Escherichia coli ATCC^® 25922^TM was used as the background strain. This microorganism is a recommended CLSI control strain used worldwide for antimicrobial susceptibility testing (including quinolones). First, four LLQR isogenic strains were tested using a global transcriptomic approach: ATCC 25922 (wild-type, MIC for ciprofloxacin 0.004 μg/mL); EC14 (E. coli ATCC 25922 coding for QnrS1, MIC for ciprofloxacin 0.125 μg/mL); EC19 (E. coli ATCC 25922 with deleted marR gene and coding for QnrS1, MIC for ciprofloxacin 0.5 μg/mL); and EC24 (E. coli ATCC 25922 with the Ser83Leu substitution in GyrA and coding for QnrS1, MIC for ciprofloxacin 1 μg/mL) (Table 1; Machuca et al., 2014). All of these were susceptible to quinolones according to CLSI breakpoints (CLSI). Second, E. coli ATCC 25922 and two isogenic mutants harboring chromosomal and/or PMQR mechanisms (EC04: E. coli ATCC 25922 with the Ser83Leu substitutions in GyrA and Ser80Arg in ParC, MIC for ciprofloxacin 0.5 μg/mL; and EC59: E. coli ATCC 25922 with the Ser83Leu and Asp87Arg substitutions in GyrA and Ser80Arg in ParC, deleted marR gene and coding for QnrS1, MIC for ciprofloxacin 32 μg/mL) were used for sdhC (succinate dehydrogenase complex subunit C) or cyoA (cytochrome oxidase subunit II) inactivation, as described (Table 1; Datsenko and Wanner, 2000; Machuca et al., 2014). qnrS1 gene was cloned into pBK-CMV vector as described previously (Machuca et al., 2014). Empty pBK-CMV vector was used as control and also introduced into E. coli ATCC 25922 and EC04 for comparison.

Escherichia coli ATCC 25922 (wild-type) and isogenic EC14, EC19, EC24 (LLQR) strains were tested to evaluate the global response to relevant fixed concentrations of ciprofloxacin (1 μg/mL, the breakpoint for reduced susceptibility according to CLSI and near to 1/2 of serum Cmax) (Mandell et al., 2010). Transcriptomic profile of E. coli ATCC 25922 was also compared to EC14 in absence of ciprofloxacin as control. All were susceptible to quinolones according to CLSI breakpoints (CLSI). Cultures were started from single colonies and grown overnight in 25 ml of LB (Luria Bertani medium, Becton Dickinson, Le Pont-de-Claix, France). These cells were diluted 1:100 and grown to cell concentrations of 4x10⁸ cells/ml (OD_600nm = 0.4, exponential phase) for treatment. Three biological replicates per genotype were incubated at 1 μg/mL ciprofloxacin for 60 min (i.e., 250xMIC for E. coli ATCC 25922, 8xMIC for EC14, 2xMIC for EC19 and 1xMIC for EC24). Approximately 10⁹ cells (2 ml) were taken for RNA isolation. The pellet was processed immediately. The pellet was pre-incubated in RNAprotect Bacteria Reagent (Qiagen, Hilden, Germany). RNA extraction was performed using the RNeasy Mini Kit (Qiagen, Hilden, Germany). Contaminating DNA was removed from RNA samples with TURBO DNA-free (Ambion, United States). The RNA concentration was measured using NanoDrop ND-1000 (Thermo Scientific). The integrity of the RNA samples was analyzed with the BioAnalyzer 2100 (Agilent Technologies) using the RNAnano 6000 kit (Agilent Technologies). Labeling, microarray hybridization, scanning, and data processing were performed in the Genomics Unit of the Centro Nacional de Biotecnología¹. The data obtained from each LLQR mutant was always compared with the E. coli ATCC 25922 wild-type strain. The FIESTA program was used for analysis² and the p-value was determined according to the FDR algorithm (p < 0.05 was considered significant). Microarray RNA data are available at GEO NCBI³ (accession number GSE86341).

