When Escherichia coli is exposed to redox-cycling drugs, its SoxR transcription factor is activated by oxidation of its [2Fe–2S] cluster. In aerobic cells these drugs generate superoxide, and because superoxide dismutase (SOD) is a member of the SoxRS regulon, superoxide was initially thought to be the activator of SoxR. Its many-gene regulon was therefore believed to comprise a defence against superoxide stress. However, we found that abundant superoxide did not effectively activate SoxR in an SOD- mutant, that overproduced SOD could not suppress activation by redox-cycling drugs, and that redox-cycling drugs were able to activate SoxR in anaerobic cells as long as alternative respiratory acceptors were provided. Thus superoxide is not the signal that SoxR senses. Indeed, redox-cycling drugs directly oxidized the cluster of purified SoxR in vitro, while superoxide did not. Redox-cycling drugs are excreted by both bacteria and plants. Their toxicity does not require superoxide, as they poisoned E. coli under anaerobic conditions, in part by oxidizing dehydratase iron–sulfur clusters. Under these conditions SoxRS induction was protective. Thus it is physiologically appropriate that the SoxR protein directly senses redox-cycling drugs rather than superoxide.
SoxR is a dimeric transcription factor, with each monomer containing a [2Fe–2S] cluster (Hidalgo et al., 1995). When aerobic E. coli is exposed to redox-cycling drugs such as paraquat or menadione, the cluster undergoes a reversible one-electron oxidation and gains the ability to activate the transcription of soxS, the gene immediately adjacent (Ding et al., 1996; Gaudu and Weiss, 1996; Ding and Demple, 1997; Gaudu et al., 1997). The resulting SoxS protein is itself a transcription factor that activates the expression of more than 100 genes in the SoxRS regulon (Pomposiello et al., 2001). When oxidative stress abates, the oxidized SoxR is returned to its reduced state via reducing systems encoded by rseC and rsxABCDGE (Koo et al., 2003). Proteolysis rapidly degrades the extant SoxS protein, ending the response (Griffith et al., 2004). SoxR is found in many bacteria, although in non-enterics SoxS is absent and SoxR directly binds the promoter region of each of the regulon members (Dietrich et al., 2008).
What environmental circumstances might create superoxide stress inside bacteria? Superoxide is continuously generated by the adventitious oxidation of redox enzymes (Imlay, 2008), but studies with E. coli indicate that the basal amount of SOD is enough to keep its steady-state concentration at non-toxic levels (Gort and Imlay, 1998). Superoxide does not cross membranes, so exogenous superoxide cannot penetrate cells (Korshunov and Imlay, 2002). To date, then, the only situation that elevates intracellular superoxide to levels that warrant SOD induction are the conditions that were originally explored by Hassan and Fridovich – the presence of redox-cycling drugs (Hassan and Fridovich, 1978) (The structures of three classes of redox-cycling drugs are shown in Fig. 1). It is now recognized that these compounds are released by both plants and bacteria as devices to inhibit the growth of competitors (Paiva et al., 2003; Inbaraj and Chignell, 2004). For example, plumbagin, a naphthoquinone, was originally isolated from the plant Plumbago (van der Vijver, 1972); juglone, another quinone, occurs naturally in the Juglandaceae family and is recognizable as the yellow residue on the leaves and seeds of the black walnut (Inbaraj and Chignell, 2004). Both compounds are effective herbicides that allow the parent plant to dominate a habitat. Phenazines are commonly excreted by bacteria, including Pseudomonas, Streptomyces and Pantoea agglomerans (Turner and Messenger, 1986). They exert toxic effects on other bacteria. In addition, man-made viologens such as paraquat (PQ, methyl viologen) are also used as herbicides. Each of these drugs can penetrate into the cell interior, where they abstract single electrons from the reduced flavins or metal centres of redox enzymes. The reduced drug can then transfer the electron to oxygen, generating superoxide (Hassan and Fridovich, 1979). This redox-cycling behaviour can elevate intracellular superoxide formation by orders of magnitude above the usual rate.
Thus the induction of SOD by the SoxRS regulon provides a critical defence against these drugs. Other components of the regulon focus on limiting the intracellular levels of these drugs. For example, the acrAB encodes a drug efflux system (Ma et al., 1995). The micF gene encodes an antisense RNA that represses synthesis of the OmpF outer membrane porin (Aiba et al., 1987), while waaYZ encodes an LPS modification function (Lee et al., 2009); together, the induction of these genes apparently reduces the permeability of the cell envelope so that redox drugs cannot easily enter. Other induced proteins – NfsA, YgfZ and NfnB – detoxify redox drugs by modifying them (Liochev et al., 1999; Rau and Stolz, 2003; C. N. Lin, pers. comm.). The presence of these genes in the SoxRS regulon confirms that this system exists to defend the cell from such redox-cycling drugs.
