|Home | About | Journals | Submit | Contact Us | Français|
Hydroxyurea (HU) specifically inhibits class I ribonucleotide reductase (RNR), depleting dNTP pools and leading to replication fork arrest. While HU inhibition of RNR has been recognized for decades, the mechanism by which it leads to cell death remains unknown. To investigate the mechanism of HU-induced cell death we used a systems-level approach to determine the genomic and physiological responses of E. coli to HU treatment. Our results suggest a model by which HU treatment rapidly induces a set of protective responses to manage genomic instability in the majority of the cell population. Continued HU stress activates iron uptake as well as the toxins MazF and RelE whose activity causes the synthesis of incompletely translated proteins and stimulation of the envelope stress response system. These effects alter the properties of one of the cell’s two terminal cytochrome oxidases in the electron transport chain, causing an increase in the production of superoxide. The increased superoxide production from the respiratory chain together with the increased iron uptake fuels the formation of hydroxyl radicals that contribute to HU-induced cell death. This work significantly expands our understanding of HU-mediated cell death and more broadly suggests a pathway whereby replication fork arrest leads to cell death.
Hydroxyurea (HU) is commonly used in both prokaryotes and eukaryotes to study DNA damage-independent replication fork arrest (Lopes et al., 2001; Sogo et al., 2002; Timson, 1975). HU is exquisitely specific for inhibiting class I ribonucleotide reductase (RNR), the enzyme responsible for the synthesis of dNTPs under aerobic conditions in many organisms including Escherichia coli, yeast, and humans (Rosenkranz et al., 1967; Sinha and Snustad, 1972). Several lines of evidence have indicated that RNR is the primary, if not only, cellular target of HU (Fuchs and Karlstrom, 1973; Kren and Fuchs, 1987; Sjoberg et al., 1986; Sneeden and Loeb, 2004). Depletion of dNTP pools through HU treatment leads to replication fork arrest and subsequent genomic instability (Ahmad et al., 1998; Nakayama et al., 1994), most likely through substrate starvation (Foti et al., 2005; Godoy et al., 2006; Timson, 1975). Although HU has been used to study replication fork arrest for decades, very little is known about the downstream physiological effects of HU-dependent replication fork arrest, especially the mechanism by which HU leads to loss of cell viability and eventual lysis.
An intriguing but unexplained observation is that prolonged HU stress results in cell death and lysis mediated in part by the mazEF and relBE toxin/antitoxin modules (Godoy et al., 2006). Depletion of thymine pools (thymineless death) has similarly been suggested to involve MazF activation in E. coli (Sat et al., 2003). Our observation that variants of the translesion DNA polymerase UmuC, acting in combination with the translesion polymerase DinB and the umuD gene products, may mitigate such mazEF- or relBE-induced death has led us to suggest that HU-induced death of E. coli is brought about, not by stalled replication forks directly, but rather through a series of downstream processes involving the toxin/antitoxin pairs mazEF and relBE (Godoy et al., 2006).
To understand how HU-dependent replication fork arrest leads to cell death we used a systems-level analysis to identify the genomic and physiological effects of HU treatment. We present a mechanism whereby HU treatment rapidly induces several survival responses including upregulation of the SOS response, downregulation of cell division inhibition and induction of RNR synthesis. However, as HU stress continues, toxin modules MazF and RelE are activated triggering a cascade of events that eventually results in the production of deleteriously hydroxyl radicals. The production of hydroxyl radicals is exacerbated by increased iron uptake, and these harmful reactive oxygen species contribute to the majority of HU-mediated cell death in E. coli.
When exponentially growing E. coli strain MC4100 is treated with 100 mM HU in liquid culture, cell growth quickly ceases. However, cells do not begin to die until 2 to 3 hours into the treatment, and cell survival declines to approximately 0.1 % by 6 hours (Figure 1). We examined gene expression profiles of exponentially-growing MC4100 cultures following 1 h of treatment with +/− 100mM HU. At this 1 h time point, HU treated cultures do not show decreased survival, but do show growth inhibition (Figure 1A). We hypothesized that expression profiles at this time during HU treatment would allow us to gain insights into the early cellular events that lead to cell death and lysis.
RNA from 3 independent cultures treated −/+ HU were analyzed using Affymetrix Antisense Genome microarrays. The expression results were integrated into an E. coli microarray expression database (Faith et al., 2007 PLOS Biology http://m3d.bu.edu/) for analysis comparing the HU results to over 500 additional expression profiles (Faith et al., 2007). We examined significance (z-score) of gene expression on a gene-by-gene basis through comparison of expression levels +/− HU treatment in the framework of the mean and standard deviation of expression for each gene across all conditions in the compendium (see Experimental Procedures for additional details). This allowed us to determine the statistically significant changes in expression due to HU treatment in terms of units of standard deviation by subtracting, on a gene-by-gene basis, the z-score for an untreated control from the z-score for an HU treated sample. This procedure allowed us to filter our expression profile to identify those genes that showed altered transcript level specifically in response to HU (Table S1).
