The multiple antibiotic resistance operon of enteric bacteria controls DNA repair and outer membrane integrity

1The Escherichia coli multiple antibiotic resistance (mar) locus was identified as a determinant for cross-resistance to tetracyclines, quinolones and β-lactams. Widespread among enteric bacteria, the system confers clinically relevant antimicrobial resistance in E. coli –. The mar phenotype results from induction of an operon designated marRAB. In wild-type cells, marRAB expression is stochastic and enhanced by many antimicrobial compounds,. Conversely, in some clinical isolates, marRAB expression is constitutive; mutations prevent auto-repression by MarR,. Induction of marRAB results in expression of MarA, a transcription factor that controls multidrug efflux and porin production–. Like all members of the AraC-XylS family, MarA uses a dual helix-turn-helix motif to bind non-palindromic DNA targets. This DNA sequence, called the ‘marbox’, is conserved in Gram-negative bacteria,. Proteins related to MarA, such as SoxS, Rob and RamA, also recognise the marbox and have overlapping regulatory effects–. However, interplay between these factors is complex. For example, Rob binds DNA with low specificity, but high affinity, and is usually sequestered in a non-functional state,. Conversely, MarA and SoxS have a lower affinity for the marbox but form ternary complexes with RNA polymerase that enhance recognition of marbox-containing promoters. It is likely that Rob behaves differently to other marbox binding proteins because Rob has a multimerisation domain and can interact with DNA using only one of its two helix-turn-helix motifs–.

2Models that depend on drug efflux and reduced porin production cannot explain all mar phenotypes,. For example, substantial resistance to minocycline is retained in ΔmarR cells lacking tolC or acrAB. Such observations have stimulated attempts to define the complete MarA regulon. Transcriptome analyses identified hundreds of genes putatively controlled by MarA,. However, only three genes were common to independent studies. Similarly, while 10,000 copies of the marbox occupy the E. coli genome, most are non-functional,. Consequently, only a handful of experimentally confirmed MarA binding sites are listed in the Ecocyc database (Supplementary Table 1).

3In this work, we have mapped genome-wide DNA binding by both MarR and MarA using chromatin immunoprecipitation and DNA sequencing (ChIP-seq). We show that the mar regulon is extensive and encompasses both DNA repair and lipid trafficking systems. These regulatory events allow the mar locus to combat quinolone-induced DNA damage and penetration of the outer membrane by antimicrobial compounds. Although previously unrecognised, these pathways reside within a ‘core’ mar regulon shared by many enteric bacteria.

4We used ChIP-seq to map global DNA binding by the marRAB-encoded transcription factors and the RNA polymerase σ70 subunit. Experiments were done using the enterotoxigenic E. coli strain H10407. The strain shares 3766 genes with E. coli K-12 and has 599 additional genes encoded by 25 discrete chromosomal loci and 4 plasmids (p948, p666, p58 and p52). We expected to identify a primary mar regulon, shared by most E. coli species, and additional targets specific to toxigenic strains. The MarA, MarR and σ70 binding profiles of are shown in Fig. 1a. In each plot, genes are illustrated by blue lines (tracks 1 and 2), MarA binding is in green (track 3), σ70 binding is in orange (track 4) and MarR binding is in black (track 5). As expected, MarR bound only to the marRAB promoter while MarA and σ70 bound at many loci. We used MEME to identify sequence motifs associated with the 33 MarA binding peaks. Only one statistically significant (E-value < 1e−12) motif was found (Fig. 1b, top panel). The motif closely resembles known MarA binding sites (Fig. 1b, compare top and bottom panel). For all peaks we determined the distance to the nearest start codon and sorted these distances into 100 bp bins. The distribution of peaks among the bins is illustrated in Fig. 1c; MarA and σ70 most frequently bind the 100 bp preceding the 5′ end of a gene. Of the 33 MarA binding peaks, 15 were within 150 bp of a binding peak for σ70 (Fig. 1c, inset). To support our ChIP-seq analysis, we tested binding of purified MarA to DNA fragments overlapping all 33 peaks. As a control, we also tested five DNA fragments from elsewhere in the genome. All but one of the DNA sequences derived from MarA ChIP-seq peaks bound MarA in vitro (Supplementary Fig. 1a). However, there was no binding to any of the control sequences (Supplementary Fig. 1b). The ChIP-seq and in vitro DNA binding data are summarised in Table 1. Five MarA targets were specific to E. coli H10407. Of these targets, one did not bind MarA in vitro, four were within prophage remnants, and only one was near to the 5′ end of a gene. Note that our list of MarA targets will not include sites occluded by other proteins in vivo. For example, most known marboxes preferentially bind SoxS or Rob; we did not expect to isolate these loci (Supplementary Table 1). Consistent with this, our ChIP-seq identified only the known marboxes with a high affinity for MarA (Supplementary Table 1).