Gene functions (COG, Clusters of Orthologous Groups) (Tatusov et al., 2000) were identified using the following bioinformatics resources for E. coli: ECOCYC⁴, ECOGENE⁵, Gene Expression Database⁶, KEGG⁷, BPROM⁸, and DAVID⁹. Venn diagrams were created using the http://bioinfogp.cnb.csic.es/tools/venny/index.html website.

Real time RT-PCR was used, as described (Rodríguez-Martínez et al., 2012), to confirm specific transcriptome data (sdhC, sucD, nouH, recA, focA, tnaA genes), both in presence and absence of ciprofloxacin. The LightCycler FastStart DNA Master SYBR Green I Kit (Roche) was used for amplification. To normalize expression levels, target gene transcripts were calculated relative to the mdh gene, using the 2^-ΔΔCT method. The primers used for gene amplification are indicated in Supplementary Table 1.

Antimicrobial susceptibility was determined by microdilution according to CLSI guidelines (CLSI). The quinolones tested were ciprofloxacin, norfloxacin, ofloxacin, levofloxacin, moxifloxacin and nalidixic acid (Sigma–Aldrich, Madrid, Spain).

Potentiating antibacterial activity by enhancing microbial ROS production has been demonstrated previously (Brynildsen et al., 2013). To test the impact of ROS modulation (by sdhC or cyoA deletion) on bacterial viability in E. coli ATCC 25922, EC04 and EC59 strains, time-kill curves were assayed in MHB (Mueller-Hinton broth, Becton Dickinson, Le Pont-de-Claix, France) at ciprofloxacin and ofloxacin concentrations of 2xMIC (Table 1). Growth in drug-free broth was evaluated in parallel, as a control. Cultures were incubated at 37°C, and shaken at 180rpm. An initial inoculum of 10⁶ CFU/mL from a fresh overnight culture was used in all experiments. Bacterial concentrations were determined at 0, 1, 2, 3, 4, and 24 h by colony counting on drug-free agar.

In parallel, the time courses of E. coli ATCC 25922, EC04 and EC59 cells treated with carboxin (500 μM), an inhibitor of succinate dehydrogenase (Brynildsen et al., 2013), and a fluoroquinolone (ciprofloxacin or ofloxacin at 2xMIC), were compared with treatment with fluoroquinolones alone. Carboxin (Sigma–Aldrich, Madrid, Spain) was dissolved in 100% ethanol (Brynildsen et al., 2013).

In order to establish the role of LLQR phenotypes on the differential expression in response to quinolones, the expression profiles of isogenic E. coli carrying LLQR harboring frequent chromosomal and PMQR mechanisms (EC14 coding for QnrS1; EC19 with deleted marR gene and coding for QnrS1; and EC24 with the Ser83Leu substitution in GyrA and coding for QnrS1) (Machuca et al., 2014) were evaluated in the presence of significant clinical concentrations of ciprofloxacin (1 μg/mL, the breakpoint for reduced susceptibility according to CLSI and near to 1/2 of serum Cmax) (Mandell et al., 2010; Clinical and Laboratory Standards Institute [CLSI], 2016) and compared with those of wild-type E. coli ATCC 25922 (Figure 1 and Supplementary Tables 2, 3). The MIC values of ciprofloxacin for these susceptible LLQR strains (EC14, EC19, EC24) were 32-, 125- and 250-fold higher compared to the wild-type strain. As expected, significant differences were observed in the numbers of differentially up- and downregulated genes (>5-fold, p < 0.05). Most genes were upregulated and the differences were proportional to the MIC values for strains with LLQR (EC14: 452/203 genes; EC19: 574/256 genes; EC24: 695/381 genes, upregulated and downregulated, respectively) (Figure 1A). Expression of >2500 genes, whether by upregulation or downregulation, varied more than > 2-fold, reflecting the marked differential response to ciprofloxacin in LLQR strains. Under these this clinically significant concentration of ciprofloxacin, variations of > 10-fold for EC14/EC19/EC24 (271/339/450 genes), of > 50-fold (34/32/68 genes) and of > 100-fold (8/8/33 genes) were observed (Figure 1B). Just a few genes had differential expression of more than 200-fold for LLQR cells compared to the wild-type (16 genes for EC24).