Consistent with previous observations (Gibert et al., 1990), we found a significant increase in gene and protein expression of the components of class I RNR, NrdA and NrdB, in response to HU treatment (Table 1, Figure S1A, B). In addition, genes encoding E. coli anaerobic class III RNR (nrdDG) and a cryptic RNR (nrdEF) (Jordan and Reichard, 1998) showed significant transcriptional increases (Table 1). We were surprised to find substantial induction of nrdDG, since any NrdDG produced would be irreversibly inactivated under our aerobic conditions. These observations suggest that upregulation of all RNR genes in response to HU-induced dNTP pool depletion is a strong survival response.
We also observed that components of the primosome, PriA and PriB, were substantially upregulated by HU treatment (Table 1). PriA and PriB can assemble forks on either the leading or lagging strand (Heller and Marians, 2005; Heller and Marians, 2006; Lovett, 2005). The PriA pathway for replication restart is most efficient on fork structures without gaps in the leading strand. This structure could be formed after the collapse of a replication fork. An alternative PriC-dependent pathway preferentially utilizes forks with large gaps in the leading strand. We did not observe upregulation of priC (Table 1), an observation consistent with the mode of fork damage induced by HU treatment. Upregulation of primosome components may help restart replication forks stalled or collapsed following HU treatment. Notably, induction of primosome components is not part of the SOS response (Simmons et al., 2008a).
Consistent with previous work (Barbe et al., 1987), our microarray analysis revealed that numerous genes in the DNA damage responsive SOS network were induced by HU treatment (Table 1). The SOS response involves the upregulation of 57 genes involved in numerous aspects of DNA repair and other cellular functions (Simmons et al., 2008a). In agreement with the transcriptional data, SulA and RecA protein levels showed a significant increase following HU challenge (Figure S1C, D).
Recent work has shown that a small subpopulation of cells can account for large transcriptional changes in microarray analyses of the SOS response (Britton et al., 2007; Simmons et al., 2008b). To determine the population of SOS-induced cells after HU treatment, we used the pL(lexO)-GFP construct in which GFP expression is controlled by LexA (Dwyer et al., 2007) as a single-cell marker of SOS induction. No GFP fluorescence was observed in the absence of HU (data not shown) but we did observe GFP fluorescence soon after HU treatment (Figure 1Bi). After a 30 min exposure to HU, GFP fluorescence was clearly visible in the vast majority of cells (Figure 1Bii), indicating that SOS induction occurred uniformly in the bulk of the population.
To explore the contribution of the SOS response to HU survival, we tested the sensitivity of the lexA3 mutant strain to increasing amounts of HU (Figure 1C). The lexA3 strain carries a non-cleavable form of LexA effectively preventing SOS induction (Friedberg, 2005; Mount et al., 1972). The lexA3 strain was approximately 10 fold more sensitive relative to the parental strain over a range of HU concentrations indicating that activation of the SOS-regulated genes helps cells survive HU exposure. HU is much less stable in liquid culture then in solid LB agar (Lovett, 2009), possible due to increased oxidation of HU in liquid. We found that 100 mM HU was required for significant killing in liquid cultures (Figure 1A) while 5-30 mM HU was sufficient to observe growth difference between strains in LB agar plating assays (Figure 1C).
HU treatment induces extreme filamentation (Godoy et al., 2006) which could occur due to SulA-mediated inhibition of septum formation. However, upregulation of SulA alone cannot completely account for the observed increase in filamentation, since HU-induced filamentation still occurs in a ΔsulA background (Figure S1E, F). A functionally linked cluster of genes involved early in cell division showed significant down-regulation in our microarrays including several genes involved in septum formation (e.g., ftsZ, ftsQ, zipA) and a positive regulator of ftsZ, rcsB (Table 1). In agreement with the transcriptional result, we observed a decrease in FtsZ protein level (Figure S1G). To determine if the downregulation of genes required for septum assembly was sufficient to perturb septum formation, we analyzed Z-ring formation at the single-cell level using an ftsZ-GFP fusion (Weiss et al., 1999). FtsZ-GFP does not functionally complement endogenous FtsZ, but is able to decorate the wild-type protein forming Z rings allowing for visualization by fluorescence microscopy (Ma et al., 1996). Induction of FtsZ-GFP by IPTG was comparable in HU-treated and non-treated cultures (Figure S1H). Under non-stress conditions, we observed septal ring formation in MC4100 cells at mid-cell (Figure 1D). However, when ftsZ-GFP was induced following HU treatment, the elongated cells did not show Z-ring formation (Figure 1D). In fact we did not observe Z-rings in any HU treated cells (data not shown). Together these results strongly suggest that the decreased expression of ftsZ and other septum assembly genes is sufficient to perturb normal septum formation.