5We focused on the 28 MarA binding sites shared with E. coli K-12. Implications for antibiotic resistance were assessed using phenotypic profiling data (Fig. 1d; Supplementary Fig. 2). In the heatmap, rows represent E. coli strains lacking individual MarA-targeted genes. Columns indicate treatment with different antibiotics. Where columns and rows intersect boxes are coloured according to growth relative to wild-type cells. Hence, red boxes identify MarA targets required for innate resistance to the corresponding antibiotic. Of particular note are xseA and the mlaFEDCB operon. These encode the large subunit of exonuclease VII and a lipid trafficking ABC transport system, respectively–. The data indicate that xseA is a determinant for quinolone tolerance. Conversely, mlaFEDCB mediates sensitivity to several antibiotics including tetracyclines (Fig. 1d).

6Figure 2a shows co-binding of MarA and RNA polymerase to the xseA locus. The sequence of the regulatory region is in Fig. 2b. The putative marbox (green) is immediately adjacent to the xseA promoter (PxseA; orange). To confirm binding at the marbox, we used electrophoretic mobility shift assays (EMSA) and DNAseI footprinting. For the EMSA experiments we used the xseA1 and xseA2 DNA fragments; the 5′ end of each fragment is denoted by an inverted triangle in Fig. 2b. The −36C mutation, present only in xseA2, ensures marbox inactivation. The EMSA experiment is shown in Fig. 2c. As predicted, MarA bound to the xseA1 fragment (top panel) but not the xseA2 fragment (bottom panel). In DNAseI footprinting, MarA protects the marbox from digestion and induces hypersensitivity at adjacent sites (Fig. 2d). The xseA1 and xseA2 fragments were also fused to lacZ in the reporter plasmid pRW50. Measurements of β-galactosidase activity suggest a role for the marbox in transcription activation (Fig. 2e). Consistent with this, raising the intracellular concentration of MarA stimulates PxseA only in the presence of the marbox (Supplementary Fig. 3). Note that this approach avoids deletion of marA and the possibility of compensatory regulation by SoxS or Rob.

7Consistent with previous observations, the minimum inhibitory concentration (MIC) of ciprofloxacin decreased in cells lacking xseA (Supplementary Table 2). To determine if the ciprofloxacin hypersensitivity phenotype was dependent on both xseA and the marbox, we used genetic complementation. Hence, we constructed derivatives of plasmid pBR322 encoding xseA downstream of PxseA. This plasmid was able to rescue growth of an xseA::kan E. coli strain in the presence of ciprofloxacin but pBR322 with no insert was not (Supplementary Fig. 4). In the absence of ciprofloxacin, deletion of the marbox had little effect (Fig. 2f, top). However, in the presence of ciprofloxacin, the marbox was essential for growth (Fig. 2f, bottom).

8Exonuclease VII comprises large and small subunits; XseA and XseB, respectively. However, MarA binding was only detected upstream of xseA (Table 1). Hence, XseA could act independently of XseB to increase quinolone tolerance. Briefly, XseA consists of four functional regions: an N-terminal OB-fold, a central catalytic domain, a coiled-coil and a short C-terminal domain (Supplementary Fig. 5a). The OB-fold is responsible for DNA binding, the catalytic domain mediates exonuclease activity, the coiled-coil binds XseB and the C-terminal domain contains three α-helices. To investigate the role of these activities, we introduced point mutations at key positions in xseA encoded by pBR322. The mutations, previously characterised by Poleszak et al. are described in Supplementary Fig. 5a. Inactivation of any xseA determinant required for exonuclease VII function, including the interaction between XseA and XseB, results in hypersensitivity to ciprofloxacin (Supplementary Fig. 5b). Consistent with this, Nichols et al. also concluded that both components of exonuclease VII were required for innate levels of ciprofloxacin resistance. We conclude that XseA acts via its described exonuclease VII activity in our assays.

