Analysis of the Oxidative Stress Regulon Identifies soxS as a Genetic Target for Resistance Reversal in Multidrug-Resistant Klebsiella pneumoniae

ABSTRACT In bacteria, the defense system deployed to counter oxidative stress is orchestrated by three transcriptional factors, SoxS, SoxR, and OxyR. Although the regulon that these factors control is known in many bacteria, similar data are not available for Klebsiella pneumoniae. To address this data gap, oxidative stress was artificially induced in K. pneumoniae MGH78578 using paraquat and the corresponding oxidative stress regulon recorded using transcriptome sequencing (RNA-seq). The soxS gene was significantly induced during oxidative stress, and a knockout mutant was constructed to explore its functionality. The wild type and mutant were grown in the presence of paraquat and subjected to RNA-seq to elucidate the soxS regulon in K. pneumoniae MGH78578. Genes that are commonly regulated both in the oxidative stress and soxS regulons were identified and denoted as the oxidative SoxS regulon; these included a group of genes specifically regulated by SoxS. Efflux pump-encoding genes and global regulators were identified as part of this regulon. Consequently, the isogenic soxS mutant was found to exhibit a reduction in the minimum bactericidal concentration against tetracycline compared to that of the wild type. Impaired efflux activity, allowing tetracycline to be accumulated in the cytoplasm to bactericidal levels, was further evaluated using a tetraphenylphosphonium (TPP+) accumulation assay. The soxS mutant was also susceptible to tetracycline in vivo in a zebrafish embryo model. We conclude that the soxS gene could be considered a genetic target against which an inhibitor could be developed and used in combinatorial therapy to combat infections associated with multidrug-resistant K. pneumoniae.


Introduction
Oxygen started accumulating in the biosphere about 2-3 billion years ago. Many organisms harvest energy by oxidizing organic compounds with oxygen acting as the terminal electron acceptor. This molecule, therefore, has become essential for life, at least for aerobic organisms. As a natural consequence of aerobic metabolism, the production of toxic reactive oxygen species (ROS) namely, hydrogen peroxide (H2O2), superoxide radical (O2 •-), and the generation of hydroxyl radical (HO • ), becomes inevitable in an oxygen-rich environment. Different ROS will not only oxidize macromolecules (such as DNA, proteins, lipids) but also extract iron from proteins containing an ironsulfur cluster creating a highly reactive HO • rich intracellular environment 1,2 , detrimental for bacteria. Therefore, to survive the effects of ROS, bacteria deploy a variety of adaptive responses. These are well characterized in bacteria like Escherichia coli 3

, but as yet, not in Klebsiella pneumoniae
In bacteria, the primary antioxidant defence systems employ superoxide dismutase (SOD) and catalase (CAT) enzymes 1,4 . However, these systems may prove inadequate to protect bacteria under circumstances of extreme and prolonged oxidative stress. During these stress conditions, bacteria can activate the OxyR and SoxRS systems in response to hydrogen peroxide 5 and redox-active compounds 6 , respectively. Both the OxyR and SoxRS work by transcriptionally activating genes whose protein products either protect or repair the damage caused by intracellular ROS accumulation. In the SoxRS system, the activation of a target gene occurs via a two-step process wherein SoxR acts as a sensory SoxS protein is a transcriptional activator belonging to the XylS/AraC family 7 . SoxR dependent induction of SoxS, in turn, activates the transcription of many other genes (denoted collectively as the SoxRS regulon) whose primary functions involve anti-oxidative action, detoxification, efflux of redoxactive compounds, changes in membrane permeability and protecting DNA thereby rescuing bacteria from the deleterious effects of increased intracellular levels of ROS [9][10][11] . In E. coli genes that were regulated by SoxS were identified [12][13][14][15][16][17] . Overall, the biological role of the SoxRS regulon can be summarized as (1) prevention of oxidative damage (2) recycling of damaged macromolecules and (3) regeneration of nicotinamide adenine dinucleotide phosphate.
Though the transcriptional organization of soxS is well characterized in some pathogens, these data are lacking for K. pneumoniae. These bacteria are a member of the ESKAPE group, one of six pathogens responsible for most drug-resistant nosocomial infections 18 . Since oxidative stress is known to mediate antibiotic resistance in pathogens we were interested in identifying how K. pneumoniae responds to oxidative stress and what its impact might be on antimicrobial resistance. In this study, RNA-seq was used to describe the transcriptional architecture of K. pneumoniae MGH78578 during exposure to a ROS inducing agent, paraquat, revealing that the regulon was controlled by the soxRS two-component system. In addition, RNA-seq analysis of the isogenic mutant K. pneumoniae MGH 78578ΔsoxS was carried out and used to describe the 'oxidative soxS regulon' -a stringent set of genes regulated via soxS. K. pneumoniae MGH 78578ΔsoxS was highly susceptible to tetracycline.
Susceptibility of the mutant to tetracycline coupled with increased accumulation of tetraphenylphosphonium (TPP + ) in the bacterial cytoplasm was supported at least in part by the downregulation of acrAB-tolC and the global regulator marRAB in K. pneumoniae MGH 78578ΔsoxS. Since the mutant was highly avirulent in a zebrafish model, we predict that soxS can be used as a genetic target to inhibit infections associated with MDR K. pneumoniae.