We also observed several functional classes of genes whose expression changes we could not easily explain as survival responses. Our microarray analysis indicated that many genes encoding the Fep, Fec, Fhu and Feo iron uptake systems (Postle and Larsen, 2007) were significantly upregulated after 1 h of HU treatment (Table 1). Iron is required for numerous biological purposes, including playing a catalytic role in class I RNR (Stubbe, 2003), and so it seemed possible that increased iron uptake in response to HU represents an effort to replenish RNR. However, excessive amounts of iron can be deleterious to E. coli, catalyzing the production of highly destructive OH• through Fenton chemistry (Jakubovics and Jenkinson, 2001). To test whether such a strong HU-induced iron uptake had an overall beneficial or detrimental effect on E. coli we disrupted tonB, which is a component of the TonB/ExbB/ExbD inner membrane protein complex. TonB is important for iron uptake because it couples the dedicated outer-membrane iron transporters FepA, FecA and FhuE to the proton-motive force (Wandersman and Delepelaire, 2004; Wiener, 2005). We found that disruption of tonB increased resistance of the strain to HU (Figure 2A) suggesting that the strong upregulation of iron uptake has an overall negative impact on cell survival under HU stress.
We asked if the increased iron uptake occurred on the same time scale as HU-protective responses such as the SOS response. To explore the temporal relationship between iron uptake and HU treatment, we observed MC4100 carrying plasmid pL(furO)-GFP that expresses GFP under the control of the iron-responsive transcriptional regulator Fur. Fur controls expression of three major iron transport systems upregulated in our microarray, Fec, Fep and Fhu. In contrast to induction of the SOS-regulated GFP response, we did not observe Fur-regulated GFP expression after 15 min of HU treatment and observed only minor induction at 30 min (Figure 1Biii, iv). Under the same exposure conditions, we did observe a significant increase in Fur-regulated GFP expression after 1 h of HU treatment consistent with our microarray data (Figure 2B). Interestingly, our microscopy indicated that there were two subpopulations of GFP-expressing cells after 1 h of HU treatment. Both subpopulations showed GFP fluorescence above the untreated control, but one subpopulation was more fluorescent than the other.
We further explored the relationship between iron uptake and cell death by staining MC4100 cells carrying pL(furO)-GFP with propidium iodide. Propidium iodide is used as a marker for cell death as it is only able to enter and stain bacteria that have damaged cell membranes. After 3 h of HU-treatment we found that 20% of cells stained with propidium iodide and that 98% of propidium iodide stained cells also showed increased levels of Fur-regulated GFP expression (n = 810 cells). In sum, these results indicated that, unlike the uniform SOS survival response, the induction of genes involved in iron uptake occurs later, to two distinguishing levels and correlates with cell death.
We had speculated that increased iron uptake might be detrimental to the cell by increasing OH• formation (Imlay et al., 1988). To test this idea, we monitored the fluorescence of the dye, 3′-(ρ-hydroxyphenyl)-fluorescein (HPF), in cultures after HU treatment (Kohanski et al., 2007). HPF is oxidized with high specificity by OH• to a fluorescent product (Setsukinai et al., 2003). 1 h after HU treatment, we observed OH• formation in MC4100 that steadily increased until reaching a maximum at 4 h (Figure 2C). The first major increase in OH• formation coincides exactly with the first major decrease in cell viability after HU exposure (Figure 1A). This correlation suggests that HU-induced cell death may be due, at least in part, to increased OH• levels.
To test if the increased OH• production we observe following HU treatment is causally related to cell death, we quenched OH• using the potent OH• scavenger thiourea (Novogrodsky et al., 1982; Repine et al., 1981; Touati et al., 1995). While similar in structure to HU, thiourea is not a potent inhibitor of RNR because it lacks the key hydroxyl group that inactivates RNR and thus does not affect culture growth (Figure 2D) (Stubbe, 2003). Addition of thiourea to cultures essentially prevented HU-induced cell death (Figure 2D) and strongly decreased OH• formation (Figure 2E). In other words, thiourea changes HU from a bactericidal stress into a bacteriostatic stress. Thiourea did not interfere with the HU-dependent arrest of DNA synthesis (Figure S2B). Since thiourea has been shown to scavenge another radical, HOCl, as well (Wasil et al., 1987), we cannot rule out the possibility that thiourea may be quenching additional unknown radicals that also contribute to HU-induced cell death. Nevertheless, the virtually complete abrogation of cell death brought about by thiourea treatment strongly suggests that OH• and possibly other radicals play a causal role in the chain of events that leads to cell death in response to prolonged HU treatment.
OH• formation via the Fenton reaction occurs when hydrogen peroxide (H2O2) interacts with ferrous iron. There are a number of enzymes which can remove harmful reactive oxygen species (ROS), and in E. coli, ahpCF encodes for one of the key H2O2 scavenging enzymes (Seaver and Imlay, 2001). We examined survival of an ΔahpC strain and found that loss of ahpC increased sensitivity to HU over a range of HU concentrations (Figure 2F). This result strongly supports our inference that HU induces OH• formation via Fenton chemistry and that this ROS contributes to HU-mediated cell death.