9Exonuclease VII is known to influence DNA repair,. Hence, the simplest interpretation of our data is that MarA activates xseA to reduce DNA damage. We tested this prediction by visualising DNA damage in Hoechst-stained E. coli cells. As expected, in the presence of ciprofloxacin, xseA::kan cells were filamentous with abnormal nucleoid morphology (Fig. 2g, compare top two panels). Plasmid pBR322 encoding xseA was able to negate this phenotype, while deletion of the PxseA marbox blocked complementation (Fig. 2g, bottom two panels). We confirmed that the unusual nucleoid morphology was indicative of DNA fragmentation using pulse field gel electrophoresis (PFGE) (Fig. 2h). Chromosomal DNA ran as a tight band (lanes 1–3) but smearing was evident upon ciprofloxacin treatment of cells lacking xseA (lanes 4, 5). As in the microscopic analysis, genetic complementation required an intact marbox (lanes 6, 7). For direct comparison, lanes 3, 5, 6 and 7 equate to the top through bottom panels in Fig. 2g, respectively.

10We next turned our attention to the mlaFEDCB locus that also co-binds MarA and σ70 (Fig. 3a). We confirmed binding of MarA to the mlaFEDCB regulatory DNA using EMSA and DNAse I footprinting assays (Fig. 3c, d). The experiments were done using the mlaF1 and mlaF2 DNA fragments. The 5′ end of each fragment is marked by an inverted triangle in Fig. 3b. The results are consistent with MarA binding to the predicted marbox (green in Fig. 3b). However, the location of nearby promoters is unknown. To identify MarA regulated promoters, we used in vitro transcription assays. A DNA fragment containing the mlaFEDCB intergenic region was cloned upstream of the λoop terminator in plasmid pSR. The resulting plasmid was used as a template for RNA synthesis (Fig. 3e). Note the 108 nt RNAI transcript is derived from the plasmid replication origin and serves as an internal control. Transcripts of 128, 148 and 157 nt in length were also observed (Fig. 3e). These initiate within the mlaFEDCB intergenic DNA at sites denoted by a bent arrow (Fig. 3b). Each messenger RNA start site maps downstream of promoter −10 and −35 elements. We refer to the promoters as mlaP1, mlaP2 and mlaP3 (Fig. 3b). MarA activates mlaP2 by binding adjacent to the −35 hexamer (Fig. 3b, e). Hence, deletion of the marbox reduces transcription derived from the mlaFEDCB intergenic region in lacZ fusion assays (Fig. 3f). Similarly, MarA overproduction activated mlaFEDCB only in the presence of the marbox (Supplementary Fig. 3). Inactivation of mlaP1, which overlaps mlaP2 and the marbox, increased the stimulatory effect of the marbox on mlaP2 activity (Supplementary Fig. 6).

11We quantified an eight-fold reduction in doxycycline MIC for mlaE::kan compared to wild-type cells (Supplementary Table 2). To confirm that doxycycline hypersensitivity was dependent on both mlaFEDCB and the marbox, we used genetic complementation. Thus, we constructed derivatives of plasmid pBR322 carrying mlaFEDCB and the upstream regulatory DNA. This plasmid was able to rescue growth of BW25113 mlaE::kan cells in the presence of doxycycline (Supplementary Fig. 7). As expected, deletion of the marbox had little effect in the absence of doxycycline (Fig. 3g, left). Conversely, in the presence of doxycycline, the marbox was essential for growth (Fig. 2g, right).

12In Gram-negative bacteria, cell surface barrier function relies on outer membrane asymmetry; the inner and outer leaflets comprise phospholipids and lipopolysaccharides, respectively. The ABC transport system encoded by mlaFEDCB removes unwanted phospholipids from the outer leaflet. Hence, mlaFEDCB could enhance drug resistance by improving outer membrane barrier function. Alternatively, ABC transport activity could support drug efflux. To assess these hypotheses, we compared drug accumulation and efflux in wild-type and mlaE::kan cells. Accumulation increased dramatically upon disruption of mlaE (Fig. 3h), but efflux was identical in both strains (Fig. 3i). Hence, our data are consistent with increased cell surface permeability but not defective efflux.