Figure 2 shows gene expression classified according to the COG functional categories (Tatusov et al., 2000) (showing values equal to or more than 30-fold, that is, large differences, p-value < 0.05) (see also Supplementary Table 3). Among these genes, four functional groups were chiefly affected, related to the number of genes involved. “Energy production and conversion genes (EPC)” and “Amino acid metabolism and transport genes (AMT)” tended to be genes upregulated at higher levels (≥30-fold). These differences were proportional to MIC values for strains with LLQR (EC14: 9/4 genes; EC19: 15/11 genes; EC24: 41/24 genes for EPC and AMT genes, respectively) (Figure 2). A reduction of genes related to “Bacteriophage Activation (BA)” was observed counter to MIC values for LLQR strains (mostly upregulated in EC14 and EC19) (Figure 2), while there was a considerable number of highly altered genes without functional prediction (both up- and downregulated).

Figure 3 shows differentially expressed genes equal to or greater than 30-fold (p-value < 0.05) (regarded as major differences in terms of expression). Interestingly, the strains with LLQR were proportionally reorganized in response to ciprofloxacin stress (Figure 3). In this regard, 141 genes showed altered expression for at least one strain with LLQR, and 14.9% of these were shared in all three strains. In total, under ciprofloxacin-induced stress, 25 and 47 genes were specifically affected in EC14 and EC24, respectively, while neither of these strains shared any genes with altered expression. EC19, on the other hand, shared 20 and 27 genes with EC14 and EC24, respectively. These data indicate that, with respect to EC14 and EC24, EC19 is at a transitional stage in its response to ciprofloxacin (see also Supplementary Table 3).

The transcriptome microarray data for six specific genes was validated by quantitative RT-PCR, confirming both the upregulation and downregulation results in presence of ciprofloxacin (Supplementary Table 1). No significant differences were observed for these genes in absence of ciprofloxacin (data not shown). No significant differences were observed when E. coli ATCC 25922 and EC14 were compared in absence of ciprofloxacin (data not shown).

Multiple genes implicated in oxidative metabolism and modulation of ROS [related to the TCA cycle (sdhC: Succinate dehydrogenase; pta: Phosphotransacetylase; sucA-D: Succinyl CoA synthase complex; up to 495-fold), aerobic respiration (cyoA-E: Cytochrome oxidase; nuoG or nuoH: Ubiquinone oxidoreductase; atpH: ATP synthase; up to 108-fold), glucose metabolism (gnd: Gluconate-P dehydrogenase; zwf: Glucose-6-phosphate 1-dehydrogenase; up to 5.5-fold) or detoxification systems (dps: Stress-induced Fe-binding, ahpC: Alkyl hydroperoxide reductase, sodB: Superoxide dismutase, katG: Catalase-peroxidase; up to 12.4-fold)] were highly upregulated in LLQR mutants (Figure 4 and Supplementary Table 2). Proteome profiling based on LC-MS/MS (liquid chromatography–mass spectrometry) showed a similar trend with proteins related to ROS modulation over-represented in LLQR mutants (Supplementary Table 4). Differences observed were proportional to MIC values for strains with LLQR. These data indicate that strains with LLQR would produce lower levels of ROS in the presence of clinical concentrations of ciprofloxacin (Brynildsen et al., 2013; Dwyer et al., 2014).

Genes involved in DNA repair and the SOS response are highly upregulated in response to quinolones (Dwyer et al., 2007; Baharoglu and Mazel, 2014). In our experiment, the relative concentrations of ciprofloxacin used were 250xMIC for E. coli ATCC 25922, 8xMIC for EC14, 2xMIC for EC19 and 1xMIC for EC24. Under this clinically significant concentration of ciprofloxacin, a clear reduction of SOS system gene expression was observed in strains with LLQR [related to SOS system activation (recA: the multifunctional DNA recombination and repair protein and master regulator of the SOS system; up to -30.7-fold), DNA repair by homologous recombination (ruvA: Holliday junction recognition protein; up to -4.75-fold), DNA nucleotide excision repair (uvrB: Excision nuclease subunit B; up to -20.3-fold) or DNA repair by translesion synthesis (umuC: Translesion DNA polymerase V; up to -7.9-fold)] (Figure 5 and Supplementary Table 2).