Results and discussion
SoxS is the major transcriptional regulator when K. pneumoniae MGH78578 is exposed to redox compound-based oxidative stress. Experimentally, oxidative stress can be induced in bacteria by exposing cultures to either redox compounds like PQ (paraquat) or H2O2. PQ is 1,1-dimethyl-4,4bipyridinium and is a widely used nonselective herbicide, found to induce oxidative stress by enhancing ROS levels -superoxide anion radical ( • O2 -) in a dose-dependent manner, as exemplified in Vibrio cholera, E. coli and others 14,19 . Firstly, we started by assessing the inhibitory concentration of PQ in K.
pneumoniae MGH78578 using broth microdilution and determined the MIC to be 15.62 µM. Thereafter, the following transcriptomic experiments were carried out at sub-minimal inhibitory concentrations (half the MIC). Here, we used RNA-seq to investigate the genome-wide transcriptional architecture of multi-drug resistant K. pneumoniae MGH78578 following exposure to a sub-inhibitory concentration of PQ. K. pneumoniae MGH78578 was exposed to 7.8 µM PQ for 30 minutes to induce oxidative stress. Approximately 57 million uniquely mapped reads were generated across all six libraries accounting for more than 9 million reads/library (Table S1-WS1), data that was sufficient for robust transcriptional analysis 20 . The expression levels of 5,185 K. pneumoniae MGH78578 chromosomal genes and the resident plasmid encoding genes (including plasmids pKPN3, pKPN4, pKPN5, pKPN6, and pKPN7) were calculated using the Voom approach (limma package) 21 . We confirmed the reproducibility of the RNA-seq data by calculating the Spearman coefficients for the biological replicates of all libraries based on the normalized read counts. In all six libraries, the coefficient was found to be ~ 0.96 to 0.99, confirming the statistical significance between replicates (Fig. S1).
Here, we describe the 'oxidative stress regulon' of K. pneumoniae MGH78578 by identifying the genes that were differentially regulated in MGHPQ versus MGHwt libraries. The oxidative stress regulon comprised 1,366 genes which were differentially regulated (Table S1-WS2 and Fig. 1A). Of these 11.5% (n=158) were highly up-regulated (>4-fold) and 22.5% (n=309) were up-regulated (2 to 4fold). A total of 49 genes (3.7%) were highly down-regulated (>4-fold) while a further 147 (11.12%) were down-regulated (2 to 4-fold) (Table S1-WS2). Upon analysis, the most induced K. pneumoniae MGH78578 gene was found to be soxS (145-fold) indicating that the soxRS regulon was highly active in PQ exposed K. pneumoniae MGH78578. The transcriptomic response of bacteria to oxidative stress is specific to the agent causing oxidative stress -extracellular H2O2 triggers OxyR regulon while PQ induces the SoxRS regulon, as exemplified in Escherichia coli 3 . Based on these E. coli data, we hypothesized that exposure to PQ should induce the SoxS regulon in K. pneumoniae MGH78578. Since no data was available in K. pneumoniae, we put our hypothesis to the test using RT-qPCR targeting the soxS gene. Our RT-qPCR data confirmed that the expression of the soxS transcript improved with increasing concentration of PQ ( Fig. 2A).
Exposure to H2O2, however, generated a different response in other bacteria. Exposure of V. cholerae to oxidative stress increased the activity of SOD and CAT enzymes 19 . However, in V.
cholerae, the level of CAT did not increase post-exposure to PQ but rather increased during exposure to H2O2. Our results describing PQ exposed K. pneumoniae MGH78578 support this observationnone of the catalases (encoded by genes KPN_RS06170, KPN_RS06615, and KPN_RS09805) were differentially regulated (Table S1-WS2). However, the SOD (encoded by sodA, sodB, and sodC) was highly up-regulated -sodA alone was highly up-regulated (14-fold) while sodC was up-regulated (⁓ 3fold) in PQ induced cells. We did not find sodB to be differentially regulated within PQ treated K. pneumoniae MGH78578. It is tempting to speculate that selective differential regulation of SOD and not CAT in K. pneumoniae MGH78578 could be the response to • O2induced by PQ.
pneumoniae MGH78578 soxS regulon with the E. coli soxS regulon published earlier 3 . Of the 59 soxS regulated in E. coli K-12 genes, 44 were also found to be similarly regulated by soxS in K. pneumoniae MGH78578 (Fig. 1C).
To add stringency to our data, we further compared the oxidative stress regulon to the soxS regulon to identify those genes that belonged to the 'oxidative soxS' regulon. The oxidative soxS regulon represented a stringent set of K. pneumoniae MGH78578 genes that were regulated by soxS alone. The genes belonging to this regulon had a characteristic statistically significant expression pattern -upregulated in the MGHPQ and down-regulated in MGHΔsoxSPQ (i.e. SoxS induced); down-regulated in MGHPQ and up-regulated in MGHΔsoxSPQ (i.e. SoxS repressed). In total, 256 genes belonged to the 'oxidative soxS' regulon. Of these 222 genes were found to be 'SoxS induced' while 34 were 'SoxS repressed' (Table S1-WS3). Examples include soxS, acrAB, tolC among others, all of which were soxS induced. Of the 44 genes commonly identified in our soxS regulon and E. coli K-12 3 , 30 were identified to belong to the more stringent 'oxidative soxS' regulon. Our 'oxidative soxS regulon' identified many genes that were previously shown to be regulated by SoxS. A discussion of these genes is included in the supplementary file S1.