However, several observations have arisen that do not fit the notion that SoxR responds to superoxide per se. First, workers found that paraquat could induce sodA in E. coli even in anaerobic habitats, if nitrate was supplied (Privalle and Fridovich, 1988). Under those conditions superoxide could not be present. Further, experiments showed that SoxRS is poorly induced in SOD- mutants of E. coli, despite the accumulation of sufficient superoxide to disable several key metabolic pathways. The same study showed that SoxRS activation by paraquat could not be suppressed by SOD overproduction; together these results indicated that superoxide was neither sufficient nor necessary for SoxRS activation (Gort and Imlay, 1998). Both groups suggested alternative mechanisms, including the possibilities that redox drugs might either directly oxidize SoxR or that they might deplete NADPH and thereby slow the SoxR reducing system. More recently, studies in Pseudomonas aeruginosa replicated the observation of anaerobic SoxR activation by paraquat/nitrate, although the authors pointed out the caveat that nitric oxide, which can be formed as a by-product of nitrite reduction, might activate SoxR by degrading its iron–sulfur cluster (Dietrich et al., 2006). Finally, the SoxR regulons of both P. aeruginosa and Pseudomonas putida do not seem to involve enzymes, such as SOD, that explicitly defend cells against superoxide (Kobayashi and Tagawa, 2004; Palma et al., 2005; Dietrich et al., 2006).
These results reopen the question: What is the physiological activator of SoxRS? As a corollary, to what extent might these drugs threaten cells through mechanisms that do not involve superoxide?
Superoxide is not the physiological activator of SoxR
The growth defects of E. coli catalase/peroxidase mutants, which cannot scavenge endogenous H2O2, are substantially suppressed by the concurrent induction of the OxyR stress response. For example, while the catalase/peroxidase mutants grow steadily in aerobic media, the addition of an oxyR mutation eradicates growth (Park et al., 2005). A similar effect is created by the addition of mutations in individual genes that OxyR controls, including dps, suf, fur and mntH (Park et al., 2005; Jang and Imlay, 2007; Varghese et al., 2007; Anjem et al., 2009). We wondered whether, in an analogous way, the full toxic effect of endogenous superoxide in SOD- mutants (sodA sodB) is muted by the induction of the SoxRS regulon. Therefore, a soxS null mutation was introduced into a SOD- mutant, and the growth rates of the two strains were compared under several aerobic conditions. No differences were found (Fig. 2A). Similar results were obtained when several individual SoxS-controlled genes were deleted from the SOD- strain, including zwf and fpr (data not shown).
This result could be explained in either of two, non-exclusive ways: SoxRS might not be significantly active in SOD- mutants, and/or the genes it controls might not protect the cell from O2-. We examined both possibilities in turn.
The degree of SoxRS activation was determined with soxS::lacZ fusions. Using different growth conditions, our lab previously reported that soxS is not significantly activated in SOD- mutants (Gort and Imlay, 1998). Using the growth conditions of the preceding experiment, we reproduced those results: the soxS gene was only marginally induced (∼3-fold over the basal level) in SOD- mutants, and the degree of expression was far higher (more than 40-fold over the basal level) when paraquat was added (Fig. 2B). Therefore, although this level of O2- is sufficient to block key biosynthetic and catabolic pathways, it is a poor inducer of SoxRS compared with paraquat.
These results suggested that in paraquat-treated cells the SoxR protein senses a signal other than O2-. To test this idea more directly, we examined whether a 20-fold overproduction of SOD could diminish the SoxRS response to three classes of redox-cycling drugs: viologens (e.g. paraquat), quinones (menadione) and phenazines (phenazine methosulfate) (The structures are shown in Fig. 1). Strains lacking sodA were used in order to avoid the complications of sodA induction via SoxRS. If O2- were the signal, one might expect the soxS::lacZ dose–response curves to be shifted in the overproducing strain so that a 20-fold higher level of drug is required for equivalent induction. In fact, the curves were unchanged (Fig. 2C). We conclude that O2- is not the inducer of SoxRS.
Most SoxRS-controlled genes do not mitigate superoxide stress
Our second question was whether SoxRS would confer protection against O2- in the SOD- mutant if the regulon were strongly induced by genetic means. The soxS gene was overexpressed directly from the tac promoter using psoxS, and enzyme assays confirmed that this construct overexpressed zwf as strongly as when 100 µM paraquat was added, indicating that the SoxRS regulon was fully induced. We observed that, like its SOD- parent (PN134), the SoxRS-induced strain remained unable to grow in glucose medium unless aromatic, sulfur-containing and branched-chain amino acids were supplemented (data not shown). Even in amino-acid supplemented medium, SoxRS induction did not noticeably improve the growth rate of the SOD- mutant.
The primary known toxic action of O2- in E. coli is to damage the [4Fe–4S] clusters of dehydratases (Kuo et al., 1987; Gardner and Fridovich, 1991a,b; Flint et al., 1993). We measured the impact of soxS overexpression on the activity of one of these enzymes, 6-phosphogluconate dehydratase (Edd). In fact, soxS overexpression significantly improved Edd activity in SOD- mutants (Fig. 3), from 25% of wild-type activity to about 60%. Although SoxS affects the expression of ca. 100 genes (Pomposiello et al., 2001), this effect could be completely ascribed to the induction of yggX, a SoxRS-responsive 11 kD cytoplasmic protein. YggX has been proposed to be involved in Fe–S cluster metabolism during oxidative stress (Gralnick and Downs, 2001; Pomposiello et al., 2003). The activity improvement that was conferred by soxS overexpression could be quantitatively replicated by overexpression of yggX itself, and soxS was no longer protective in mutants that lack yggX. Dynamic measurements showed that YggX improved Edd activity by accelerating Fe–S cluster repair (Fig. S1).
There appear to be more than 100 genes in the SoxRS regulon, many of which have unknown functions (Blanchard et al., 2007). The gene sodA, encoding the Mn-containing SOD, defends against O2-; acnA and fumC encode oxidant-resistant isozymes that replace damaged Fe–S dehydratases; and yggX facilitates Fe–S cluster repair under stress. The preceding data suggest that the other genes probably do little to ameliorate O2- toxicity per se. As mentioned above, many other known members of the SoxRS regulon seem to be focused upon reducing the intracellular levels of redox-cycling drugs.