Our results indicate that OH• play a causal role in HU-mediated cell death, yet blocking iron uptake by deleting tonB results in only a modest increase on survival in the presence of HU. Thus, while increased iron likely contributes to HU-mediated cell death by promoting OH• formation, our results suggest that there must be additional factors responsible for the high levels of OH• production after HU treatment.
Although RNR is widely thought to be the sole cellular target of HU (Fuchs and Karlstrom, 1973; Kren and Fuchs, 1987; Sjoberg et al., 1986; Sneeden and Loeb, 2004), it was still possible that the increased iron uptake and OH• production, which contribute to cell death, were off target effects of HU treatment. To address these issues we first treated MC4100 with guanazole and again monitored for increased iron uptake and hydroxyl radical formation. Guanazole is a highly specific inhibitor of RNR that is chemically distinct from HU (Foti et al., 2005; Jones et al., 2007; Moore and Hurlbert, 1985). Treatment of MC4100 with 100 mM guanazole resulted in approximately 0.1 % survival after 6 h, similar to HU treatment (Figure 3A). We monitored iron uptake using plasmid pL(furO)-GFP. Treatment of MC4100 carrying pL(furO)-GFP with either HU (Figure 3B) or guanazole (Figure 3C) resulted in a substantial increase in Fur-regulated GFP fluorescence, as measured by flow cytometry. Furthermore, treatment of MC4100 with guanazole also strongly induced OH• production (Figure 3D). These results support the hypothesis that both increased iron uptake and OH• production are due to inhibition of RNR activity rather than an off-target effect of HU.
To determine whether increased iron uptake and OH• were due strictly to the inhibition of RNR or to the downstream effects of replication fork arrest, we tested a temperature sensitive dnaE mutant (Strauss et al., 2005) for iron uptake and OH• production at the permissive and non-permissive temperatures. dnaE codes for the α-catalytic subunit of DNA polymerase III holoenzyme and is essential for DNA synthesis (Gefter et al., 1971; Wickner and Kornberg, 1974), and incubation at the non-permissive temperature leads to replication fork arrest in the dnaE(ts) mutant. At the permissive temperature, the dnaE(ts) strain was identical to the parental strain with respect to growth, expression of iron uptake and OH• formation (data not shown). After 5 h at the non-permissive temperature of 40° C, the dnaE(ts) strain viability had decreased to 0.1 % (Figure 3E). Interestingly, at the non permissive temperature the dnaE(ts) strain did not exhibit a change in iron uptake response (Figure S2B, C) but did show a significant increase in OH• formation (Figure 3F) compared to the wild type control strain. In fact, the first major increase in OH• production was observed at the same time as the first major decrease in dnaE(ts) mutant cell survival. These observations support our conclusion that, in addition to the increase in iron uptake caused by HU treatment, there must be additional cellular responses resulting from the blocked replication fork that play a major role in OH• production and subsequent cell death.
The major source of reactive oxygen species in the cell is from the inappropriate reduction of oxygen during electron transport through the respiratory chain (Imlay and Fridovich, 1991). It was recently shown that lethality of a temperature sensitive DNA polymerase III mutant could be overcome (Strauss et al., 2005) by disrupting one the major respiratory cytochrome oxidases, cytochrome oxidase bd (Poole and Cook, 2000; Strauss et al., 2005). We hypothesized that replication fork arrest was ultimately disrupting electron transport chain activity which was leading to increased ROS production. Agreeing with this, we found that deleting cydB, which encodes subunit b of one of the major respiratory terminal cytochrome oxidases, prevented HU-induced cell death (Figure 3G). Furthermore, deletion of cydB nearly completely prevented HU-induced OH• formation (Figure 3H).
Our observations that the dnaE(ts) mutant produced OH• at the non permissive temperature and that the ΔcydB mutant exhibited increased survival following HU treatment point toward replication fork arrest as a key element underlying ROS formation during HU stress. It was also recently shown that the action of MazF can mediate cell death in response to certain antibiotics by producing reactive oxygen species (Kolodkin-Gal et al., 2008). Interestingly, we had previously shown that disruption of either toxin-antitoxin pairs, mazEF or relBE, in the E. coli HM21 and P90C strain backgrounds increased HU resistance (Godoy et al., 2006). These observations led us to ask whether the mazEF and relBE toxin/antitoxin modules contributed to the death of HU-treated strain MC4100 via an OH•-mediated mechanism, and if this is connected to HU-mediated replication fork inhibition. When challenged with HU, we found that in strain MC4100, deletion of mazEF led to increased resistance to HU (Figure 4A). The MC4100 ΔrelBE strain, however, was indistinguishable from the parental strain for HU survival (Figure 4B). Strikingly, we found that deletions of both mazEF and relBE act synergistically to promote HU resistance to levels far above that observed in a ΔmazEF strain alone (Figure 4C) and similar to that observed with the ΔcydB strain. Deletion of relBE did not have a detectable effect on OH• formation while deletion of mazEF caused a small decrease in OH• formation (Figure S2D, E). The ΔmazEF ΔrelBE strain exhibited a stronger reduction in OH• levels (Figure S2F), consistent with the observed increase in survival (Figure 4C). We also note that our microarray results showed that HU induced only small increases in expression of mazF relative to mazE and of relE relative to relB (Table 1) suggesting that HU treatment appears to be affecting toxin activation mainly at a post-transcriptional level.