13Our model predicts reduced barrier function due to accumulation of phospholipids in the outer leaflet of the outer membrane. This should coincide with increased cell surface hydrophobicity. To test this, we measured partitioning of wild-type and mlaE::kan cells in a two solvent system (aqueous PUM buffer and p-xylene). In this assay, cells with increased surface hydrophobicity migrate to the organic phase. Consequently, turbidity of the aqueous phase is reduced. Consistent with this, mlaE::kan cells were retained in aqueous suspension less efficiently than the parent strain (Fig. 3j). Similar results were obtained in crystal violet adsorption assays; the hydrophobic dye was bound more efficiently by the mlaE::kan strain (Fig. 3k). As expected, similar phenotypes arose in complementation experiments if the marbox was deleted in the pBR322 derivative carrying mlaFEDCB (Supplementary Fig. 8).

14The SoxS and Rob proteins share 42 and 43% sequence identity with MarA across their helix-turn-helix determinants, respectively. Rob usually binds marboxes with a higher affinity than either MarA or SoxS. Indeed, Rob can bind DNA fragments containing no marbox more tightly than MarA and SoxS bind canonical targets. Hence, we next sought to better understand the comparative affinity of MarA, SoxS and Rob for the xseA and mlaFEDCB promoters. In control experiments, we confirmed that Rob had appreciable non-specific DNA binding activity (Supplementary Fig. 9a) and also bound a known marbox with higher affinity than MarA or SoxS (Supplementary Fig. 9b). Interestingly, most well-defined marboxes preferentially bind SoxS rather than MarA (Supplementary Table 1). This was not the case for the xseA or mlaFEDCB promoters, which bound MarA but not SoxS (Supplementary Fig. 10).

15Overexpression of MarA-like regulators is associated with clinically relevant resistance to quinolones and tetracyclines in many enteric bacteria. Hence, we quantified conservation of MarA targets among the Enterobacteriaceae. The result of the analysis is illustrated in Fig. 4. As expected, MarA binding sites associated with efflux systems and porin expression were found in most species. The xseA marbox was similarly distributed. However, the best conserved marbox occurred upstream of mlaFEDCB; this sequence was present in all but the most divergent Cedecea neteri genome. Overall, the E. coli MarA regulon is best conserved among Escherichia spp. and Shigella spp. However, a core regulon is shared by many Enterobacteriaceae. For example, Salmonella, Citrobacter, Klebsiella, Enterobacter and Raoultella spp. share a similar subset of MarA target genes including tolC, micF, xseA and mlaFEDCB.

16MarA-like proteins are frequently implicated in the development of clinical resistance to quinolone and tetracycline family antibiotics,. For example, constitutive MarA production reduces the rate of killing by norfloxacin and ofloxacin in E. coli. This provides an opportunity for subsequent beneficial mutations to arise,. Of the MarA target genes, we identified loss of xseA, rather than tolC, had the biggest impact on quinolone activity (Fig. 1d). Hence, we demonstrate a direct molecular link between MarA expression and reduced DNA damage (Fig. 2). This overlooked feature of the mar system in E. coli is likely to be widespread; the position and sequence of the marbox at PxseA is conserved in many enteric bacteria (Fig. 4). Deregulation of the mar operon is also associated with evolution of resistance to tetracyclines. Indeed, clinically relevant levels of tetracycline resistance have been attributed solely to overproduction of MarA in E. coli. Interestingly, substantial resistance to some tetracycline family antibiotics is retained in marR mutants lacking tolC or acrAB. Hence, the mar regulon must encode other determinants for drug resistance. Our data show that the mlaFEDCB operon is of particular importance; the locus couples MarA to the control of lipid trafficking and outer membrane integrity (Figs. 1, 3). We note that tetracycline family antibiotics vary in their properties; doxycycline and minocycline are more hydrophobic than tetracycline. Hence, increased surface hydrophobicity, due to inactivation of mlaFEDCB, renders cells most sensitive to the former two drugs (Fig. 1d). Consistent with our interpretation, others have noted a positive correlation between compound hydrophobicity and resistance provided by the mar system,.

