the cytoplasm. For that circumstance virtually all bacteria maintain inducible defensive regulons governed by either the OxyR or the PerR transcription factors.

Defensive tactics: the OxyR regulon

The OxyR protein is more widely distributed. It features a sensory cysteine residue that is directly oxidized by H₂O₂, thereby generating a disulfide bond that alters its binding to DNA (Fig. 1A) (4, 11). In some bacteria OxyR acts as a repressor, and its oxidation inactivates this function and stimulates gene expression (39, 40, 71, 72, 103). More commonly OxyR acts as an activator protein that recruits RNA polymerase. In E. coli the oxidized protein stimulates the transcription of about two dozen genes that are well-suited to address the problems that H₂O₂ causes (Table 1) (118). To drive down the H₂O₂ level, OxyR strongly induces both an NADH peroxidase (AhpCF) and catalase (KatG)(44). A mini-ferritin, Dps, is synthesized to sequester unincorporated iron (10, 31). This has the effect of substantially suppressing the rate of DNA damage, especially in collaboration with the induction of Fur and Yaa proteins (70, 109, 117). The MntH manganese importer is induced, apparently to enable Mn—which does not react with H₂O₂—to displace Fe as the catalytic cofactor of mononuclear enzymes (3, 49, 101). Induction of the Suf iron-sulfur assembly system enables the repair of damaged iron-sulfur clusters. Although the cell normally uses an Isc system for this purpose, Isc does not work well when iron levels decline, a situation that ensues from the action of Dps (47, 83). Similarly, the iron-dependent enzymes in the heme biosynthetic pathway are replaced (HemF) or induced (HemH) in order to sustain this process as intracellular iron becomes scarce (75a).

Three regulon members typically serve to reduce disulfide bonds: glutathione reductase, glutaredoxin 1, and thioredoxin 2. The glutaredoxin assists in deactivating OxyR once H₂O₂ has dissipated (4), but the roles of the others are less obvious. Deletion of these genes does not cause obvious sensitivity to H₂O₂, which inspires other ideas about their purpose in this regulon (see below.)

How does OxyR sense H₂O₂?

Recent studies have analyzed both the specificity and the sensitivity of OxyR. E. coli OxyR features a hyperreactive thiol (Cys199) that is quickly oxidized by H₂O₂ to a sulfenic acid (−SOH). The adducted residue moves from the hydrophobic cleft in which it is normally buried, and it swings into the proximity of Cys208, with which it condenses to form a disulfide bond (11). This shift locks the domain into a conformation that activates the protein as a transcription factor. The protein functions as a dimer, and it displays some cooperativity that may avoid activation when H₂O₂ is scant (60). The fact that OxyR uses cysteine oxidation to detect H₂O₂ enables it to work like a rheostat, given that disulfide formation is among the few amino acid oxidations that are reversible.

The surprising aspect of this chemistry is that cysteine per se reacts very sluggishly with H₂O₂ (2 M⁻¹ s⁻¹) (114). With such a rate constant the half-time for oxidation by 0.5 μM H₂O₂—an intracellular concentration that suffices to block E. coli growth (101)—is approximately one week. In sharp contrast, the sensing cysteine residue on OxyR exhibits a rate constant of 10⁵ M⁻¹ s⁻¹ and detects micromolar H₂O₂ within seconds (4). What is the source of this kinetic improvement?

The hyper-reactive behavior of OxyR C199 has not been deconstructed in detail, but it resembles that of the catalytic cysteine residues on peroxiredoxins, the ubiquitous peroxidases (including E. coli AhpC) that degrade H₂O₂ through cycles of cysteine oxidation and reduction (18, 104). As with OxyR, their reaction involves a nucleophilic attack of a cysteine thiolate upon H₂O₂ (Fig. 1A) with cleavage of the dioxygen bond. A nearby cationic residue sets the stage by ensuring deprotonation of the cysteine (81)—but this should provide only an order-of-magnitude improvement in reactivity. A plausible explanation for the remaining stimulation is depicted in Fig. 1B. The peroxidactic cysteine residue of peroxiredoxins participates in hydrogen-bond networks that shield its charge within a largely hydrophobic cleft (89). Entry of H₂O₂ may shift these bonds towards the H₂O₂ itself, freeing the thiolate. Several aspects of this are conducive to catalysis. Release of the cysteine residue generates a nucleophile whose potency is enhanced in a low-dielectric environment. The hydrogen bonds that shift towards H₂O₂ likely polarize its di-oxygen bond, making it more conducive to attack, and one of these residues probably protonates the hydroxide leaving group to pull the reaction forward. The cysteine residue of OxyR is invariably flanked by arginine and histidine residues that may play the same role.