We have shown above the increased iron uptake has overall negative effects on HU survival. To determine the relationship between increased iron uptake and toxin induced cell death, we constructed a ΔmazEF ΔrelBE ΔtonB strain and tested its sensitivity to HU. We found that our ΔmazEF ΔrelBE ΔtonB was more resistant to HU than our ΔmazEF ΔrelBE strain (Figure 4D). This suggests that increased iron uptake enhances toxin-mediated cell death, which likely occurs via an OH•-dependent mechanism. Consistent with these data, we found that deleting tonB in combination with ΔmazEF ΔrelBE results in a significant decrease in the levels of OH• following HU treatment (Figure 4E).
We next considered how toxin activity could lead to radical-mediated cell death. A number of chromosomally encoded toxins have functions which directly affect protein synthesis. HipA was recently shown to phosphorylate the translation-related protein EF-Tu (Schumacher et al., 2009), and the toxins MazF and RelE are ribonucleases that cleave mRNA (Christensen and Gerdes, 2003; Zhang et al., 2005) thereby disrupting protein translation. Improper protein synthesis can lead to problems with protein folding, and this has been associated with protein carbonylation (a marker of oxidative stress) (Dukan et al., 2000) and ROS formation following treatment with aminoglycosides (Kohanski et al., 2008). Recent work has also shown that MazF overexpression can lead to protein carbonylation (Kolodkin-Gal et al., 2008). We hypothesized that HU stimulates MazF and RelE activity which cleave mRNA leading to truncated proteins, protein misfolding and ultimately oxidative stress and cell death.
To test this idea, we investigated protein translation following HU treatment to see if replication fork inhibition affects protein synthesis and if such an affect is toxin-dependent. Disruption of protein translation can lead to stalled ribosomes that can be rescued by tmRNA (Moore and Sauer, 2007). tmRNA frees stalled ribosomes by releasing both mRNA and the incompletely translated protein. We used the tmRNA system developed by Moore et al., which adds a 6X His peptide tag to the end tmRNA released proteins, to monitor the affect of toxin activity during HU treatment (Moore and Sauer, 2005). We found that HU treatment induced a significant increase in tmRNA His-tagged proteins (Figure 5A). This increase in tmRNA His-tagged proteins was alleviated by deleting both toxins MazF and RelE. This suggests that toxin activity during HU treatment likely leads to the improper translation of proteins.
Improperly translated proteins can also lead to membrane stress (Alba and Gross, 2004). To test if the HU treatment leads to membrane stress, we first used an rpoHP3-lacZ reporter to measure RpoE activity following HU treatment. RpoE is a sigma factor whose activity is induced by membrane stress (Alba and Gross, 2004). After only 1 h of HU treatment we observed a 2-fold increase in RpoE activity (Figure 5B). This increase in RpoE activity was prevented by deletion of both toxins MazF and RelE confirming that toxins stimulate envelope stress, likely through a mechanism involving improperly translation proteins. We were unable to assess RpoE activity at later time points since the strain lacking both toxin-antitoxin pairs rapidly approached stationary phase which induces RpoE activity (Kabir et al., 2005).
To confirm that membrane stress affects HU-mediated survival, we looked at the responses of ΔcpxA and ΔdegP mutant strains to HU treatment. Disruption of membrane proteins also leads to the activation of the CpxAR two-component signaling system (CpxA is the sensor protein and CpxR is the cognate transcription factor), which monitors the fidelity of proteins trafficked across the inner membrane and works in conjunction with σE to activate elements of the envelope stress response system such as the periplasmic protease, degP (MacRitchie et al., 2008; Ruiz and Silhavy, 2005). Recent evidence suggests that mistranslated proteins can induce OH• through Cpx-mediated perturbation of envelope stress and metabolism, and that deletion of the cpxA or degP significantly decreased OH• production (Kohanski et al., 2008). In agreement with this, we found that deletion of cpxA resulted in a striking rescue from HU-induced cell death (Figure 5C), comparable to that achieved by deleting cydB or mazEF, relBE and tonB. Deletion of cpxA also substantial reduced HU-induced OH• production (Figure 5D). Furthermore, we also observed that deletion of degP also conferred significant resistance to HU-induced cell death (Figure 5C).
If HU-induced membrane stress leads to OH• formation by disrupting the normal electron flow through the electron transport chain, then we hypothesized that HU treatment should also lead to changes in membrane potential. We HU treated cells with the bis-oxonol dye DiBAC which diffuses across depolarized yet intact cell membranes (Jepras 1997). We examined cells 3 h after HU treatment as cell death appears to have begun by this time (Figure 1A) and observed a significant number of DIBAC stained cells (Figure 5F). Furthermore we found that only cells strongly expressing Fur-regulated GFP stained with DIBAC (Figure 5G). This agrees with our observation above that, overall, increased expression of iron uptake genes is associated with HU-induced cell dysfunction.