17The potential for cross-talk with other AraC family proteins complicates the study of gene regulation by MarA. We show that both the xseA and mlaFEDCB promoters preferentially bind MarA rather than SoxS (Supplementary Fig. 10). Furthermore, upregulation of MarA expression activates these promoters even when SoxS and Rob are present (Supplementary Fig. 3). Even so, we do not exclude the possibility that MarA targets described here will bind closely related proteins in some circumstances (Supplementary Figs. 9, 10). Indeed, the mlaFEDCB locus is a target for RamA in Klebsiella pneumoniae. Similarly, of 25 E. coli SoxS targets identified using ChIP-exo, 7 are bound by MarA in this study (ybaO, acrZ, ybjC, ycgZ, micF, ypeC and yjcB). Surprisingly, genes encoding transcription factors, including the global regulator FNR, were the second most common class of MarA targets (Table 1; Fig. 4). This provides an explanation for the pleiotropic effects of MarA on gene expression. Furthermore, because five MarA target genes were also a target for FNR, responses to oxidative and antibiotic stress must overlap,. In conclusion, our work identifies previously unrecognised pathways to antibiotic tolerance mediated by MarA-like regulators. We suggest that the proteins involved are excellent drug targets; inhibition of XseA and MlaFEDCB should enhance the efficacy of quinolones and tetracyclines.

18ETEC strain H10407 is described by Crossman et al. The E. coli K-12 strains JCB387 and BW25113 are described by Page et al. and Datsenko and Wanner, respectively. The xseA::kan and mlaE::kan derivatives of BW25113 were obtained from the Keio collection Plasmids pRW50 and pSR are described by Lodge et al. and Kolb et al. More detailed descriptions of strains and plasmids, along with sequences of oligonucleotides, are provided in Supplementary Table 3.

19Immunoprecipitations with anti-MarA, anti-MarR and anti-σ70 antibodies were done as described by Haycocks et al. using lysates of bacterial strain H10407 transformed with plasmid pRGM9818 encoding marA under the control of the tac promoter. This allowed us to circumvent repression of the chromosomal mar locus by MarR. Immunoprecipitations with anti-FLAG were done using lysates of H10407 transformed with pAMNF (encoding MarR-3xFLAG) or pAMNM (encoding MarR-8xMyc). Lysates were prepared from log phase cells cultured in LB medium. Anti-σ70 and anti-FLAG were purchased from Neoclone (Madison, USA) and Sigma, respectively. Anti-MarA and anti-MarR were a kind gift from Lee Rosner and Bob Martin (NIH, Bethesda). Libraries were prepared using immunoprecipitated protein–DNA complexes immobilised with Protein A sepharose. DNA fragments were then blunt ended, A-tailed and barcoded. This was done using an NEB Quick Blunting and Ligation Kit, the Klenow fragment (5′–3′ exo-, NEB) and NEXTflex ChIP-seq barcodes (Bioo Scientific). Following elution of complexes from the Protein A sepharose, crosslinks were reversed and barcoded libraries were amplified by PCR. The number of PCR cycles was determined empirically for each library. After amplification, library concentration was quantified using Qubit analysis and real-time PCR. Equimolar library concentrations were pooled and sequenced using an Illumina MiSeq instrument. Sequencing reads are stored in ArrayExpress under accession number E-MTAB-5521 and E-MTAB-5591.