The PCD and TA systems have been linked to cell death in response to quinolones (Erental et al., 2012, 2014). We analyzed patterns of expression of multiple TA systems (both type I and type II) and found both upregulated (gnsA, ymcE, yjhQ, yjhX, or yihE) and downregulated genes (hokA, hokD, symE, chpBK, yhaV, yefM, yoeB, or yfjF) in strains with LLQR (Yamaguchi and Inouye, 2011). Interestingly, the yihE gene (coding for a key kinase regulator that protects E. coli from antimicrobials like quinolones by antagonizing the MazEF TA module) was upregulated in LLQR strains (Supplementary Figure 1 and Table 2) (Dorsey-Oresto et al., 2013).

Altered expression of downstream genes linked to the mazEF pathway involved in PCD control was also detected (Erental et al., 2014). yfiD, clpX, clpP, and yfbU were upregulated; slyD and ygcR were downregulated in strains with LLQR. Interestingly, genes for the major quinolone target in Gram-negative bacteria, the DNA gyrase subunits, tended to be downregulated (gyrA: DNA gyrase, subunit A, up to -7.18-fold and gyrB: DNA gyrase, subunit B, up to -8.9-fold) in strains with LLQR. Finally, several genes implicated in the DNA mismatch repair system tended to be downregulated (mutH: up to -2.84-fold; mutL: up to -2.14-fold; mutM: up to -5.57-fold; mutY: up to -4.03-fold) in LLQR strains. mutS was not significantly affected (Supplementary Figure 1 and Table 2).

The transcriptomic data obtained indicated that oxidative metabolism (related to the endogenous production of ROS and detoxification systems) plays an important and differential role in quinolone response in strains with LLQR. We tested the impact of sdhC (Succinate dehydrogenase component) and cyoA (Cytochrome O oxidase component) deletions on bacterial viability. Inhibiting these two targets has previously been shown to enhance endogenous microbial ROS production and potentiate the antibacterial activity of quinolones in full susceptible wild-type E. coli (Brynildsen et al., 2013). Here, we tested its putative role in terms of LLQR phenotype sensitization. In terms of time-kill curves (assayed in MHB at ciprofloxacin and ofloxacin concentrations of 2xMIC), sdhC deletion increased sensitivity to fluoroquinolones (both ciprofloxacin and ofloxacin) in wild-type and LLQR strains (Figure 6). A marked reduction in viable bacteria counts was observed for ciprofloxacin after a short incubation (E. coli ATCC 25922: ΔLog = 1.4, EC04: ΔLog = 1.5, EC59: ΔLog = 0.4; after 4 h) and a long one (E. coli ATCC 25922: ΔLog = 0.6, EC04: ΔLog = 1.4, EC59: ΔLog = 2.3; after 24 h). Although it has been suggested that cyoA is implicated in the quinolone sensitization process in E. coli (Brynildsen et al., 2013), only a minor impact was observed in our study in terms of expression (data not shown).

We also wondered whether chemical inhibition of this validated target (sdhC) would increase sensitivity to these bactericidal antimicrobial agents. We treated wild-type and LLQR strains with carboxin, an inhibitor of succinate dehydrogenase (Brynildsen et al., 2013), and measured their susceptibility to ciprofloxacin and ofloxacin, respectively. The addition of carboxin alone had no effect on the growth of wild-type cells (data not shown). However, wild-type and LLQR cells treated with fluoroquinolones and carboxin demonstrated increased sensitivity compared to cells treated with fluoroquinolones alone. For ciprofloxacin, this reduction in viability was observed after a short period of incubation (E. coli ATCC 25922: ΔLog = 1.1, EC04: ΔLog = 0.9, EC59: ΔLog = 0.6; after 4 h) and a long one (E. coli ATCC 25922: ΔLog = 0.6, EC04: ΔLog = 0.3, EC59: ΔLog = 2.2; after 24 h) (Figure 6).