Gyrase inhibitors induce an oxidative damage cellular death pathway in Escherichia coli

Modulation of bacterial chromosomal supercoiling is a function of DNA gyrase-catalyzed strand breakage and rejoining. This reaction is exploited by both antibiotic and proteic gyrase inhibitors, which trap the gyrase molecule at the DNA cleavage stage. Owing to this interaction, double-stranded DNA breaks are introduced and replication machinery is arrested at blocked replication forks. This immediately results in bacteriostasis and ultimately induces cell death. Here we demonstrate, through a series of phenotypic and gene expression analyses, that superoxide and hydroxyl radical oxidative species are generated following gyrase poisoning and play an important role in cell killing by gyrase inhibitors. We show that superoxide-mediated oxidation of iron–sulfur clusters promotes a breakdown of iron regulatory dynamics; in turn, iron misregulation drives the generation of highly destructive hydroxyl radicals via the Fenton reaction. Importantly, our data reveal that blockage of hydroxyl radical formation increases the survival of gyrase-poisoned cells. Together, this series of biochemical reactions appears to compose a maladaptive response, that serves to amplify the primary effect of gyrase inhibition by oxidatively damaging DNA, proteins and lipids.


Identification and functional enrichment of significantly changing genes in response to gyrase inhibition
All genes have a natural expression range which, under typical conditions, is not fully explored. Even when a gene`s dynamic range is well-explored by experiments, a certain expression state will usually dominate. In order to achieve consistency between microarrays, we normalized our data, in conjunction with the M 3D compendium (T. Gardner, Boston University, http://m3d.bu.edu) using the RMA method of normalization (Irizarry et al., 2003). This resulted in a log-scale microarray compendium of d600 chips.
We approximated each genes fluctuation around its respective compendium mean with a gaussian model, where each expression value was replaced with the corresponding zscore around that gene`s mean. Finally, for each point of the time-series, we computed the z-score difference between the gyrase-inhibitor treatment and the untreated control, allowing us to analyze the change of expression of a gene in terms of the estimated standard deviation of that gene (Supplemental Table 1). The standard deviation of any gene in the compendium had to be biased by the experimental conditions of which the compendium is comprised. Nevertheless, we found the standard deviation interval to be a robust representation of the difference of expression even for biased genes, e.g. for LexA, which was heavily perturbed in the compendium. More importantly, this measure was designed to be independent of a gene`s dynamic range and sensitive to the statistical significance of a change of expression between the treatment and control. This allowed us to eliminate genes that change similarly over time in both a gyrase-inhibitor treatment and the control expression sets, and to focus on genes that change expression levels specifically as a function of the treatment using a robust statistical measure as our thresholding parameter. In this regard, it was preferable to the more usual log-ratio metric, which forces the choice of an arbitrary significance threshold independent of a gene`s dynamic range.
Following the detection of the significantly changing genes at each timepoint, we performed functional enrichment using Gene Ontology classification (Ashburner et al., 2000;Camon et al., 2004), and the program GO::TermFinder (Boyle et al., 2004), in order to track temporal gene expression changes ( Supplementary Figures 1 and 2, Supplementary Table 2).. Functional enrichment was performed under the hypergeometric model of random occurrence. In order to simplify the gene expression picture, we reduced the set of differentially expressing genes to the responsible transcription factors in the following manner. First, for each gene in the set of significantly changed genes we determined its transcription factor in RegulonDB 5.0 (Salgado et al., 2006). Then, starting with the most-represented regulator, we removed every gene regulated by a given transcription factor from the set of significantly changed genes, until no genes remained, or until none of the remaining genes had a known transcription factor. We used the resultant set of transcription factors as an approximation of the transcriptional program differentially expressed between gyrase inhibitor treated and untreated cultures.
In addition, we determined statistical enrichment of transcription factors` individual regulons at every time point (Supplementary Table 3). To this end, we restricted the list of differentially expressed genes, constructed as described above, to only those genes whose regulation was described in RegulonDB and a recently published set of regulatory connections (Faith et al., 2007). For each transcription factor both databases, we calculated the likelihood of finding the given number of its targets in this reduced query set using hypergeometric distribution, under the assumption that each transcription factor`s regulon was correctly and completely described by RegulonDB and the regulation map. Finally, as a separate analysis of differential gene expression, we conflated difference of z-scores across all time points by utilizing the formula, Zaverage = sum_t (Z_t)/sqrt(t) (Whitlock, 2005) (Figure 2).