K. pneumoniae MGH78578 is a multi-drug resistant isolate and its antimicrobial resistance profile is well characterized 22 . Our primary interest was in identifying how soxS modulates antimicrobial resistance in K. pneumoniae MGH78578 . So we assayed whether any K. pneumoniae MGH78578 genes conferring antimicrobial resistance were captured in our 'oxidative soxS' regulon.
We identified 11 antimicrobial resistance-encoding genes (acrAB, acrE, tolC, marRAB, cmr (mdfA), ybhT, KPN_RS15915, and KPN_RS15920) in the 'oxidative soxS' regulon and all of them were soxS induced (Table S1-WS3). Interestingly we also found that another member of the XylS/AraC family, tetD, that was absent from the 'oxidative regulon' but present in the 'soxS' regulon indicating that, at least in K. pneumoniae, tetD is positively regulated by soxS. Though tetD was shown to modulate response against redox compounds and tetracycline 23 , we did not find any evidence of differential regulation when K. pneumoniae MGH78578 was exposed to paraquat. Since many genes conferring antimicrobial resistance were modulated by soxS, we were interested in examining whether inactivation of soxS resulted in aberrations in the antimicrobial resistance pattern of K. pneumoniae MGH78578. This analysis also identified 50 conditions considered to be down-regulated, in which the mutant showed reduced metabolic respiration compared with the WT. These were found to be associated with high pH (9,5) sensitivity, and having an impact on cell wall and protein synthesis due to susceptibility to antimicrobial drugs such as tetracyclines (doxycycline, demeclocycline, chlortetracycline, minocycline) aminoglycosides (amikacin), cephalosporines (cephalothin, cefuroxime, cefotaxime), βlactams (cloxacillin, oxacillin, phenethicillin) and others such as polymyxin B and colistin (polymyxin E).
Our RNA-seq data showed that the genes encoding antimicrobial resistance such as acrAB-tolC, marRAB, and others, were differentially regulated in K. pneumoniae MGH 78578ΔsoxS and, thus, classified as soxS induced. This observation, and the phenotypic microarray associated metabolic profiling, led us to hypothesize that the soxS mutant might have a modified antimicrobial resistance profile compared to the wild type. To test our hypothesis, we assayed the MBC of both K. pneumoniae MGH 78578 and K. pneumoniae MGH 78578ΔsoxS against a panel of antimicrobial compounds. MBC assays were carried out on K. pneumoniae MGH 78578 and K. pneumoniae MGH 78578ΔsoxS against colistin, gentamycin, kanamycin, cefotaxime, and tetracycline (Fig. 3). Escherichia coli ATCC ® 29544 was used as a control. Our results showed that there was no change in the MBC values against colistin, kanamycin, and rifampicin, even though our phenotypic microarray assay recorded a down-regulation in the metabolism of the mutant compared to the wild type. It could be that the metabolic downregulation was not sufficient to cause an inhibitory effect. However, there was a significant reduction in the MBC values for K. pneumoniae MGH78578ΔsoxS compared to K. pneumoniae MGH78578 when exposed to tetracycline and cefotaxime. Therefore using a combination of phenotypic microarray and RNA-seq we show that tetracycline tolerance was soxS dependent in MDR K. pneumoniae MGH78578.
Oxidative stress is a common cause of cell death mediated by antimicrobial agents, irrespective of the class to which the compound belongs 24 . So we were interested to know whether exposure to tetracycline induced any oxidative stress in K. pneumoniae MGH78578. For this, we checked the induction of soxS in tetracycline exposed K. pneumoniae MGH78578. Proportional induction of soxS expression in response to increasing tetracycline concentration confirmed that the exposure to tetracycline induced soxS dependent oxidative stress in K. pneumoniae MGH78578 (Fig. 2B).
The soxRS associated regulation of antibiotic resistance was described earlier in several bacteria 25,26 . Similarly, the induction of ROS was also reported to modulate antibiotic resistance in other pathogenic bacteria. For example, Salmonella Typhimurium was shown to modulate its susceptibility to tetracycline when exposed to a ROS generating macrolide antibiotic, tylosin 27 . Also, in Acinetobacter baumannii, soxR overexpression led to susceptibility to tetracycline 28 . This SoxR based negative regulation of SoxS could be the reason underpinning the increased susceptibility. Even though the correlation between the expression of soxS and efflux pumps has been shown previously 29 , there is no evidence pointing to the cytoplasmic accumulation of antimicrobial compounds due to an inactive soxS based impaired efflux activity. We, therefore, proceeded to determine whether an impaired efflux activity led to the accumulation of compounds within the cytoplasm of K. pneumoniae MGH78578 ΔsoxS, leading to the bactericidal effect.