Cellular respiration reoxidizes redox-cycling drugs and enables them to activate SoxR
Redox-cycling drugs were unable to induce the manganese-containing SOD under anaerobic conditions (Moody and Hassan, 1984), which fit the notion that O2- was the activator of SoxRS. An alternative explanation, however, might be that only the oxidized drugs generate the SoxR signal; when oxygen is unavailable to reoxidize them, the drugs quickly accumulate in a reduced, inactive form. Indeed, when paraquat is added to anaerobic cultures, both the media and the cells turn the blue colour of the reduced paraquat radical species; the colour immediately dissipates when air is introduced.
We examined expression of soxS itself under these anaerobic conditions. Unlike sodA, the soxS gene could still be induced, although the induction level was much smaller than when oxygen was present (Fig. 4A).
Other workers reported that paraquat can induce sodA in the absence of oxygen if nitrate is present (Privalle and Fridovich, 1988; 1991; Privalle et al., 1989). An ambiguity attending these experiments was the fact that nitric oxide, which is formed in trace amounts during denitrification, can itself activate SoxR (Ding and Demple, 2000). We found that soxS expression was also induced strongly when nitrate was supplied to the cell (Fig. 4A). Further, fumarate, another respiratory substrate, also improved induction (see the insert of Fig. 4B). Fumarate, nitrate and oxygen are progressively better respiratory substrates for E. coli, and so their parallel effects on soxS expression suggested that paraquat might be recycled in vivo via oxidation by the electron transport chain. Indeed, the inducing effect of paraquat/fumarate was lost when we eliminated the capacity for fumarate-directed respiration, by deleting the operon encoding fumarate reductase. Deletion of the nar genes, which encode the primary nitrate reductase, substantially diminished the inducing effect of nitrate/paraquat, but the treatment could still promote soxS expression to some extent (Fig. 4A). The residual effect may have been due to the remaining nitrate reductase isozymes or, possibly, to the formation of nitric oxide. Fumarate was unable to stimulate anaerobic soxS transcription in mutants that lacked respiratory quinones (menA ubiA), but it did so in ndh nuo mutants that lack the NADH dehydrogenases through which most of reducing equivalents enter the chain (Fig. 4B). These results suggest that drugs accumulate in their reduced (non-inducing) forms unless they can transfer their electrons to an oxidized pool of respiratory quinones.
The ability of the respiratory chain to directly oxidize reduced redox drugs was confirmed in vitro. Inverted respiratory vesicles containing fumarate reductase were prepared from cells grown in anaerobic media with fumarate. Paraquat radical was generated by adding dithionite to paraquat in anaerobic buffer. The paraquat-radical spectrum was quickly bleached upon the addition of both inverted respiratory vesicles and fumarate (Fig. S2).
To verify that the role of oxygen in SoxR activation is simply to recycle reduced drugs, we sought to use an alternative oxidant, ferricyanide, in its place. Ferricyanide is a small chemical oxidant that is known to be able to oxidize the quinone pool. In the absence of respiratory substrates, the enhancing effect of aerobic respiration upon soxS expression was replicated by the provision of ferricyanide (Fig. 5). In fact, when ferricyanide was used at high concentrations, the anaerobic induction level of soxS was even higher than when oxygen was present.
To verify that anaerobic soxS expression reflects the oxidation of SoxR, this protein was overexpressed in vivo, and EPR spectroscopy was performed to monitor the redox state of its [2Fe–2S] cluster during drug exposure (Ding and Demple, 1997) (Gaudu et al., 1997). In the absence of redox-cycling drugs, the cluster exhibited an EPR spectrum that is typical of reduced [2Fe–2S]+ clusters. When paraquat was added to anaerobic cells, we observed a gradual disappearance as the paraquat concentration increased (Fig. 6A), indicating the oxidation of SoxR. Menadione and PMS had the same effect (data not shown).
The above observations collectively demonstrated that O2- is not the signal that SoxR senses. The key role of oxygen is to regenerate the oxidized drugs, which, directly or indirectly, trigger SoxR oxidation.
Redox-cycling drugs directly oxidize SoxR
The redox state of SoxR is determined by the kinetic equilibrium between its reduction and its oxidation; therefore, redox-cycling drugs must either slow the first process or accelerate the second. NAD(P)H is the probable electron source for SoxR reduction (Koo et al., 2003), and one possibility is that redox-cycling drugs might lower NADPH levels and thereby lead to SoxR activation. To test whether NADPH depletion can have this effect, we examined a zwf pnt mutant, which lacks the two primary mechanisms by which E. coli generates NADPH. In minimal glucose medium this mutant grew with a 4 h doubling time due to its extreme deficiency in NADPH. In the absence of redox drugs, the strain showed a minimally elevated expression of soxS, which is consistent with results reported by Krapp and colleagues (Krapp et al., 2010). However, it still responded strongly to PQ (Fig. S3A), suggesting that NADPH depletion itself is a weak trigger of SoxRS induction. The rsx and rseC gene products constitute the primary reducing systems for SoxR in E. coli (Koo et al., 2003). When we eliminated either or both of these, we again observed slightly higher basal levels of soxS induction, but once more the mutants showed typical SoxRS induction patterns in response to PQ (Fig. S3B). Therefore, the activating effect of the drugs is not due to an inhibition of these SoxR-reducing systems.