Phenotypic analysis of deletion mutants
To identify potential gyrase inhibitor-mediated genetic responses that contribute to cellular death in E. coli, we screened single-gene knockout strains from a BW25113 deletion library (Baba et al., 2006). Knockout strains were selected based on our microarray functional enrichment results, and we monitored changes in survival (relative to wildtype cells) upon norfloxacin exposure or CcdB expression. We analyzed a subset of genes involved in DNA damage sensing, ATP synthesis, oxidative stress response, Fe-S cluster synthesis and global iron regulation to better understand how these biochemical processes affect the ability of E. coli to survive gyrase poisoning (Supplemetary Figure   7).
We first used a recA deletion strain to examine how the inability to sense DNA damage and initiate the SOS response via RecA would affect cell survival. We observed a drastic 4-log reduction in cell survival in the first hour after application of norfloxacin, and an additional 1-log reduction over the remainder of the experiment. Given these data, the severe reaction of !recA cells treated with norfloxacin suggests that quinolone treatment of these SOS-compromised cells is overwhelming and appears to rapidly breakdown cellular functionality. We observed a similarly strong, near 3-log reduction in viability in the first hour following induction of CcdB expression. Together, these results highlight how rapidly DNA damage occurs following norfloxacin-and CcdB-mediated gyrase inhibition.
In light of the observed downregulated expression of anaerobic respiratory components and upregulation of ATP synthase component genes, we hypothesized that gyrase inhibitor-mediated supercoiling changes lead to a burst in superoxide formation owed to increased respiratory chain activity. We challenged the majority of ATP synthase component genes with norfloxacin and observed increased survival in each case.
Our norfloxacin results demonstrate that efficient killing is directly dependent on ATP, supporting published in vitro and in vivo findings (Kampranis and Maxwell, 1998;Li and Liu, 1998). Further, these results are consistent with previous work (Deitz et al., 1966;Li and Liu, 1998), in which chemical uncoupling of oxidative phosphorylation prevented DNA damage following quinolone poisoning of gyrase.
We next examined the effect of gyrase inhibition on a sodB deletion strain. While !sodB and wildtype cells treated with norfloxacin showed similar growth behavior between the 0 and 2 hour time points, !sodB cells exhibited decreased survival over the final four hours of the experiment. The delay in the phenotypic response to norfloxacin treatment in this strain implies that superoxide generation and accumulation does not occur immediately following DNA damage formation, and is consistent with the steps in our oxidative damage cell death pathway model. Expression of CcdB in !sodB cells similarly resulted in decreased survival when compared to wildtype. Together, these data suggest that SodB-related protection of Fe-S cluster proteins from superoxide damage is an important factor in the survival of gyrase-poisoned bacteria.
We then aimed to determine the role played by Fur, in cell survival following gyrase inhibition. Interestingly, we found that a fur deletion mutant responds slower and survives better than wildtype when challenged with norfloxacin. In contrast, wildtype and !fur cells expressing CcdB respond similarly over the first 3 hours of our experiments, yet do not exhibit the same ability to recover from gyrase poisoning and DNA damage formation. Relative to previous findings, in which exogenous iron import has been implicated in oxidative damage-mediated cell killing following exposure to hydrogen peroxide (Touati et al., 1995), our results imply that internal misregulation of iron is a critical factor in gyrase inhibitor-mediated cell death. This hypothesis was validated by our findings in !tonB cells. We observed no change in cellular viability following gyrase inhibiton by CcdB and norfloxacin in !tonB cells.
Considering that expression of soxS is stimulated by superoxide-mediated oxidative damage to the Fe-S cluster of its transcription factor, SoxR, we next studied whether impairment of Fe-S cluster synthesis affects cell killing by gyrase inhibition. We tested single knockouts of Fe-S cluster synthesis operon genes, iscRSUA. !iscR, !iscU and !iscA deletion mutants all responded to application of norfloxacin in a manner similar to wildtype cultures. We also tested the phenotypic effect of norfloxacin addition on !sufS cells and found that this strain also behaved like wildtype cells.
Lastly we monitored the effect on growth of norfloxacin treatment in both !soxS and !sodA backgrounds. We found that both deletion strains behaved comparably to norfloxacin-treated wildtype cells.