Reduction in MBC is due to the impaired efflux pump activity in K. pneumoniae MGH 78578
ΔsoxS cells. Since the K. pneumoniae, MGH78578ΔsoxS was susceptible to tetracycline, we were interested in understanding the mechanism underpinning the observation. Our RNA-seq data revealed that the genes encoding the AcrAB-TolC efflux pump were highly SoxS dependent because they were >4 fold up-regulated in the PQ regulon and >8 fold down-regulated in K. pneumoniae MGH78578ΔsoxS. Tetracycline is one of several structurally diverse substrates of the efflux pump AcrAB-TolC 30 . Hence we hypothesized that the deletion of the soxS gene could lead to a reduction in the expression of the AcrAB-TolC efflux pump. This feature could then account for the accumulation of tetracycline in the cytoplasm to bactericidal levels.
To test our hypothesis, we assayed the efflux activity of wild type K. pneumoniae MGH78578 and K. pneumoniae MGH78578ΔsoxS by measuring the accumulation of tetraphenylphosphonium (TPP + ) ions using previously described protocols 31 . We first tested whether K. pneumoniae MGH 78578ΔsoxS had an intact outer membrane. In this case, both wild type K. pneumoniae MGH78578 and K. pneumoniae MGH78578 ΔsoxS were first exposed to low concentrations of polymyxin B (PMB), an antibiotic that causes outer membrane destabilization, and then assayed the accumulation of TPP + . Our results showed that K. pneumoniae MGH78578ΔsoxS were more susceptible to PMB and a concentration of 6 g/ml was sufficient to induce the depolarization of the plasma membrane. In comparison, for the wild type K. pneumoniae MGH78578, a concentration of PMB of 9 g/ml was required. Nonetheless, alterations in membrane voltage (maximum amount of TPP + ) were similar for both wild type K. pneumoniae MGH78578 and the isogenic K. pneumoniae MGH78578ΔsoxS mutant showing that neither the outer nor the inner plasma membranes were compromised in K. pneumoniae MGH78578ΔsoxS (Fig. 4A). This finding was supported by our earlier RNA-seq data which showed that membrane-associated genes that were differentially regulated during 1-(1-naphthylmethyl)piperazine (NMP) (a chemosensitizer) treatment 32 were not differentially regulated in the soxS regulon.
Next, we investigated whether the efflux pump activity was compromised in K. pneumoniae MGH78578ΔsoxS compared to that of wild type K. pneumoniae MGH78578. The aim was to confirm/refute our hypothesis that the impaired pump activity could result in the accumulation of tetracycline within K. pneumoniae MGH78578ΔsoxS. We previously established that the treatment of K. pneumoniae MGH78578 with NMP destabilized the bacterial outer membrane before efflux pump inhibition and that this phenotype was concentration-dependent 32 . Hence we used different concentrations of NMP to test the efflux pump inhibition of K. pneumoniae MGH78578ΔsoxS cells compared to that of K. pneumoniae MGH78578. Initially, we treated wild type K. pneumoniae MGH78578 with NMP and assayed the cells for TPP + accumulation. As expected, in wild type K. pneumoniae MGH78578, NMP impaired efflux pump activity and induced cytoplasmic TPP + accumulation at a concenration of 30 µg/mL. However, for K. pneumoniae MGH78578ΔsoxS, 15µg/mL NMP was sufficient to inhibit the efflux pump activity and cause TPP + accumulation (Fig. 4B). The increased sensitivity of K. pneumoniae MGH78578ΔsoxS to NMP was observed also at 120 µg/mL wherein this agent increased the accumulation of TPP + in K. pneumoniae MGH78578 but induced a partial depolarization of the plasma membrane and leakage of the accumulated cation in K. pneumoniae MGH78578ΔsoxS. These results show that the efflux pump activity was impaired in the mutant.
To further confirm the effect of the outer membrane destabilization on TPP + accumulation, we pre-treated both K. pneumoniaeMGH 78578 and K. pneumoniae MGH78578ΔsoxS first with PMB to permeabilize the outer membrane and then re-tested for NMP mediated TPP + accumulation. These data indicated that TPP + was accumulated at 15 µg/mL for K. pneumoniae MGH78578ΔsoxS either indicating that both efflux pump inhibition and outer membrane destabilization could cause TPP + accumulation in the bacterial cytoplasm (Fig. 4C). The respiration activity of K. pneumoniae use oxidative stress to control bacterial infections in zebrafish larvae. We observed that K. pneumoniae MGH78578 ΔsoxS