The obvious alternative is that redox-cycling drugs directly oxidize SoxR. To examine this possibility, we purified SoxR protein from the soxR-overproducing strain XA90/pKOXR. SoxR was eluted in an oxidized form, which showed a spectrum typical of [2Fe–2S] with absorption maxima at 414 and 450 nm (Fig. S4). It could be reduced under anaerobic conditions by dithionite, generating an EPR signal that is characteristic of reduced [2Fe–2S]+ clusters. When we treated the reduced SoxR protein with redox drugs in the absence of oxygen, this EPR signal rapidly disappeared (Fig. 6B), corresponding to the oxidation of the SoxR cluster to the [2Fe–2S]2+ form.
These data were consistent with the fact that small-molecule mediators had been used to determine the SoxR reduction potential (Ding et al., 1996); however, the rates of SoxR oxidation by potential oxidants have not been reported. The oxidized cluster absorbs light at 414 nm, and so we used visible spectroscopy to measure the rates at which redox drugs react with SoxR in the absence of oxygen. The rate constants are listed in Table 1. To use these values in calculating the rate at which SoxR might be oxidized by a drug in vivo, it is necessary to know the intracellular concentration of the drug. E. coli cells that had been cultured with paraquat were harvested, washed and lysed, and the released paraquat was reduced with dithionite and quantified by spectroscopy. We found, for example, that incubation with 200 µM paraquat – an inducing dose – led cells to accumulate at least 700 µM intracellular paraquat. This value is likely to be an underestimate, as some paraquat was lost during the washing process, but it is sufficiently high to predict that intracellular paraquat can oxidize SoxR within the time frame of the biological response (see Discussion).
Table 1. Rate constants for the oxidation of purified SoxR and fumarase A by univalent oxidants.
Oxidation rate constants were also measured when reduced SoxR was exposed to molecular oxygen and to hydrogen peroxide. These values are very low (Table 1), and they explain why neither aeration nor physiological (low-micromolar) H2O2 stress activates SoxR in vivo (Zheng et al., 2001). Finally, the question of whether O2- can directly oxidize SoxR (Eq. 1) was examined in two ways:
First, exposure of reduced SoxR to xanthine oxidase, a potent source of superoxide, did not lead to any apparent loss of [2Fe–2S]+ absorbance over 5 min. An equivalent exposure inactivated fumarase A by > 90% within 20 s. Second, the presence of reduced SoxR did not detectably diminish the rate at which xanthine-oxidase-generated O2- transferred electrons to cytochrome c (Fig. S5). Given the rate constant with which O2- reacts with cytochrome c, we deduce that the rate constant with which SoxR degrades O2- must be lower than 1000 M−1 s−1 (Experimental procedures). This value is far too low for SoxR to respond to physiological levels of O2- (see Discussion).
Redox-cycling drugs exert toxic effects in the absence of oxygen
Anaerobic paraquat/nitrate treatment induced a sodA::lacZ fusion to < 1% of its paraquat/oxygen level. This lack of SOD induction makes physiological sense, because O2- scavengers are not needed in anaerobic environments. The anaerobic regulatory protein Fnr, which is a repressor of sodA, apparently overrides the activating effects of oxidized SoxR transcription (Tardat and Touati, 1991; Hassan and Sun, 1992; Compan and Touati, 1993). In contrast, paraquat/nitrate treatment induced the soxS::lacZ fusion to about 80% of its paraquat/oxygen level. In fact, we found that, unlike sodA, another SoxR-controlled protein, fumarase C, was induced by paraquat/nitrate to at least half of its paraquat/oxygen level: induction was 3.3-fold under the anaerobic conditions, compared with 5.5-fold in aerobic medium. If we assume that this response occurs because it is useful, we infer that redox drugs must have oxygen-independent toxic effects that SoxR has evolved to suppress.
Consistent with this notion, redox drugs slowed bacterial growth in anaerobic nitrate media, although higher drug doses were needed than in aerobic media (Fig. S6). 120 µM PMS almost completely blocked the growth of the wild type strain under anaerobic conditions; upon exposure to 40 µM PMS, cells lagged and then grew, suggesting that adaptation occurred. A lower dose created a branched-chain amino acid auxotrophy; wild-type cells eventually adapted, while a ΔsoxR mutant did not (Fig. 7). Branched-chain auxotrophy results from the oxidation of the two [4Fe–4S] cluster dehydratases in that pathway, dihydroxyacid dehydratase and isopropylmalate isomerase (Kuo et al., 1987; Flint et al., 1993; Jang and Imlay, 2007). Therefore, the implication was that PMS can damage vulnerable dehydratases even in the absence of oxygen, and that cells respond by inducing SoxRS.
The obvious hypothesis was that redox drugs might directly oxidize the clusters of dehydratases, as they do that of SoxR. Indeed, assays confirmed that paraquat, menadione and PMS treatments each inactivated fumarase B inside anaerobic cells (Fig. 8A). This observation explains why Fnr does not override the induction of fumC, which encodes a cluster-free fumarase isozyme. These drugs were also able to directly inactivate purified fumarase A in vitro (Fig. 8B). Fumarase was protected from drugs by its substrate, L-malate (Fig. S7A), indicating that the drugs damaged the enzyme by interacting with its active site; activity was fully restored when the inactivated enzyme was subsequently incubated with ferrous iron, dithiothreitol, IscS and cysteine, a cocktail that rebuilds damaged iron–sulfur clusters (Fig. S7B). The anaerobic inactivation constants were measured in vitro to be 102–104 M−1 s−1 at 0°C.