A hydrogen-bond network is configured to fit H₂O₂ should also enhance the rate constant by disfavoring water in the active site, given that H₂O is too small a molecule to bridge a hyrogen-bond network tailored for H₂O₂ (91). The upshot is that thiol-based peroxidases achieve a level of reactivity (10⁷ M⁻¹ s⁻¹) that is appropriate for their physiological roles, and similar features may have allowed OxyR to do so as well. A new structure of the Pseudomonas aerugionas OxyR protein suggests that H₂O₂ binding might engage a H-bond network that ultimately links the deprotonation of cysteine with the protonation of the product hydroxide group (47a).

In E. coli OxyR quickly responds to the sub-micromolar levels of H₂O₂ that threaten to disrupt metabolism. The facultative bacterium Vibrio vulnificus takes it a step further, employing two OxyR proteins that are calibrated to sense distinct levels of H₂O₂: The more sensitive OxyR is activated by the low levels of H₂O₂ that are endogenously generated upon aeration, whereas the less sensitive one responds to more-severe influxes of H₂O₂ from the external environment (52).

Has OxyR evolved to respond to stresses other than H₂O₂?

Because OxyR responds to adduction of its sensory thiol, it can also be activated by electrophiles (116). Experimenters have commonly used the synthetic chemical diamide as a proxy for such compounds. An interesting question is whether OxyR has an authentic role in sensing stressors other than H₂O₂. Seth et al. reported that OxyR becomes nitrosylated during anaerobic nitrate reduction, presumably by nitric oxide that escapes from the system (97). Indeed, an oxyR mutant exhibited poorer growth in this circumstance, suggesting that some elements of the regulon help the cell to tolerate this stress.

Many bacteria contain transcription factors other than OxyR that appear to be dedicated to responding to electrophiles, whose attack upon thiols can indirectly generate disulfide bonds. Both the Spx system of Bacillus subtilis and the SigR/RsrA system of Streptomyces coelicolor each induce a mixture of glutaredoxins, thioredoxins, chaperones, and proteases, which collectively avert protein aggregation from the denaturing effects of disulfide formation and thiol adduction (48, 51, 84, 94, 119). These bacteria additionally contain OxyR or PerR systems that respond to H₂O₂—but in these bacteria, the latter systems do not include disulfide-reducing genes that are part of the E. coli OxyR regulon. Thus it seems possible that E. coli is exceptional in multitasking OxyR with responding to both stresses: H₂O₂, which primarily threatens iron enzymes, and electrophiles, which react with cysteine residues much more avidly than does H₂O₂.

PerR of Bacillus subtilis: an alternative to OxyR

Gram-positive bacteria often employ a different sensor of H₂O₂: PerR. This dimeric protein belongs to the Fur family of metal-binding transcription factors, which have evolved in their various forms to sense iron (Fur proper), manganese (Mur), zinc (Zur), nickel (Nur), or H₂O₂ (PerR) (19, 38). Each subunit contains a structural site that irreversibly binds zinc, plus at least one regulatory site that reversibly binds the activating metal (Fig. 2A). Only the metal-bound forms of the proteins are competent to bind DNA, and they generally work as repressors. In Bacillus subtilis PerR, the regulatory metal is ferrous iron, and it is affixed in a distorted square pyramid with four co-planar ligands and a single axial ligand (45). When H₂O₂ levels rise, H₂O₂ accesses the sixth coordination site and oxidizes the iron atom. The ferryl or hydroxyl radical that is thereby formed oxidizes the histidine ligands of the iron atom to 2-oxohistidine (63). The modified residues can no longer bind the metal, and the resultant apoprotein undergoes a global conformational change that eradicates its ability to bind DNA (Fig. 2B). Because holo-PerR acts as a repressor, the effect is to derepress regulon members whenever H₂O₂ levels rise.