The most common ROS released from the respiratory chain is superoxide (Imlay and Fridovich, 1991). Our results demonstrating a connection between toxin activity and membrane stress together with our earlier results showing ΔcydB is substantially less sensitive to HU treatment suggests that HU induced changes in membrane composition and function may lead to superoxide formation. Superoxide is converted to hydrogen peroxide (H2O2) that can react with ferrous iron to produce OH• (Jakubovics and Jenkinson, 2001). It is thus possible that toxin-mediated superoxide formation can contribute to an increase in Fenton chemistry, which would explain the significant decrease in cell death in the ΔmazEF ΔrelBE ΔtonB strain. To monitor superoxide production we used plasmid pL(soxS)-GPF which expresses GFP under the control of the superoxide sensing regulator SoxS (Dwyer et al., 2007). We observed an increase in SoxS- regulated GFP expression following HU treatment. SoxS-regulated GFP expression was significantly decreased in the ΔmazEF ΔrelBE mutant (Figure 5H) which demonstrates that HU-mediated toxin activity contributes to cell death via superoxide formation.
Our work describes the previously unknown chain of events by which HU-induced DNA replication arrest leads to cell death. On the basis of our findings, we propose the following mechanism whereby HU treatment results in cell death (Figure 6). HU inhibits RNR leading to dNTP depletion and an arrest of normal DNA replication (Figure 6A). dNTP depletion results in the activation of toxins MazF and RelE. The activity of these toxins produces improperly translated proteins leading to membrane stress and activation of membrane stress responses (Figure 6B). These effects alter the properties of one of the cell’s two terminal cytochrome oxidases in the electron transport chain, causing an increase in the production of superoxide, which is then converted to hydrogen peroxide (Figure 6C). The hydrogen peroxide reacts with free ferrous iron resulting in the formation of OH•, which contribute to cell death (Figure 6D). HU treatment also induces a strong iron uptake response (Figure 6E), which may be part of an initial survival response to increased demand for synthesis of the iron-utilizing class I RNR. However, due to replication fork arrest, toxin activity and membrane stress, HU-induced iron uptake becomes misregulated and exacerbates OH• production (Figure 6E).
Our work also shows that HU initially induces cell survival responses beyond the classic SOS response. Specifically, we show that HU induces genes required for primosome formation and downregulates genes required for cell division. Induction of primosome genes could help restart collapsed replication forks that have resulted from HU treatment. Downregulation of cell division genes would retain all copies of the chromosome in one compartment, thus leaving open the possibility of recombinational repair.
It is striking that HU-induced death, which takes several hours and is preceded by specific cellular changes and associated with cell lysis, can be substantially prevented by adding thiourea or by inactivating cydB, cpxA, or the combination of mazEF, relBE, and tonB. These systems appear to converge on radical formation as a contributing element in HU-mediated cell death, and these observations suggest that there is one predominant mode of death for HU-treated cells, which is in contrast to cells treated with various bactericidal antibiotics for which there are additional modes of death besides that mediated by OH• radicals (Kohanski et al., 2007).
Interestingly, Kohanski et al. (2007) showed that treatment of E. coli with the DNA gyrase inhibitor norfloxacin also induces radical production and cell death, both of which can be partially alleviated by thiourea. However, thiourea rescue of norfloxacin-induced cell death is not as complete as it is for HU. Thus while both HU and norfloxacin inhibit DNA replication and induce lethal radical production, radicals appear to play a more significant role in HU-induced cell death. We believe this difference is due to the exact mode of replication arrest these two compounds elicit and how the cell interprets these modes of replication arrest. A clue that this may be the case come from microarray studies which show that norfloxacin does not induce iron uptake but does cause internal iron release by destabilizing iron-sulfur clusters (Kohanski et al., 2007). Furthermore norfloxacin-mediated radical production is not dependent on toxins while HU radical production is.
Our observations establish a direct link between HU-mediated killing and the respiratory chain (Figure 6C) and suggest that, under the stressed condition resulting from prolonged exposure to HU, the cydAB-encoded heterodimeric cytochrome bd quinol oxidase plays a contributing role in cell death by generating superoxide. The involvement of this particular cytochrome in HU-mediated cell death is intriguing. Typically, cytoplasmic superoxide is not generated by the action of cytochrome oxidases (Imlay, 2003), and both cytochrome bd and cytochrome bo3 have molecular features that prevent the release of incompletely reduced oxygen as superoxide (Tsubaki et al., 2000). Nevertheless, our results suggest that, under conditions of HU stress, cytochrome bd oxidase switches from a protective role to a diametrically opposite role of promoting cell death by producing superoxide. From this perspective, it is intriguing that the d heme, which binds the O2, is only found in this enzyme (Dueweke and Gennis, 1991).