The experiments reported here indicate that the SoxR response comprises a general defence against redox-cycling drugs. The apparent mechanism of SoxR activation is depicted in Fig. 9. Redox-cycling drugs directly oxidize SoxR; when respiratory substrates are present, the reduced drugs then transfer electrons to the quinones of the respiratory chain. The latter reaction recycles the drugs, enabling them to continue to oxidize both SoxR and other target enzymes in the cell.
SoxR is configured to be a non-specific sensor of univalent oxidants
Phenazines, quinones and viologens are efficient oxidants of exposed flavins and metal centres, as their one-electron reduction does not require the scission or creation of new bonds and therefore does not require facilitation by particular enzymic functional groups. Thus the ability of these redox-cycling drugs to oxidize redox enzymes is determined largely by whether they can get close to their cofactors. The crystal structure of SoxR shows that its [2Fe–2S] sensory cluster is exposed at the surface of the protein (Watanabe et al., 2008), which means that redox drugs of a wide variety of shapes and sizes can approach and oxidize it; this protein is therefore ideal as non-selective sensor of univalent oxidants. In contrast, SODs have evolved so their redox metals are buried at the bottom of a narrow cleft that excludes molecules larger than diatomic superoxide (Tainer et al., 1982), and in fact they do not react with other molecules. This comparison suggests that the structure of SoxR evolved to sense a variety of oxidants, rather than superoxide.
The oxidizability of SoxR is assisted by the low potential of its [2Fe–2S] cluster (−285 mV), which provides a strong driving force for electron transfer to even moderate oxidants (Ding et al., 1996). The SoxR cluster also has a key feature that the dehydratase [4Fe–4S] clusters lack: when it is oxidized it does not degrade. Instead the oxidized [2Fe–2S]2+ cluster remains intact and is quickly reduced again when oxidants are removed, presumably allowing the regulon to respond quickly to changing circumstances (Ding and Demple, 1997). In contrast, the degraded clusters from oxidized [4Fe–4S] dehydratases are rebuilt slowly, particularly in iron-poor environments. In this respect SoxR provides an interesting contrast with the E. coli Fnr transcription factor, whose [4Fe–4S] cluster is destroyed as the crux of its oxygen-sensing mechanism (Lazazzera et al., 1996). Perhaps SoxR is calibrated to sense the degree of redox stress, whereas Fnr provides a black-or-white report on oxygenation.
Can SoxR sense superoxide?
The fact that SoxR reacts directly with redox drugs does not, of course, mean that it cannot also react with superoxide. However, in our experiments the rate of its reaction with superoxide fell below the detection limit, indicating that the rate constant was less than 1000 M−1 s−1. Prior work suggested that in aerobic E. coli the cytoplasmic superoxide level is about 0.1 nM, and a 10-fold increase to 1 nM creates enough stress that metabolic pathways begin to fail (Gort and Imlay, 1998). At the latter concentration – even employing 1000 M−1 s−1, the upper limit for SoxR reactivity – one calculates that the half-time for SoxR oxidation would be 8 days. Thus SoxR is not capable of sensing the amount of superoxide that comprises a threat to the organism.
One can also calculate that paraquat, not the superoxide it generates, is the more important oxidant of SoxR in paraquat-stressed cells. Our measurements show that when E. coli is treated with 200 µM paraquat the intracellular rate of non-respiratory oxygen consumption (a.k.a., cyanide-resistant respiration) reaches a maximum of ca. 1.1 mM s−1; this value indicates the formation of 2.2 mM s−1 superoxide, a very high rate. Yet, given that even uninduced cells contain 3600 U ml−1 SOD, one calculates that the steady-state level of superoxide in these paraquat-challenged cells is only 80 nM. This concentration is high enough to oxidize dehydratases (k = 106–107 M−1 s−1) with a half-time of 10 s, but the half-time for the oxidation of SoxR by superoxide (k < 1000 M−1 s−1) must exceed 2 h. In contrast, SoxR oxidation by the 700 µM intracellular paraquat that accumulates in this circumstance (k = 30 M−1 s−1) should exhibit a half-time of only 30 s. Thus even in extremely superoxide-stressed cells, superoxide levels are inadequate to activate SoxR – but the drug itself is a sufficient inducer.
One might wonder, then, why SOD mutants exhibit slight SoxRS activation. In these cells the superoxide level may rise far higher than in SOD-proficient cells, even when the latter are treated with redox drugs. A level of up to 5 µM is theoretically possible (Imlay and Fridovich, 1991), which would oxidize SoxR with a half-time of about 2 min – conceivably fast enough to partially override the reducing activity of the Rse/Rsx systems. However, as the preceding calculations show, these levels of superoxide cannot be achieved in wild-type cells that contain SOD, even with toxic levels of redox drugs. It is also possible that the metabolic dysfunction of SOD- mutants diminishes the efficiency of the Rse/Rsx systems.