was inefficient in killing zebrafish larvae compared to the wild type (Fig. 5). At 1 dpi (days post-infection), a survival rate of 100% was recorded in embryos injected with wild type K. pneumoniae MGH78578, K. pneumoniae MGH 78578 ΔsoxS, E. coli, and DPBS. However, at 2 dpi, the survival rate of the embryos injected with wild type K. pneumoniae MGH78578 decreased to 80% while the survival rate in those embryos injected with K. pneumoniae MGH78578 ΔsoxS and in the avirulent/un-inoculated controls remained unaltered. At 3 dpi, the survival rate dropped further down to 50% for the embryos injected with K. pneumoniae MGH78578 while being maintained at 90% for the K. pneumoniae MGH78578 ΔsoxS infected embryos (Fig. 5). The survival rate remained unaltered in the case of E. coli and DPBS injected embryos over the time course of infection. Recent studies report that high neutrophil recruitment and zebrafish lethality is observed with K. pneumoniae if directly injected into the blood 34,36 . We anticipate two possibilities for the sensitivity of K. pneumoniae MGH78578 ΔsoxS in zebrafish larvae. First, in vertebrates, extracellular bactericidal action is initiated by neutrophils at a distance by activating an NADPH oxidase-dependent production of superoxide 37 .
The avirulent phenotype of K. pneumoniae MGH78578 ΔsoxS could be due to the inefficiency in combating the extracellularly produced, neutrophil originated superoxide in the blood. Secondly, K.
pneumoniae MGH78578 ΔsoxS exhibited a down-regulated expression of acrAB-tolC which could result in an avirulent phenotype as seen previously in Salmonella Typhimurium 38 .
By impairing soxS, multi-drug resistant K. pneumoniae MGH78578 infections can be treated by tetracycline. Currently, various strategies are being investigated to mitigate the threat of AMR in bacteria 39 . It was thought that restricting the use of a particular antibiotic would restore susceptibility to that compound over time by eliminating the selective advantage. But it has been observed that AMR is persistent over the decades 40 . Recent research has elaborated on the possibility wherein resistance can be reversed. One strategy here made use of defined drug-adjuvant combinations to reverse resistance so that conventional antibiotics continue to be effective 41 . With this broad goal in mind, we endeavored to identify genetic targets that regulate resistance and develop strategies to reverse resistance by inhibiting them. Our transcriptomic and phenotypic data has shown that by inhibiting soxS, susceptibility to tetracycline can be restored in a multi-drug resistant K. pneumoniae. We were further interested to investigate whether soxS mediated tetracycline susceptibility can be demonstrated in an in vivo zebrafish model.
For this, we treated 4 hpf zebrafish embryos with increasing concentrations of tetracycline. At 48 hpf, K. pneumoniae MGH78578, K. pneumoniae MGH78578 ΔsoxS, and DPBS were microinjected into the blood circulation. Post-injection, zebrafish larvae were collected at different time points and again bacterial counts were enumerated (Fig. 6). It was shown recently that exposure to tetracycline induced ROS production in zebrafish larvae 42 . So, we hypothesized that K. pneumoniae MGH78578 ΔsoxS will be impaired in its ability to survive in tetracycline treated zebrafish larvae due to increased sensitivity to either ROS production in tetracycline treated larvae or tetracycline alone. It is also possible that healthy zebrafish larvae could clear K. pneumoniae MGH78578 ΔsoxS from the system due to normal exposure to peroxides synthesized from neutrophils. Confirming our hypothesis, K. pneumoniae MGH78578 ΔsoxS was completely cleared from the tetracycline treated zebrafish larvae in 24 hours. However, K. pneumoniae MGH78578 ΔsoxS was cleared even in tetracycline untreated zebrafish larvae confirming that the selective advantage was lost in the bacterial mutant making it susceptible to the immune system of zebrafish larvae (Fig. 6). It should also be noted that the clearance was much more marked in tetracycline treated larvae suggesting that K. pneumoniae MGH78578 ΔsoxS was cleared from the system due to a cumulative effect of both immune system and tetracycline induced ROS production. Overall, we show that soxS can be used as a genetic target to treat multi-drug resistant K. pneumoniae infections.