RecA, SoxS and Fur response to gyrase inhibition
On the basis of our phenotypic and microarray results, we contend that gyrase inhibitor-mediated DNA damage promotes the formation of superoxide radicals. In turn, we propose that sustained oxidation of Fe-S clusters by superoxide results in the breakdown of iron regulatory dynamics. To monitor the occurrence of these biochemical events, we designed sensor constructs that activate green fluorescent protein (GFP) expression in response to DNA lesion formation, Fe-S cluster oxidation by superoxide, and derepression of iron-related genes, respectively. Measurement of GFP fluorescence by flow cytometry allowed us to monitor population responses to norfloxacin treatment and CcdB expression at single-cell resolution (Supplementary Figure 6).
Our DNA damage sensor construct employs an engineered promoter that relies upon LexA repression for regulation of gfp transcription, and is thus sensitive to RecAstimulated autocleavage following DNA lesion formation and recognition (Little, 1991).
Gyrase poisoning of wildtype bacteria resulted in significant GFP expression from this sensor. To monitor Fe-S cluster oxidation, we placed gfp under the transcriptional control of the native soxS promoter to create an Fe-S cluster damage sensor; the soxS promoter is regulated by the redox state of the SoxR Fe-S cluster, and superoxidetriggered oxidation activates transcription (Hidalgo et al., 1998). Gyrase inhibition by either norfloxacin or CcdB resulted in GFP expression from this sensor, demonstrating that attack Fe-S clusters by superoxide is occurring and that superoxide attack is sustained. To determine if DNA and oxidative damage promote derepression of iron uptake and utilization genes, we constructed an iron regulation sensor construct that employs the transcription factor Fur as the mediator of gfp transcription. As sustained Fe-S cluster attack by superoxide should increase the intracellular concentration of kfree ironl, an increase in GFP expression would suggest that iron misregulation is occurring.
Accordingly, following norfloxacin treatment and CcdB expression, respectively, we observed increased expression of GFP from our iron regulation sensor construct.
Given the phenotypic effect of genetic-level uncoupling of oxidative phosphorylation and impairment of Fe-S cluster repair by inhibiting de novo cluster synthesis on gyrase inhibitor-mediated cell death, we next monitored the DNA damage, Further, these results highlight the important role of redox cycling in the breakdown of iron regulatory dynamics.
We also monitored the DNA damage, oxidative damage and iron regulatory responses in our !fur, !recA and !sodB strains. GFP expression levels from each sensor, in each strain, were consistent with our phenotypic results.

Time course sensor construct and hydroxyl radical response of wildtype to norfloxacin treatment
We monitored, via flow cytometry, gfp expression from our DNA damage sensor, Fur sensor, and iron-sulfur cluster damage sensor as well as hydroxyl radical formation every hour over a 6-hour treatment with 250ng/mL of norfloxacin (Supplementary Figure   9). In all cases, we see an increase in DNA damage, Fur derepression, iron-sulfur cluster damage, and hydroxyl radical formation over the first 3-hours of norfloxacin treatment followed by stabilization of gfp expression and hydroxyl radical levels. The similar patterns of expression suggest a connection between iron homeostasis, iron-sulfur cluster stability, hydroxyl radical levels and DNA damage. The insets show untreated wildtype cultures followed with the sensor constructs and HPF dye, respectively, over the same time course. In all cases, we do not observe an increase in DNA damage, Fur derepression, iron-sulfur cluster damage, and hydroxyl radical levels as a function of growth alone.  Fig. 4a)and without CcdB.

Gyrase poisoning and hydroxyl radical formation
In the Fenton reaction, free ferrous iron reacts with hydrogen peroxide to generate highly destructive hydroxyl radicals (Imlay, 2003). In our model, production of hydroxyl radicals is the cytotoxic end-product of redox cycling following a gyrase inhibitorinduced superoxide burst and Fe-S cluster damage. To detect the generation of hydroxyl radicals following gyrase inhibition by norfloxacin or CcdB, we employed the fluorescent reporter dye, 3'-(p-hydroxyphenyl) fluorescein (HPF) (Setsukinai et al., 2003), which is oxidized by hydroxyl radicals with high specificity (Supplementary Figure 11).
As expected, we observed a significant increase in hydroxyl radical-induced fluorescence upon addition of norfloxacin to wildtype cells, the first direct demonstration of DNA gyrase inhibitor-induced hydroxyl radical generation. The largest detectable increase in fluorescence was observed in !sodB cells, while we did not detect hydroxyl radical production in !atpC cells. Additionally, !iscS cells expectedly exhibited a small increase in hydroxyl radical levels following norfloxacin treatment. We also observed decreased hydroxyl radical formation in !fur cells relative to wildtype, in line with the observed increase in !fur survival following application of norfloxacin. Interestingly,