20The Fastq files obtained after DNA sequencing were converted into Fastq Sanger format, using FastqGroomer, and aligned to Genbank reference sequences (FN649414.1, FN649418, FN649417, FN649416 or FN649415) using BWA for Illumina. The reference sequences correspond to the H10407 chromosome and plasmids p948, p666, p58 and p52, respectively. The resulting SAM files were converted to BAM format using SAM-to-BAM. For each experiment, coverage per base was determined using multiBamSummary. Subsequent processing was done using R. Data were normalised to the same average read depth and mean coverage per base was determined for each pair of replicates. The immunoprecipitations with anti-MarR were poorly efficient; enrichment of the marRAB promoter was evident but most peaks were associated with highly transcribed genes. These data served as a useful control for the anti-MarA data set; non-specific signals were removed by subtracting the anti-MarR signal from the equivalent anti-MarA value. Conversely, the anti-FLAG immunoprecipitations, with lysates of H10407 expressing MarR-3xFLAG, were highly efficient and isolated only the marRAB promoter. To select peaks for MarA or σ70 binding, we used Artemis to generate a coverage plot and selected peaks scoring >2.7-fold (for MarA) or >3-fold (for σ70) over background. A small number of peaks were called twice because they oscillated above and below the set threshold. Such duplications were removed manually. Four peaks for MarA binding were added manually after visual inspection. The peak centres were set as the centre of the region passing the cut-off rounded to the nearest integer.

21After defining the MarA peak centres, we created a set of genome features in gff file format. Feature boundaries were 100 bp either side of each peak centre. The 201 bp DNA sequence corresponding to each feature was extracted using Artemis and submitted to MEME to search for motifs. The expected number of sites was set to one per sequence and the minimum motif width was set to 15 bp. A single statistically significant motif (E-value <1e−12) was recovered and this matched the known MarA binding consensus. The E-value is derived by MEME, from the motif’s log likelihood ratio, taking motif length and background DNA sequence into account. To determine the distance between peak centres for σ70 and MarA, we used the fetch closest non-overlapping feature tool in Galaxy. Peaks for σ70 and MarA described as overlapping were those centred were within 150 bp of each other.

22Excluding E. coli species, we used BLASTp to search for genomes encoding MarA. We manually removed genomes encoding MarA with changes in the amino-acid sequence of either DNA recognition helix; this is indicative of altered DNA binding specificity. Many of the resulting 161 genomes were derived from closely related strains of the same species. These were omitted so that 29 representative genomes remained. Phylogeny was determined using the sequence of the polA gene from each organism and BLASTn pairwise alignments. We used BLASTn to search the 29 genome sequences for a match to the 201 bp DNA sequence derived from each MarA ChIP-seq peak. If multiple hits were obtained from a single genome, only the best match was used. If required, pairwise alignments were optimised manually to remove alignment gaps within marbox sequences. If no sequence match was identified, this was scored as ‘region absent’. An identified marbox was scored as ‘conserved’ if it matched either the equivalent E. coli sequence, or the consensus marbox (5′-gcactaattgctaaa-3′) in at least 14 of the 15 possible positions. Sites were scored as ‘conserved with mismatches’ if the above criteria were satisfied at 13 of the 15 marbox positions. Sequences falling below this threshold we scored as ‘not conserved’.

23To locate regions of the E. coli K-12 genome equivalent to those in strain H10407, we used BLASTn. We compared the sequence of the marbox, intergenic region (defined as 200 bp upstream of the target gene or entire region between convergent genes) and target gene. Of the 28 shared marboxes, all but 6 were within identical intergenic regions. Of the six intergenic regions that differed, four contained a single base change and two had differences in just two positions. None of the intergenic regions contained insertions or deletions, consistent with identical juxtaposition of promoter elements and regulator binding sites between strains. Target genes were a minimum of 97% identical at the nucleotide level and 13 genes had 100% identity.

24Purified MarA, SoxS and Rob were a gift from Lee Rosner and Bob Martin (NIH, Bethesda). RNA polymerase was purified using a method derived from Burgess and Jendrisak. Briefly, E. coli strain MG1655 was grown overnight in 3 l of LB broth. Cells were harvested by centrifugation and resuspended in 100 ml of lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 2 mM MgCl2, 0.1 mM DTT, 2 mM EDTA, 1 mM 2-mercaptoethanol, 5% glycerol, 0.2% Triton X-100 and 0.25 mg/ml lysozyme). One protease inhibitor cocktail tablet (Roche) was added per 20 ml of buffer. Cell lysis and DNA shearing was done using 4 × 30 s pulses, at 20% output, with a Misonix XL2020 tip sonicator. Lysates were cleared by centrifugation at 39,000×g for 45 min at 4 °C. Following filtration (0.45 µm filter) Polymin P and ammonium sulphate precipitation were done as described in Burgess and Jendrisak. Precipitated protein was resuspended in TGED buffer (10 mM Tris-HCl pH 7.9, 5% glycerol, 0.1 mM EDTA and 0.1 mM DTT) containing 100 mM NaCl and passed through a HiPrep Heparin FF column (GE Healthcare). The column was washed with 0.1 M NaCl TGED and RNA polymerase was eluted in TGED using a gradient to 1 M NaCl. RNA polymerase containing fractions were pooled and protein precipitated using ammonium sulphate. After resuspension in TGED, RNA polymerase was further purified using a Mono Q HR column (GE Healthcare). Column washing and protein elution were as described in the previous step. RNA polymerase containing fractions were pooled and dialysed against −80 °C storage buffer (TGED, 0.1 M NaCl, 50% glycerol).