Conclusion.
Apart from elucidating the PQ oxidative stress regulon and the oxidative SoxS regulon, we propose that a combination of tetracycline and a SoxS inhibitor can be used to treat infections associated with MDR K pneumoniae. Tetracycline will induce oxidative stress in the host and the SoxS inhibitor can impair the ability of the pathogen to survive oxidative stress. The advantage of this approach is that it is independent of any particular bacterial resistance mechanism, can be used against strains with any resistance profile. Certainly, the next step in this approach is to construct/identify a molecule that will specifically and post-transcriptionally inhibit the synthesis of soxS mRNA or posttranslationally inhibit the SoxS protein. It will be worthwhile to test if our observations will also apply to other MDR pathogens like Salmonella Typhimurium, E. coli among others. Nevertheless, our results address the immediate concern of antimicrobial resistance in pathogens of importance to human health whilst providing a 'proof of concept' for an approach that requires further experimental investigation to achieve the therapeutic objective.

Bacterial strain
MDR Klebsiella pneumoniae MGH78578 (ATCC ® 700721) was isolated from a sputum sample in 1994 and was purchased from the American Type Culture Collection. This strain was selected mainly because it is a multi-drug resistant type strain 43 and its drug resistance profile was recently published 22 .
Moreover, the efflux pumps present in this strain are well characterized 44,45 . Further, the whole genome sequence of this strain is available in NCBI (Reference Sequence NC_009648.1) which was convenient for mapping RNA-seq data. This bacterium was grown in Müeller-Hinton (MHB) broth and Müeller-Hinton agar (MHA) (Sigma, Dublin, Ireland).