Plasmid construction, cell strains, and reagents
Basic molecular biology techniques were implemented as previously described (Sambrook and Russell, 2001). All plasmids were constructed using restriction endonucleases and T4 DNA Ligase from New England Biolabs (NEB, Beverly, MA).
Our DNA damage and iron regulation sensors were based on the design of the P LlacO-1 promoter (Lutz and Bujard, 1997). PCR was used to build each promoter, which employed LexA and Fur operator sites to regulate expression, respectively, of the green fluorescent protein gene, gfpmut3b (Cormack et al., 1996). To construct our oxidative damage sensor, we PCR amplified the native soxS promoter from XL-10 cells and cloned it upstream of the gfpmut3b gene.

Phenotypic analysis
In our experiments, we compared the growth of untreated, norfloxacin treated (250ng/mL), CcdB-(uninduced cultures containing the ccdB riboregulator) and CcdB+ (induced cultures containing the ccdB riboregulator) wildtype BW25113 (lacI q rrnB T14 !lac@ WJ16 hsdR514 !araBAD AH33 !rhaBAD LD78 ; ) cultures. In our specific single-gene knockout experiments, we studied the growth of deletion strains contained in a BW25113 deletion library (Baba et al., 2006). To determine statistically significant changes in gene expression due to norfloxacin treatment, or CcdB expression, we subtracted the expression z-score of each gene in our uninduced control dataset from the corresponding z-score in our perturbed (norfloxacin-treated or CcdB-expressing) sample dataset. This was done for each experimental time-point (0, 30, 60, 120, 180 minutes), e.g., the z-score of recA expression from our uninduced sample at 30 minutes was subtracted from the recA zscore from our norfloxacin-treated sample at 30 minutes. This allowed us to determine the difference in expression between an uninduced control set and a gyrase-inhibitor treated data set in terms of units of standard deviation, a robust metric. A gene was considered to have significantly changed expression when its z-score difference was greater than two units of standard deviation, with the sign determining over-and underexpression.
Following the identification of significantly changed genes at each time-point, we performed functional enrichment using the gene ontology (GO) classification system found on EcoCyc (Keseler et al., 2005). In doing this, we were able to group genes by GO annotated pathways. To track the changes in the cellular transcriptional program over time, we utilized the transcription factor regulatory information contained in RegulonDB (version 4) (Salgado et al., 2006). Using both sets of information, we were able to categorize significantly changed genes in functional units.

Sensor construct experiments using the flow cytometer
To monitor the occurrence of DNA damage, oxidative damage to Fe-S clusters and changes in iron regulation we employed our respective engineered sensors which respond to these biochemical events by activating expression of gfpmut3b. All data were

Measurement of hydroxyl radicals using HPF dye
To detect hydroxyl radical formation following norfloxacin treatment or CcdB expression we used the fluorescent reporter dye, 3'-(p-hydroxyphenyl) fluorescein (HPF, Invitrogen), which is oxidized by hydroxyl radicals with high specificity. All data were collected using the Becton Dickinson FACSCalibur flow cytometer described above (Becton Dickinson). The following PMT voltage settings were used: E00 (FSC), 300 (SSC), and 825 (FL1). Calibrite Beads (Becton Dickinson) were used for instrument calibration. Flow data were collected, converted and analyzed as above.
In all experiments, cells were grown overnight, then were diluted 1:1,000 in 50mL LB (+30 µg/ml kanamycin for CcdB expressing cells) supplemented with 5µM HPF. CcdB expression was induced by addition of 1mM IPTG and 0.25% arabinose at an OD 600 of 0.3-0.4, while 250ng/mL norfloxacin was added to drug treated cultures.