25For EMSA experiments, DNA fragments were prepared using PCR as described by Shimada et al. with oligonucleotides listed in Supplementary Table 3. Protein binding and subsequent electrophoresis were done as described by Chintakayala et al. For footprinting experiments, DNA fragments were prepared as described by Grainger et al. Protein binding, DNA digestion and electrophoresis were done as described by Singh and Grainger. Briefly, DNA fragments were labelled at one end using [γ-32P]-ATP and T4 polynucleotide kinase and used at a final concentration of ~10 nM in footprinting reactions. All reactions contained excess of herring sperm DNA (12.5 µg/ml) as a non-specific competitor. Our in vitro transcription assays were done as described by Haycocks et al. Briefly, supercoiled pSR plasmid carrying promoter inserts (16 µg/ml) was pre-incubated with MarA in buffer containing 20 mM Tris pH 7.9, 5 mM MgCl2, 500 µM DTT, 50 mM KCl, 100 µg/ml BSA, 200 µM ATP, 200 µM GTP, 200 µM CTP, 10 µM UTP and 5 µCi [α-32P]-UTP. Purified E. coli RNA polymerase was added to start reactions. DNAseI digested DNA and in vitro generated RNA transcripts were analysed on 6% DNA sequencing gels (molecular dynamics). The results were visualised using a Fuji phosphor screen and Bio-Rad Molecular Imager FX. Raw gel images are in Supplementary Fig. 11.

26β-galactosidase assays were done as described previously, using the protocol of Miller. All assay values are the mean of three independent experiments with a standard deviation equivalent to <10% of the mean β-galactosidase activity. Cells were grown aerobically at 37 °C to mid-log phase in LB medium unless stated otherwise.

27A single colony of each bacterial strain was used to inoculate 5 ml of LB broth that was incubated overnight at 37 °C. The OD650 of overnight cultures was recorded so that equivalent OD650 units could be used for sub-culturing each strain in fresh LB medium. The sub-cultures were then placed in a shaking incubator at 37 °C and, at 40 min intervals, 200 μl was transferred to a 96-well plate to measure OD650 units. We used the MIC brothmic microtitre double dilution method to assess antibiotic sensitivity. These assays were done in a final volume of 100 μl of LB medium in a 96 well round bottomed microtitre plate. Each well contained 10 colony forming units of E. coli, 100 μl of LB medium and antibiotics as required. The microtitre plate was covered with a sterile lid and kept in a gently shaking incubator for 24 h at 37 °C. The MIC was the lowest antibiotic concentration that prevented bacterial growth. Results were only accepted if the observed MIC for the control NCTC E. coli 10418 and ATCC E. coli 25922 strains was within one doubling dilution of the expected result.

28Cells were grown in 1.5 ml of LB medium in the presence or absence of 0.005 µg/ml ciprofloxacin for 36 h at 37 °C. Cells were harvested by centrifugation and washed three times with PBS (140 mM NaCl, 2 mM KCl, 8 mM Na2HPO4, 1.5 mM KH2PO4). After washing, cells were resuspended in 10 μl of PBS containing Hoechst 33258 (5 µg/ml) and 40% (v/v) glycerol and left at room temperature for 10 min. Microscope slides were prepared by spreading 5 μl poly-l-lysine (10 mg/ml; Sigma) onto the centre of a slide. After drying, the slide was loaded with 5 μl of the cells suspension and a cover slip applied. The slides containing the cells were imaged using a Nikon Eclipse 90i microscope containing a Nikon Intensilight C-HGFI lamp, Hamamatsu ORCA ER camera (pixel size 6.45 μm) and Nikon Plan Apo VC ×100 Oil immersion lens (Numerical Aperture 1.4). The magnification was ×100 and a DAPI filter set allowed detection of Hoechst 33258-stained nucleoids. Exposure time was 90 ms, the excitation filter range was 340–380 nm and barrier filter range was 435–485 nm. Microscopy was done at room temperature and slides were imaged within 30 min of preparation. Images were analysed using the Nikon’s NIS elements software.