Strikingly, B. subtilis PerR reacts with H₂O₂ at essentially the same rate constant as does E. coli OxyR (10⁵ M⁻¹ s⁻¹) (62). Other mononuclear iron proteins, such as E. coli peptide deformylase, have similar inactivation rate constants (2), indicating that this value is largely dictated by the fundamental rate of Fenton reactions. However, B. subtilis also employs a true Fur protein whose iron cofactor seems to be resistant to oxidation by H₂O₂. Inspection of this discrepancy elicited evidence that in the Fur protein a coordinating glutamate residue probably acts as a bidentate ligand to the iron atom, thereby occluding its sixth coordination site (87). The analogous aspartate ligand of PerR does not. Given that H₂O₂ must directly contact iron in order to oxidize it (28), H₂O₂ oxidizes PerR but not Fur.

The PerR regulon of Bacillus subtilis includes many of the same genes that OxyR controls in other bacteria: genes encoding a catalase, a Dps homologue (MrgA), and Fur protein (7) (Table 1). Yet PerR offers several intriguing departures from the OxyR mechanism of H₂O₂ sensing. First, given that there is no known mechanism of oxo-histidine reduction, the reaction is presumably irreversible, perhaps making the system more of an on/off switch. This arrangement is a rarity: With few exceptions (78) bacterial regulators are reversible, and their activities are set by the dynamic equilibrium between activating and deactivating events. It is not clear whether the irreversibility of PerR deactivation provides any advantage.

Second, like other mononuclear proteins, PerR can bind manganese in place of iron, both in vitro and in vivo (9). Given that Mn is not oxidized by H₂O₂, the upshot is that PerR does not respond to H₂O₂ whenever the cytoplasm is rich in manganese. As manganese helps to defend cells against H₂O₂, this arrangement suggests that a primary goal of the PerR and OxyR responses may be to raise the Mn/Fe ratio (115); if that ratio is already high, no induction is needed.

Finally, PerR does not control the synthesis of disulfide-reducing redoxins, as their synthesis is governed by the Spx response regulator (119), which is not responsive to H₂O₂. PerR appears to be focused exclusively on the detection and prevention of Fenton chemistry.

The threat of redox-cycling compounds

Because O₂⁻ is a charged molecule, it cannot penetrate into bacteria from the outside (56, 73); therefore, any cytoplasmic O₂⁻ stress is due to O₂⁻ that is generated internally. Endogenous superoxide (O₂⁻) that is formed by enzyme autoxidation is kept very scarce (~10⁻¹⁰ M) by the action of superoxide dismutases (SODs) (41). E. coli mutants that lack cytoplasmic SODs exhibit growth defects (8) that have been traced back to deficiencies in the same enzymes that H₂O₂ inactivates: [4Fe-4S] dehydratases and mononuclear iron enzymes (20, 24, 35, 59, 101). These mutants are unable to synthesize branched-chain or aromatic amino acids, and they cannot catabolize TCA-cycle substrates.

The same phenotypes arise even in wild-type E. coli if it encounters viologens (e.g., paraquat), quinones, or phenazines (37). These cyclic organic chemicals can penetrate into cells, where they directly abstract electrons from redox enzymes and transfer them to oxygen. Toxic doses of O₂⁻ are thereby produced. Indeed, quinones and phenazines are released by various bacteria and plants, often in an effort to poison microflora (43, 85, 107). In 1977 it was discovered that E. coli responds to this stress by inducing higher titers of SOD (36). This adaptation is mediated by two proteins dubbed SoxR and SoxS (superoxide) (32, 106). SoxR is a [2Fe-2S]-containing transcription factor that senses the stress; it then induces the transcription of soxS. SoxS is a second transcription factor that then activates scores of defensive genes scattered around the chromosome (6). Several of these encode proteins that help suppress the toxicity of O₂⁻: superoxide dismutase; aconitase A and fumarase C, which are superoxide-resistant isozymes of vulnerable [4Fe-4S] enzymes (67, 108); and YggX, which facilitates cluster-repair processes (30, 34, 88). Like OxyR, the SoxRS system is not activated during normal aerobic growth and does not control the basal level of defensive systems; instead, it becomes active only when forcing conditions emerge.