Another unique feature of cytochrome bd oxidase is that its most prominent stable form is the one-electron reduced oxy state (Fe2+-O2) in which oxygen is at the level of superoxide (Belevich et al., 2005; Lorence and Gennis, 1989; Yang et al., 2008). Thus, it is possible that some modification of the cytochrome bd oxidase under conditions of prolonged HU stress could alter its function so that it becomes a producer of superoxide. Such a modification of function could result from a proteolytic cleavage in the stressed cells, perhaps in its protease-sensitive periplasmic loops (Dueweke and Gennis, 1991), in a manner reminiscent of the cleavage of RseA by DegS in response to envelope stress, or in a membrane-spanning domain by RseP (Ades, 2008). Other possibilities include C-terminal truncation by translation of an incomplete message, posttranslation modification, or modulation of the native enzyme by other components present in the stressed cell envelopes. Such alterations could, for example, potentially interfere with the efficiency of electron transfer from ubiquinol to the O2-binding d heme or change the status of the putative O2 channel that has been postulated to flicker between open and closed conformations under normal conditions (Belevich et al., 2007). The clustering of cytochrome bd oxidase in mobile patches in the membrane (Lenn et al., 2008) might have the effect of speeding up killing by concentrating hydroxyl radical production in particular regions of the cell.
Toxin/antitoxin modules are found both on plasmids and on bacterial chromosomes. It is widely accepted that the toxin/antitoxin pairs located on plasmids serve as “addiction” modules by killing bacteria that have lost the plasmid (Couturier et al., 1998; Engelberg-Kulka et al., 2006; Gerdes et al., 2005; Hayes, 2003; Jensen and Gerdes, 1995). There is evidence suggesting that toxin/antitoxin modules serve as growth modulators that help bacterial cells to manage a variety of physiological stresses (Buts et al., 2005; Gerdes et al., 2005). The proposal that chromosomally-encoded TA pairs may mediate cell death in response to certain stresses has proven to be controversial (Engelberg-Kulka et al., 2006; Gerdes et al., 2005; Kolodkin-Gal and Engelberg-Kulka, 2006; Magnuson, 2007; Saavedra De Bast et al., 2008; Tsilibaris et al., 2007).
A recent study has shown that mazEF-mediated cell death appears to occur via ROS-dependent and independent mechanisms, which depends on the production of a density-dependent pentapeptide extracellular death factor (EDF) (Kolodkin-Gal et al., 2008). Furthermore, it has been suggested that the commonly used E. coli strain MG1655 is largely resistant to mazEF-mediated killing because if fails to produce the EDF and responds very poorly to it (Kolodkin-Gal and Engelberg-Kulka, 2008). Although the killing that results from HU exposure is mediated in part by mazEF, it differs from the mazEF-mediated killing discussed above in two important respects. First, we found that HU-induced cell death is independent of culture density (Figure S3A), indicating that activation of HU-induced cell death responses may not require the pentapeptide extracellular death factor described by Kolodkin-Gal and colleagues (Kolodkin-Gal et al., 2007). Second, we found that HU induces cell death and lysis in strains MC4100 and MG1655 with similar kinetics (Figure S3B). Our results suggest that, while HU induces cell death responses that are at least partially dependent on toxin activation, these responses are complex (Figure 6) and quite different from those previously studied (Saavedra De Bast et al., 2008; Tsilibaris et al., 2007).
The mechanism by which HU-mediated dNTP depletion activates the MazF and RelE toxins and causes envelope stress is not yet clear. We hypothesis that HU-dependent depletion of dNTPs and subsequent replication fork arrest leads to gaps in the DNA that interrupt transcription and subsequent translation of several genes including toxin/anti-toxin pairs. Since antitoxins are more labile than their toxin counterparts (Buts et al., 2005; Engelberg-Kulka et al., 2006) inhibition of transcription and translation could lead to rapid loss of antitoxin and activation of toxin. Possible support for this hypothesis comes from our previous observation that certain umuC mutations, such as umuC122::Tn5, permit DNA pol IV (DinB) and DNA pol V (UmuD’2C) to carry out limited mutagenic DNA synthesis that makes cells resistant to killing by HU (Godoy et al., 2006). The Km’s of DNA pol IV and pol V for dNTPs are strikingly lower than those for the replicative DNA polymerase, explaining why they can act in dNTP-depleted cells (Godoy et al., 2006). This limited DNA synthesis presumably stabilizes the replication fork by filling in gaps that would otherwise accumulate under low dNTP conditions.
In this work, we used a systems-level approach to examine the transcriptional response of E. coli following HU treatment, and in doing so expanded our understanding of the events following HU-mediated RNR inhibition and replication fork arrest. We have explored the protective responses induced by HU treatment as well as responses important for cell death, and we implicate toxin-antitoxins, envelope stress, cytochromes and iron-misregulation as key players in an HU-mediated, free radical-based cell death pathway (Figure 6). Moving forward, it will be interesting to see if the cell death mechanism we propose in this work extends past HU inhibition to encompass replication inhibition in general.