29Preparation of DNA for PFGE was based on the method described by Heath et al. Overnight cultures of BW25113 and derivative strains were grown in presence or absence of 0.005 µg/ml ciprofloxacin at 37 °C for 36 h. After growth, cells were recovered, washed with PIV buffer (10 mM Tris-HCl pH 7.6, 1 M NaCl) and suspended in PIV at a final OD650 of 1.7 units. The suspensions were incubated at 37 °C for 10 min and mixed with an equal volume of 1% PFGE grade agarose (Amresco Agarose LF) at 42 °C. Agarose plugs were prepared by pouring the suspension into PFGE moulds. After solidification, plugs were transferred to EC Lysis buffer (6 mM Tris-HCl pH 7.6, 1 M NaCl, 100 mM EDTA, 0.2% deoxycholate, 1% N-lauroylsarcosine, 1 mg/ml lysozyme, 20 µg/ml RNase) and incubated overnight in a gently shaking incubator at 37 °C. Buffer was removed and plugs were washed five times with TE buffer (10 mM Tris-HCl, pH 7.5, 1 mM EDTA). After washing, plugs were incubated overnight at 37 °C in ESP buffer (0.5 M EDTA pH 9.5, 1% N-lauryl sarcosine, 50 µg/ml proteinase K). The following day, plugs were washed five times with TE buffer and then stored in TE buffer at 4 °C until used. PFGE was done over 24 h in 0.5× TBE using a CHEF-DR II module system. The initial and final switch times were set to 60 and 120 s, the temperature was 14 °C, voltage was 6 V/cm and the included angle was 120°. The gel was stained using ethidium bromide.

30Accumulation of doxycycline was measured using protocols derived from Mortimer and Piddock. About 10 ml of fresh LB broth was inoculated with 250 μl of an overnight culture and placed in a shaking incubator at 37 °C. When the culture had obtained an OD650 of ~0.7 units, the cells were harvested by centrifugation and resuspended in 10 ml of ice-cold PBS. After washing, cells were resuspended in PBS with a final OD650 value of 20 units. The suspension was then transferred to a sterile universal container and left to equilibrate at 37 °C with magnetic stirring. Doxycycline was added to the cell suspension at a final concentration of 35 μg/ml. At timed intervals, after the addition of doxycycline, 500 μl of cell suspension was added to an Eppendorf tube containing 1 ml of ice-cold PBS. After mixing, cells were recovered by centrifugation at 4 °C. Samples were washed with ice-cold PBS to remove residual doxycycline and stored on ice until the end of the time course. After being resuspended in 1 ml of 100 mM glycine (pH 3), and incubated for 2 h at room temperature, cell debris was removed by centrifugation and fluorescence was measured. Efflux assays were based on protocols described by Blair et al. Cells were initially grown as described above. After washing with PBS, cells were resuspended in PBS at an OD650 of 0.3 units. Ethidium bromide was added to a final concentration of 5 μg/ml and the efflux inhibitor chlorpromazine was added to a concentration of 50 μg/ml. The suspension was incubated at 25 °C for 60 min. Once loaded with ethidium bromide, cells were washed with PBS and resuspended in PBS containing 0.4% v/v glucose to induce efflux.

31Partitioning of cells between PUM buffer and p-xylene was done as described by Rosenberg et al. Crystal violet binding assays were done as described by Halder et al.

32ChIP-seq reads have been deposited in the ArrayExpress database under accession codes E-MTAB-5521 and E-MTAB-5591. The authors declare that all other data supporting the findings of the study are available in this article and its Supplementary Information files, or from the corresponding author upon request.