Phenotypic assay (OmniLog)
The comparison of K. pneumoniae MGH78578 WT with the ΔsoxS mutant was evaluated using the OmniLog (Biolog, Inc., Hayward, CA) phenotypic microarray. Microplates PM1 through PM20 with the exception of PM5 were used. These plates contain several carbon, nitrogen, sulphur and phosphorous-substrates, ions, osmolytes and chemicals at different concentrations and pH 46

Isolation of RNA from oxidatively stressed bacterial cells
Before RNA isolation, wild type Klebsiella pneumoniae MGH78578 and K. pneumoniae MGH78578 ΔsoxS were grown until mid-exponential state (MEP) following an earlier standardized protocol 22 .
MEP grown bacterial cells were treated with paraquat (7.81 µM) for 30 minutes to generate oxidative stress conditions. RNA was then extracted from both oxidative stressed and MEP grown (control) cells using QIAGEN RNeasy Mini Kit following manufacturers guidelines. Contaminating DNA was removed from the RNA sample using the Turbo DNase I kit (Thermo Fischer Scientific). RNA was then quantified using both Qubit RNA Broad Range Assay and the Nanodrop.

Sequencing RNA isolated from oxidatively stressed bacterial cells
The library preparation and subsequent sequencing were carried out commercially at the Center For An average of 1.48 Gbp of raw sequence data was obtained per sample, in 125 bp single-end reads.

Mapping of sequenced reads
The sequence quality of the RNA-seq reads was analyzed using FastQC [https://www.bioinformatics.babraham.ac.uk/projects/fastqc/]. Sequence reads were aligned and mapped against the reference genome of K. pneumoniae MGH78578 (Reference Sequence NC_009648.1) using Segemehl with default mapping parameters 22,48 and uniquely mapped reads were used and considered for the differential gene expression computational analysis. Read counts (number of reads that aligned to a specific gene) for each gene were quantified using custom Perl scripts.

Computational analysis of RNA-seq data
Computational analysis of RNA-seq data was performed using R (version 3.5.2, https://www.rproject.org/). To calculate the expression level of genes, the raw read counts were normalized using the VOOM function 21 in the limma package 49 . More specifically, counts were converted to log2 counts per million (log2 CPM), quantile normalized, and precision weighted using the VOOM function. A linear model was then fitted to each gene, empirical Bayes moderated t-statistics and its corresponding pvalues were used to assess differences in expression 21,50 . To account for multiple comparisons, Benjamini-Hochberg corrected p-values were computed. As reads for duplicated coding genes (paralogs) or duplicated small RNAs cannot be mapped unequivocally, these genes appear in the analysis as unmapped. The sequence reads can be visualized in the Integrated Genome Browser (version 9.0.0) 51 . The read depth was adjusted with the cDNA library with the lowest number of reads 52 . RNA sequencing data were analyzed using the fold change parameters as follows: highly up-regulated (>4), up-regulated (2 to 4 fold), no change in expression (0.5 to 2 fold), down-regulated (0.25 to 0.5 fold) and highly down-regulated (less than 0.25 fold).

Construction of the K. pneumoniae MGH78578 ΔsoxS mutant
A modified λ-Red system was used to construct an in-frame deletion in multi-drug resistant K. pneumoniae MGH78578 53 . Here three plasmids are employed. The first, plasmid pIJ773 which serves as a template to amplify the apramycin resistance gene, aac(3)IV, and flanking FRT sites. The second plasmid, pACBSR-Hyg contains the λ-Red system comprising beta, gam, and exo genes which are under the control of an arabinose-inducible promoter and facilitate homologous recombination between the knockout cassette and the target locus in the chromosome. The third plasmid, pFLP-Hyg, contains the FLP recombinase, which was used to excise the apramycin selection marker from the chromosome via the FRT sites. The antibiotic apramycin was used to select plasmid pIJ773, while hygromycin was used to select both plasmids pACBSR-Hyg and pFLP-Hyg. K. pneumoniae MGH78578 was susceptible to both apramycin and hygromycin.  Table S1 WS-1.