Strains and plasmids are shown in Table S2. All strains were grown in Luria-Berani (LB) or M63 minimal medium (+ 0.2% glucose) at 37 °C For ftsZ-GFP assays, LB was supplemented with 0.2 % glucose and 1.0 μM to 2.5 μM IPTG. Where indicated, antibiotics were used at the following concentrations: ampicillin (100 μg/ml except for BWD05 that used 25 μg/ml), kanamycin (30 μg/ml) and spectinomycin (100 μg/ml). Allele transfers were done by P1 transduction.
MC4100 was grown to exponential phase. 3 independent cultures were diluted and grown +/− 100mM HU (Sigma) for 1 h, after which RNA was isolated using the Qiagen RNeasy extraction kit and samples were treated with DNase treated using dna-free (Ambion). cDNA preparation and microarray analysis were performed as described (Dwyer et al., 2007). Results are reported as the average ± standard error.
The resulting microarray *.CEL files were combined with *.CEL files from arrays which comprise the M3D compendium (Faith et al., 2007) (http://m3d.bu.edu; E_coli_v3_Build_3) and RMA-normalized (Bolstad et al., 2003) with RMAexpress, for a total of 530 RMA-normalized E. coli expression arrays. Each gene’s standard deviation of expression, σ, was calculated across the entire compendium, and used to construct the z-scale difference between that gene’s normalized expression following hydroxyurea treatment versus the untreated control:
This allowed us to measure each gene’s change in expression for a given experiment in units of standard deviation, a form of the z-test. We used a 2 tier z-score cutoff platform to identify statistical significance gene expression changes. Genes with z-score of > +2 or < −2 were scored as highly significant and genes with z-scores between +1 and +2 or between −1 and −2 as significant. As described in the manuscript, we tested candidate genes that showed statistically significant gene expression changes from our microarray analysis for biological significance in response to HU treatment.
Liquid culture assays were performed as described (Godoy et al., 2006). Exponentially growing cultures were diluted to OD600 0.01 or 0.0001 prior to treatment. Where indicated, HU, guanazole and thiourea were added to 100 mM for liquid culture assays. For temperature sensitivity assays, exponentially growing cultures where diluted to OD600 0.01 and then shifted to the non-permissive temperate. For chronic assays, strains were serially diluted and plated on LB agar containing increasing amounts of HU. Cfu were counted after 24 h growth at 37°C. We required 100 mM HU to observe significant cell death in liquid culture and between 5 mM to 30 mM to observe significant cell death in plating assays. This is due the decreased stability of HU of liquid culture (Lovett, 2009).
At least 3 independent cultures were grown to mid-log +/− 100 mM HU for 1 h. Cultures were lysed by bead beater. Immunoblotting was performed as previously described (Robichon et al., 2005), loading equal amounts of protein lysate for each sample. Relative protein expression levels were quantitated using a FluorChem HD2 Imaging System (Alpha Innotech Corp.) and expressed as the average ± standard error. LacZ assays were performed as previously described (Sambrook, 2001).
Strains were grown to exponential phase, diluted, and grown for 1 h +/− 100 mM HU. Preparation of cells for live-cell microscopy was done essentially as described (Godoy et al., 2006). Flow cytometry analysis was performed as described (Dwyer et al., 2007). Flow cytometry experiments measured the entire population of cells, and all plots represent analysis of an equal number of cells (≥ 50 000 cells). The relative fluorescence values were determined by taking the mean fluorescence value for each sample at each time point and normalizing that value to the maximum mean fluorescence value that the wild type sample achieved over the time course. Each data point represents the average of three independent measurements of these relative fluorescence values, and the error bars represent the standard error of the independent measurements. Hydroxyl radical formation was measured by 3′-(p-hydroxyphenyl) fluorescein (HPF – Molecular Probes). An example of the Side Scatter and Forward Scatter for one experiment measuring HPF fluorescence is shown in Figure S4. Cell death was measured using Propidium Iodide (Molecular Probes). Cell membrane depolarization was monitored using DiBAC4 (Molecular Probes)
We would like to thank Dr. Joseph Lutkenhaus for anti-FtsZ antibodies, Dr. Bernard Strauss for strains and helpful discussion and Kaleena M. Crafton for microscopy support at the University of Michigan. This work was supported by National Institutes of Health grants GM31010, CA21615-27 (to G.C.W.) and DP10D003644 (to J.J.C), Howard Hughes Medical Institute (J.J.C.), National Science Foundation FIBR program (to J.J.C.), National Sciences and Engineering Research Council of Canada graduate scholarship (to B.W.D.), NCI postdoctoral fellowship, and start-up funds from the University of Michigan (to L.A.S) and Massachusetts Institute of Technology Center for Environmental Health Sciences National Institute of Environmental Health Sciences Grant P30 ES002109. G.C.W. is an American Cancer Society Research Professor.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.