The sensing mechanism of SoxR

SoxR is a homodimer that binds one [2Fe-2S]⁺ cluster per monomer (Fig. 3); a second domain includes a DNA-binding HTH motif (112). Electron paramagnetic resonance studies demonstrate that when cells are exposed to redox-cycling compounds, the clusters are quickly oxidized to the [2Fe-2S]²⁺ state (15, 26). Upon removal of the oxidant the clusters revert within minutes to their reduced forms. The E. coli soxS gene features an unusually long 20-bp spacing between its −10 and −35 promoter motifs, and the current model is that cluster oxidation contorts the SoxR dimer and thereby alters the conformation of the bound DNA, moving the promoter motifs to the same side of the helix so that RNA polymerase can productively bind. Physical studies have confirmed that changes in SoxR redox state distort bound DNA (23). Genetic studies indicate that the SoxR reduction to the deactivated form is catalyzed by Rsx/Rse proteins (55) that likely employ NADPH as a source of electrons (54). Thus these data collectively suggest that transcription of the soxS depends upon the dynamic equilibrium between the oxidation and reduction of the SoxR iron-sulfur clusters. When SoxR returns to its reduced form, the transcription of the full regulon quickly ceases owning to the rapid proteolysis of extant SoxS protein (33).

What is the inducer that oxidizes SoxR?

The apparent parallel with H₂O₂ detection by OxyR suggested the obvious possibility that SoxR is activated when O₂⁻ directly oxidizes its clusters. The clusters are located near the protein surface and appear accessible to a small molecule like O₂⁻ (Fig. 3) (112). Unfortunately, two efforts to directly detect cluster oxidation by O₂⁻ in vitro led to contradictory results (21, 34), suggesting that one of these complicated experiments was compromised in some way. The in vivo data, however, are clearer and somewhat surprising. In SOD⁻ mutants—which suffer debilitating levels of O₂⁻—the soxS gene is induced only 2- to 4-fold (25, 34). This activation is consistent with the idea that O₂⁻ can oxidize the clusters, but it is far less than the ~40-fold induction that occurs when wild-type cells are treated with redox-cycling compounds. Similarly modest effects were observed for members of the regulon (34, 66, 68). This induction is too minor to be effectual: The SoxRS regulon provided no protection to SOD⁻ mutants unless soxS was forcibly induced from a heterologous promoter (34). Thus high O₂⁻ concentrations are not sufficient to adequately activate SoxR.

Further, overproduction of SOD does not diminish the ability of redox agents to activate SoxR in wild-type cells (25, 29, 34, 67, 75), leading workers to conclude that these drugs oxidize SoxR independently of O₂⁻. Indeed, exposure to redox compounds can activate SoxR even in anoxic cells, from which O₂⁻ is absolutely absent (14, 34, 57).

An alternative explanation for SoxR oxidation by redox agents is that they deplete NADPH pools and thereby disrupt the ability of Rse/Rsx to keep SoxR in its reduced form. Indeed, some genes within the E. coli SoxR regulon provide mechanisms to restore NADPH pools (58, 79). However, measurements indicate that the shift in NADPH redox status [to −.314 V, calculated from (57)] is not enough to explain the near-total oxidation of SoxR [Eo’ = −.293 V (53)], and mutations that impair cellular NADPH formation do not activate SoxR (34). Thus although NADPH depletion might augment the inducing effects of redox agents, by itself it is probably not sufficient to turn on the system.

Therefore it seems likely that the redox compounds themselves directly oxidize the clusters of SoxR, just as they do the metal and flavin moieties of typical redox enzymes. This reaction has been demonstrated in vitro, with rate constants that accommodate the rates of SoxR oxidation that are observed in vivo (34). The position of the iron atom near the SoxR protein surface ensures that it can be oxidized by diverse univalent oxidants without regard for their structures.