Isolation of RNA for Quantitative Reverse Transcriptase Polymerase Chain Reaction (qRT-PCR)
Wild type K. pneumoniae MGH78578 was grown to Mid Exponential Phase (MEP) at 37ºC in Müeller-Hinton Broth, cells were then treated with paraquat at different concentrations (0, 3.905, 7.81, 200 and extracted. All assays were run in triplicate. In all conditions, RNA was extracted using QIAGEN RNeasy Mini Kit following manufacturers guidelines. Any contaminating DNA was removed from the RNA sample using the Turbo DNase I kit (Thermo Fischer Scientific). Purified RNA was subsequently quantified using both Qubit RNA Broad Range Assay and Nanodrop.

Two steps Quantitative Reverse Transcriptase Polymerase Chain Reaction (qRT-PCR)
The reverse transcriptase reaction was carried out on RNA purified from K. pneumoniae MGH78578 under the conditions mentioned earlier, using the high capacity RNA to cDNA kit (Sigma) following manufacturer guidelines. A negative control devoid of RT enzyme was also carried included. qPCR was then performed following the prime-time gene expression master mix protocol (IDT, Leuven). The expression of the housekeeping gene rpoB was used to characterize the relative expression of the gene of interest, soxS. The sequences of all primers used in the experiment is provided in Table S1 WS-1.

Determination of Minimum Bactericidal Concentration (MBC)
Previously guidelines. Triplicate MBC values for each antibiotic tested were determined using MHB broth in a 96well microtiter plate. A steel inoculator was employed to transfer inoculum from the above 96-well plate as described above to a fresh 96-well plate containing MHB without any of the antibiotics to be tested. These plates were then incubated at 37˚C for 16-18 hours, following which the MBC values were recorded.

Electrochemical measurement experiments
The efflux activity of K. pneumoniae MGH78578 and K. pneumoniae MGH78578 ΔsoxS cells was assayed measuring accumulation of tetraphenylphosphonium (TPP + ) ions. Over-night cultures of K.
pneumoniae were grown in Luria-Bertani broth, containing 0.5 % NaCl, diluted 3:100 in fresh medium, and the incubation was continued until the OD600 nm reached 1.0. The cells were collected by centrifugation at 4 ºC for 10 min at 3000 g. The pelleted cells were re-suspended in 100 mM sodium phosphate buffer, pH 8, to obtain 1,4 x 1011 cfu/ml. Concentrated cell suspensions were kept on ice until used, but not longer than 3 h.
Changes of TPP + concentration in the suspensions of thermostated and magnetically stirred cells were monitored using TPP+-selective electrodes as previously described 31 54 . Experiments were performed at 37 ºC in 100 mM sodium phosphate buffer, pH 8, containing 0.1 % glucose. OD612 nm of the cell suspension during measurements was 1.

Data availability
All the RNA sequence data generated in the study are deposited in the National Center for Biotechnological Information -Gene Expression Omnibus and are available under the accession number GSE146844. Post analysis, the differential expression of all the genes are given in supplementary Table S1, with three worksheets WS-1, WS-2 and WS-3.
Table S1-WS1: RNA-seq mapping details. Details of the RNA-seq reads mapped against different regions of the K. pneumoniae MGH78578 genome is given here. Is also given here a list of all primers used in the experiments in this paper.  Table S1-WS3: The 'oxidative soxS regulon' of Klebsiella pneumoniae MGH78578 -a stringent set of genes that are regulated by SoxS. The gene list was obtained from statistically significant genes from oxidative regulon and soxS regulon with a distinct expression pattern. Genes those were up-regulated in the oxidative regulon + down-regulated in soxS regulon and down-regulated in oxidative regulon + up-regulated in soxS regulon.

Code availability
All codes used are published programs, with citations for each provided in the references.