The sluggish reactivity of the SoxR cluster with superoxide contrasts with the high reactivities of the clusters of dehydratases, which are oxidized by superoxide at rates that are at least 1000-fold greater (106–107 M−1 s−1; Flint et al., 1993). Similarly, the rate constant for the oxidation of SoxR by H2O2 (0.8 M−1 s−1) is far lower than those of dehydratases (103–104 M−1 s−1) (Jang and Imlay, 2007). Why the disparity? The key difference between these enzymes is that the catalytic iron atom of a dehydratase cluster has an open co-ordination site (Lauble et al., 1992) (Martins et al., 2004). This arrangement allows that iron atom to bind the oxygen atoms of its substrate – but, by the same token, the same iron atom can directly bind superoxide or H2O2. Direct binding is requisite for H2O2 to react with iron, as its dioxygen bond can only be cleaved by an inner-sphere electron tranfer. Direct binding also facilitates the oxidation of clusters by superoxide, probably because the formation of an iron–superoxide complex allows time for protonation, which must occur before electron transfer. In any case, when the clusters of dehydratases are occupied by substrates or even small ligands, oxidation by superoxide or H2O2 is almost completely inhibited (Gardner and Fridovich, 1991b; Jang and Imlay, 2007).
In contrast to the catalytic iron atom of dehydratases, the two iron atoms of the SoxR cluster are each fully co-ordinated by four cysteinyl and bridging sulfur atoms (Bradley et al., 1997; Watanabe et al., 2008); thus, direct binding by oxygen species will be hindered. It follows that reactions with superoxide and H2O2 should be slow. Conversely, dehydratases and SoxR differ by only about 10-fold in the rates at which they react with molecular oxygen and redox drugs; these are oxidants that can react by outer-sphere mechanisms and so do not need to directly bind clusters in order to oxidize them.
The threat of redox-cycling drugs
The prevalence of redox-cycling drugs in real-world habitats is not clear; to date workers have studied their excretion by only a handful of plants and bacteria. The toxicity of these compounds seems to suggest that organisms excrete these drugs in order to suppress the growth of their competitors. That certainly appears to be the case with plants, such as walnut trees, which denude the ground below them through quinones that are extruded from their dropped leaves. The effect is that walnut seeds can find bare soil. A similar strategy has been attributed to bacteria that secrete such compounds. However, Wang, Newman and co-workers have recently pointed out that the release of phenazines by biofilm-forming bacteria might have the effect of permitting electron delivery to terminal oxidants, such as iron precipitates, which lie at a distance from the bacterium (Wang et al., 2010). Alternatively, this process might solubilize the iron, making it bioavailable (Wang and Newman, 2008). Ultimately, the purpose of phenazine excretion may be illuminated when workers identify the regulatory signals that control it.
In any case, these drugs are toxic to bystander bacteria. While redox compounds are best known for their ability to generate reactive oxygen compounds (Hassan and Fridovich, 1978), we found that their univalent redox activity allows them to damage iron–sulfur clusters directly. Other detrimental effects are also known. When they abstract electrons from NADPH-reduced enzymes and dump them into the respiratory chain, redox drugs drain the cell of NADPH and potentially interfere with biosynthetic processes that require it. Of course, simply by oxidizing these enzymes, the drugs act as competitive inhibitors of them. Prior work has shown that these drugs overoxidize NADH dehydrogenase II to a form that is inactive in vivo (Imlay and Fridovich, 1992).
Many of the redox-cycling drugs are also effective Michael reagents that undergo addition reactions with cellular nucleophiles – a reaction that does not involve redox cycling at all (Bolton et al., 2000; Monks and Jones, 2002; Sachdeva et al., 2005; Wang et al., 2006). In this way they form adducts with available thiols, including those of glutathione and enzyme active-site cysteine residues, and with the exocyclic amino groups of nucleic acids (Bolton et al., 2000; Monks and Jones, 2002). It is not clear which of these mechanisms comprises the greatest threat to a growing cell; presumably the answer varies from compound to compound, as facility at redox-cycling and at Michael activity need not correlate. SOD-deficient E. coli are hypersensitive to these drugs in aerobic complex medium, indicating that in the absence of SOD their toxicity is mediated primarily by superoxide (Hassan and Fridovich, 1978). However, SOD overproduction does not provide much more resistance to wild-type strains (Imlay and Fridovich, 1992), suggesting that in drug-exposed cells the SoxRS-mediated induction of SOD proceeds to a point that something other than superoxide becomes the primary threat. Interestingly, adriamycin and daunomycin are related quinone-based cancer therapeutics whose use is limited by their toxicity; most work suggests that Michael adduction, rather than redox-cycling, underlies their ability to kill mammalian cells (Cornwell et al., 2003; Lame et al., 2003; Marciniak et al., 2004; Wang et al., 2006).
Thus it is reasonable that bacteria that are targeted by these compounds use a defensive regulon that does not depend upon superoxide as a signal, that functions even in anaerobic environments, and that addresses injuries that are different than those produced by reactive oxygen species. This perspective – that the threat is the redox-cycling drugs themselves, not just the superoxide they create – may also clarify the physiological logic of the SoxRS regulon. Because NADPH depletion is a likely consequence of redox-cycling, it seems plausible that the SoxRS induces glucose-6-phosphate dehydrogenase and ferredoxin:NADP oxidoreductase in order to restore the NADPH pool. YggX evidently assists the repair of iron–sulfur clusters, whether damage is mediated by superoxide or directly by the drugs themselves. Yet the predominant strategy of the E. coli SoxRS regulon is to minimize the intracellular drug concentration through mechanisms that impede their entry, chemically modify them or pump them out. Drug excretion, in fact, is the SoxR-regulated function that seems most widespread, being found, for example, in Pseudomonad SoxR regulons that do not induce SOD at all (Park et al., 2006). Thus the emerging view is that the SoxRS regulon is perhaps best construed as a defensive system against redox-cycling drugs rather than against superoxide; superoxide is apparently only one element of the threat that these drugs comprise.