The constituency of the regulon supports the idea that the SoxRS response specifically arose to defend bacteria against redox-cycling compounds. A major element of the SoxRS response in E. coli is the induction of proteins that collectively block the entry and accumulation of redox-active compounds within the cell: lipopolysaccharide-modifying enzymes that diminish the ability of such compounds to penetrate the envelope, broad-specificity export systems that actively pump them out of the cytoplasm, and derivatizing enzymes that deactivate them through covalent modification (1, 61, 65, 69, 74, 93) (Fig. 4A). Interestingly, SoxR is an important defense against redox-active agents even under anoxic conditions, indicating that these agents can exert O₂⁻-independent mechanisms of toxicity. Their primary action under these conditions is unclear, but such compounds can form adducts to proteins by acting as Michael acceptors, they can deactivate [4Fe-4S] dehydratases by directly oxidizing the clusters, and they can deplete cellular NADPH pools by oxidizing NADPH-reducible enzymes and then dumping the electrons into the anaerobic respiratory chain (34).

Finally, it is worth noting that although H₂O₂ is an ineffective activator of SoxR (34, 118), nitric oxide (NO) can activate it by forming dinitrosyl iron complexes with the atoms of the iron-sulfur cluster (22, 82). The significance of this effect is uncertain, because dedicated NO scavenging systems are controlled by independent regulatory proteins (NsrR and NorR) that are much more responsive to NO. In one study (90) NO doses that activated the aerobic and anaerobic NO-scavenging systems by 30- and 180-fold only induced soxS 5-fold. Furthermore, this slight SoxS induction had no apparent effect upon SoxS-controlled genes, leading the authors to suggest that this degree of SoxS induction was inadequate for activation of the full regulon. That outcome dovetails with the failure of 3-fold SoxS induction to provide any protection to SOD⁻ cells (34).

A more expansive view: SoxR beyond E. coli

Studies of SoxR and SoxS in other bacteria have prompted a substantial reappraisal of their biological role. Dietrich et al. found homologs of SoxR among α-, β-, γ-, and δ-proteobacteria and among actinobacteria, but not elsewhere (13). Strikingly, only enterics employ SoxS; in all the other bacteria SoxR directly controls the expression of all regulon members. Just as striking is the fact that in these non-enterics the SoxR regulon is quite different from that of E. coli: it typically comprises only a handful of genes of unclear function (80), and these generally do not include sodA or other genes that might confer resistance to O₂⁻ (13, 14, 50, 80, 99). The implication is that the physiological purpose of SoxR in non-enterics is quite different from its purpose in E. coli.

Evidence regarding this purpose has come from studies of Pseudomonas aeruginosa and Streptomyces coelicolor. Both bacteria synthesize and excrete redox-cycling compounds: pyocyanin for P. aeruginosa and actinorhodin for S. coelicolor. Different roles have been suggested for these compounds, including toxifying competitors, signaling, assisting electron transfer to distant respiratory substrates, and solubilizing iron (14, 110, 111). The pumps that excrete these compounds, and perhaps the enzymes that complete their synthesis, are induced in both bacteria when SoxR is oxidized by either the redox agent itself or a precursor to it (13, 99) (Fig. 4B). The SoxR protein can also respond to exogenous synthetic redox compounds such as paraquat, but this is presumably adventitious, as SoxR activation does not enhance resistance to such compounds (86). The implication is that these bacteria use SoxR to control the export of endogenous redox agents into the environment.

Whether or not it is to the advantage of the producing organism, these agents can penetrate and toxify other microbes. It is possible that the latter laterally acquired SoxR as a sensor that could perceive these substances and induce defenses. In E. coli the arriving soxR gene may have piggy-backed onto a pre-existing system that responds to other toxic chemicals (Fig. 4B). SoxS is a paralog of the MarA and Rob transcription factors, which respond to and provide protection against exogenous non-redox agents (to salicylate and bile salts, for example). The three proteins are similar enough (~ 50% identity) that they recognize common regulatory sites, and so their regulons overlap extensively (76). The simple model, then, is that E. coli acquired SoxR and, after duplication of MarA or Rob, fashioned SoxS to connect SoxR to a pre-existing regulon that serves to exclude, modify, or export hazardous compounds. SoxR provided the means to activate this system in response to redox-active compounds. Notably, the reduction potential of E. coli SoxR is lower than those of P. aeruginosa and S. coelicolor SoxR proteins, and accordingly it responds to a larger range of redox agents to which it responds (98, 100). Extant antioxidant defenses, such as sodA and fumC, may then have been added to the regulon. The latter genes are significantly more responsive to SoxS than to MarA (76).