Chemicals and strains
Fumaric acid, potassium nitrate and HEPES were obtained from Fisher Scientific. Sodium dithionite was purchased from Fluka. Sodium molybdate was from Mallinckrodt. Methyl viologen (paraquat), phenazine methosulfate, menadione sodium bisulfite, potassium ferricyanide, L-amino acids, IPTG, o-nitrophenyl-β-D-galactopyranoside (ONPG), xanthine, xanthine oxidase from bovine milk, copper-zinc SOD from bovine erythrocytes, horse heart cytochrome c, ferrous ammonium sulfate, dithiotheitol (DTT), L-malic acid and 30% hydrogen peroxide were from Sigma-Aldrich.
Strains and plasmids used in this study are listed in Table S1. Null mutations were created by the Red recombinase method (Datsenko and Wanner, 2000) and were confirmed by PCR analysis. Mutations were introduced into new strains by P1 transduction with selection for linked antibiotic resistance markers (Miller, 1972). The presence of mutations was then confirmed by phenotype, enzyme assay or PCR analysis. To create plasmids overexpressing yggX, the yggX ORF was PCR-amplified from E. coli GC4468 by using the forward primer 5′- ACTGCGAATTCTAGCGGCTCCCGTGGAG and the reverse primer 5′- TCACCTCTAGAGGAGATGAGCAACGGCG. The EcoRI and XbaI sites are underlined. PCR products were digested and cloned into pCKR101 vector behind a tac promoter to generate the pyggX plasmid. The insert was verified by sequencing. The psoxS plasmid was made using the same method. The forward primer was 5′-GTGACGAATTCGCGCCGATACCGCCAC and the reverse primer was 5′-TCACCTCTAGAGCGAAGGCGATGCCGC.
Luria–Bertani medium (LB) contained (per litre) 10 g of tryptone, 5 g of yeast extract and 10 g of NaCl. Defined media contained minimal A salts (Miller, 1972) supplemented with 0.2% glucose, 1 mM MgSO4 and 5 mg ml−1 thiamine. Where indicated, L-amino acids were added to a final concentration of 0.5 mM. Nitrate medium contained 20 mM KNO3 and 5 µM sodium molybdate, and fumarate medium contained 20 mM fumaric acid. The plasmid pCKR101 and its derivatives were maintained with 100 µg ml−1 ampicillin, and pKK1 was maintained with 12 µg ml−1 tetracycline.
Anaerobic experiments were conducted in a Coy anaerobic chamber under an atmosphere of 85% N2/10% H2/5% CO2. Media and plates used in anaerobic experiments were moved into the chamber while still hot and were allowed to equilibrate with the anaerobic atmosphere for at least 24 h before use. Chemicals used in anaerobic experiments were brought into the chamber and dissolved in anaerobic buffer. Overnight cultures were diluted to approximately 0.01 OD600 and grown for four generations. These exponentially growing cells were then subcultured to an OD600 of 0.005–0.01 for subsequent experiments. Anaerobic cultures were grown in a 37°C incubator in the chamber, and aerobic cultures were grown with vigorous shaking at 37°C.
SoxR protein was purified from aerobic cultures of the soxR overexpressing strain XA90/pKOXR (Nunoshiba et al., 1992), which was kindly provided by Dr Huangen Ding from Louisiana State University. All purification steps were performed at 4°C under aerobic conditions as described (Hidalgo and Demple, 1994). The isolated enzyme exhibited the features of the oxidized cluster (Fig. S4). It lacked an apparent EPR signal (< 5% of the signal after subsequent dithionite addition), indicating that essentially all of the purified enzyme was isolated in the oxidized form.
Cell extracts were prepared by either sonication or passage through a French press. Total protein content was determined using the Coomassie Blue dye-binding assay (Pierce). β–galactosidase activity was assayed as described (Miller, 1972). The activity of Edd was assayed by the two-step procedure for determination of pyruvate (Gardner and Fridovich, 1991b). Fumarase activity was assayed by monitoring the conversion of 50 mM l-malate to fumarate at an OD of 250 nm in 50 mM sodium phosphate buffer (pH 7.3) (Massey, 1955). Oxidant-resistant fumarase C activity was measured after fumarase A and B activities had been inactivated by 2 mM H2O2 for 10 min at room temperature.
Superoxide dismutase activity was assayed by the xanthine oxidase-cytochrome c method (McCord and Fridovich, 1969). To test whether SoxR could inhibit cytochrome c reduction by competing for O2-, purified SoxR protein was reduced by dithionite anaerobically, excess dithionite was removed by centrifugation through a spin column (Amicon Ultra 3 K MWCO, Millipore) at 7000 g and 4°C anaerobically, and reduced SoxR was added to the assays before the addition of xanthine. Xanthine oxidase makes O2-, and O2- reduces cytochrome c. The rate constant for cytochrome c reduction is 2.6 × 105 M−1 s−1. By comparing the rates of cytochrome c reduction in the presence and absence of reduced SoxR, one can calculate how fast O2- is consumed by SoxR. In order to estimate how fast fumarase A reacts with O2-, purified fumarase A was added to the assays before the addition of xanthine.
Reconstitution of Fe–S clusters
Damaged clusters in the [3Fe–4S] form were chemically repaired by incubation with 100 µM Fe(NH4)2(SO4)2 and 2.5 mM DTT anaerobically for 8 min at room temperature. Reconstitution of more extensively degraded Fe–S clusters was carried out by incubation with 500 µM Fe(NH4)2(SO4)2, 5 mM DTT, 2.5 mM cysteine and purified IscS (0.16 mU) in anaerobic buffer at room temperature for 20 min. To purify IscS, the gene iscS was inserted into pET15b, and the resulting pIscS-(His)6 was overexpressed in E. coli BL21 strain. The overexpressed IscS-(His)6 was purified using His Gravitrap (GE Healthcare) (Jang and Imlay, 2010).
The reduced [2Fe–2S]+ cluster has a characteristic EPR spectrum near g = 1.93; the oxidized cluster is EPR-silent (Ding and Demple, 1997; Gaudu et al., 1997). Thus, one can monitor the redox state of SoxR by EPR. Whole-cell EPR samples were prepared from SoxR-overproducing strain XA90/pKOXR grown anaerobically in LB plus ampicillin. To overexpress SoxR protein, IPTG was added when the cells reached A600 = 0.10, and the cells were incubated at 37°C for another 2 h. Then the cells were left untreated or treated with redox-cycling drugs for 40 min anaerobically. The cells were harvested, washed quickly with minimal salts, and resuspended at 1/500th of the original culture volume in cold anaerobic 10% glycerol. The cell suspensions (350 µl) were then transferred into EPR tubes and immediately frozen on dry ice.
For study of purified SoxR, 15 µM protein was reduced by dithionite anaerobically. Excess dithionite was removed as described above. The reduced SoxR protein was then treated with redox-cycling drugs anaerobically, transferred into EPR tubes and frozen. EPR spectra of [2Fe–2S]+ clusters were obtained at the following settings: microwave power, 1 milliwatt; microwave frequency, 9.05 GHz; modulation amplitude, 12.5 G at 100 KHz; time constant, 0.032; and sample temperature, 15 K.
Inactivation of purified fumarase A
Fumarase A was purified from anaerobic cultures of the fumA-overexpressing strain SJ50 (Jang and Imlay, 2007). All purification steps were conducted anaerobically at chamber temperature (27°C). Inactivation of fumarase A was accomplished by the addition of redox-cycling drugs to the purified enzyme in anaerobic buffer. Damage was terminated by addition of 20 mM L-malate before anaerobic assay.
To inactivate fumarase A by O2-, 0.01–0.04 µM fumarase A was exposed to 50 µM xanthine and 14 mU ml−1 xanthine oxidase in aerobic buffer, which generated 0.9 µM O2- in the first 1 min. At time points, damage was terminated by the addition of > 22 U ml−1 SOD. L-malate was then added, and fumarase A activity was measured.
Oxidation of purified SoxR
The SoxR protein was purified in the oxidized form, which displayed characteristic absorption maxima at 414 and 450 nm. When reduced with dithionite, these absorption maxima are diminished (Hidalgo and Demple, 1994). Therefore, one can track the redox state of SoxR by monitoring A414. Purified SoxR protein was reduced by dithionite anaerobically. Dithionite was removed as described above. Reduced SoxR was then transferred to anaerobic cuvettes and exposed to PQ (100–700 µM), menadione (100–700 µM), PMS (30–300 µM) or H2O2 (0.2–2 mM) anaerobically. The oxidation of SoxR was monitored by following A414. When O2 was the oxidant under investigation, reduced SoxR was diluted into aerobic buffer and A414 was monitored. In the case of O2-, xanthine and xanthine oxidase were added to SoxR aerobically.
Determination of intracellular PQ concentrations
Dithionite reduces PQ to a blue radical, which absorbs at 395 nm. The strain TN530 was cultured anaerobically to an OD600 of 0.10, and PQ was added to concentrations at which SoxRS was highly induced. After 1 h incubation at 37°C, cells were harvested and washed. Crude extracts were made by French press. Dithionite was then added, and PQ concentrations were determined by the absorption at 395 nm (ε395 = 0.0132 µM−1). PQ concentrations in the extracts were then converted to concentrations in the cells based on 1 ml of 1 OD600 of cells contains 0.5 µl of cytoplasmic volume (Imlay and Fridovich, 1991).
Detection of PQ oxidation by cell membranes
The wild-type strain (GC4468) was cultured anaerobically in 600 ml of minimal medium supplemented with 0.2% glucose and 40 mM fumarate. At an OD600 of 0.4, cells were harvested, washed with 50 mM KPi pH 7.0 and resuspended in the same buffer. Cells were lysed by passage through a French press, and cell debris was removed by centrifugation at 7000 g × 20 min at 4°C. Inverted vesicles were then isolated by ultracentrifugation (27 000 r.p.m. × 2 h at 4°C). Vesicles were resuspended by pipetting in ∼2.5 ml of anaerobic 50 mM KPi pH 7.0. To detect PQ oxidation, the absorbance of buffer containing 30 µM dithionite and 20 µl of cell membranes (a total volume of 500 µl) was blanked at 385–600 nm. When 5 µM PQ was added anaerobically to the mixture, an absorption peak at 395 nm was observed, which represented the formation of PQ radical. The signal disappeared when 0.1–5 mM fumarate was added anaerobically. Bleaching was not observed if either fumarate or respiratory vesicles were omitted.
We thank Dr Mark Nilges (Illinois EPR Research Center) for assistance with EPR experiments, Dr Huangen Ding of Louisiana State University for generously providing strains, Dr Lee Macomber and Dr Soojin Jang for purifying the fumarase A and IscS proteins used in this study, and Dr Gary Olsen for assistance with phylogenetic analysis. This work was supported by National Institutes of Health Grant GM49640.