An sRNA Screen for Reversal of Quinolone Resistance in Escherichia coli

In light of the rising prevalence of antimicrobial resistance (AMR) and the slow pace of new antimicrobial development, there has been increasing interest in the development of adjuvants that improve or restore the effectiveness of existing drugs. Here, we use a novel small RNA (sRNA) screening approach to identify genes whose knockdown increases ciprofloxacin (CIP) sensitivity in a resistant strain of Escherichia coli. 5000 sRNA constructs were initially screened on a gyrA S83L background, ultimately leading to 30 validated genes whose disruption reduces CIP resistance. This set includes genes involved in DNA replication, repair, recombination, efflux, and other regulatory systems. Our findings increase understanding of the functional interactions of DNA Gyrase, and may aid in the development of new therapeutic approaches for combating AMR.

The rapid evolution of antimicrobial resistance (AMR) among bacterial pathogens and the slow development of new antibiotics have driven the search for novel approaches to counteract the resistance crisis (Levy and Marshall 2004;Davies and Davies 2010;Spellberg et al. 2004;Payne et al. 2007). Recently, re-sensitization of resistant strains to existing antibiotics has emerged as a promising strategy (Wright 2000;Bassetti et al. 2008). Compounds that reduce bacterial resistance, and that thereby restore the effectiveness of existing drugs, are a promising variety of antibiotic adjuvant (Wright 2016;González-Bello 2017). Certain b-lactamase inhibitors, for example, restore susceptibility to cephalosporins by inhibiting degradative enzymes (extended-spectrum b-lactamases, ESBLs) that are often responsible for resistance (Drawz and Bonomo 2010;King et al. 2014).
Here, we use a bacterial small RNA (sRNA) screen to identify genes whose knockdown re-sensitizes DNA gyrase-mediated resistance to the fluoroquinolone antibiotic ciprofloxacin (CIP). CIP is a synthetic antibiotic used globally for the treatment of many bacterial infections (Hooper 1999;Hooper and Rubinstein 2004;Bolon 2011); high-level resistance is typically conferred by mutations in the gyrA gene, which encodes one subunit of DNA gyrase, the primary target of quinolones (Drlica and Zhao 1997;Walsh 2000). The S83L substitution in the GyrA subunit confers a high level of CIP resistance in E. coli (Bagel et al. 1999;Bhatnagar and Wong 2019). We reasoned that CIP susceptibility might be restored in gyrA mutants by disrupting genes involved in the function of DNA Gyrase, or by altering cell permeability to CIP.
Bacterial sRNAs are widespread, non-coding RNA molecules. They are typically 50-300 nucleotides in size, trans-encoded, with distinct stem loops as secondary structures (Argaman et al. 2001;Gottesman 2004;Sharma 2005, Sharma andVogel 2009;Yoo et al. 2013). They play prominent roles in bacterial physiology by controlling gene expression post-transcriptionally. Each sRNA consists of two important regions. One is the recognition region that regulates sRNA-mRNA base-pairing through antisense short complementary base-pairing with the 59 untranslated region (UTR) or translation initiation region (TIR), and the other is the scaffold (Hfq) region that stabilizes sRNA-mRNA base-pairing (Møller et al. 2002;Zhang et al. 2002;Storz et al. 2004;Jousselin et al. 2009;Vogel and Luisi 2011;Holmqvist and Vogel 2013;Vazquez-Anderson and Contreras 2013;Lee and Moon 2018;Lee et al. 2019). Binding of sRNA to mRNA targets can reduce gene expression by inhibiting translation or promoting mRNA degradation.
In eukaryotes, RNA interference (RNAi) is used extensively for studies of gene function. RNAi mediated gene silencing through short interfering (siRNA) and short hairpin (shRNA) RNAs has become a mainstay in cancer research and is a recognized basis of target validation and drug development (Silva et al. 2005(Silva et al. , 2008Schlabach et al. 2008;Scholl et al. 2009). In prokaryotes, comparable use of sRNA as a genetic tool is promising: proof-of-principle studies have demonstrated sRNA-mediated knockdown of protein levels, and several successful screens have been carried out (Nakashima et al. 2006;Meng et al. 2012;Man et al. 2011;Sharma et al. 2011Sharma et al. , 2013. In the context of AMR, Lee et al. (2011) used a targeted sRNA screen to identify 45 genes whose knockdown reduced resistance to at least one of seventeen clinically relevant antibiotics, including essential genes that would be missed in knockout-driven screening approaches.
In this study, we randomized the antisense sequences of three naturally occurring sRNAs to generate an sRNA expression library with the potential to target diverse mRNA transcripts. We identified a number of sRNA sequences that reduce quinolone resistance on a gyrA S83L background. Further bioinformatic and functional analyses confirmed several genes whose down-regulation reduces resistance levels, and that may thus be promising adjuvant targets.

MATERIALS AND METHODS
Bacterial strains, media and plasmid construction One shot Top10 E. coli (Invitrogen, F-mcrA D(mrr-hsdRMS-mcrBC), j80lacZDM15, DlacX74 recA1 araD139 D(ara leu)7697, galU, galK, rpsL (StrR), endA1, nupG) chemically competent cells were used for the development of randomization methods and for vector maintenance. For the sRNA screen, a quinolone resistant derivative of E. coli K-12 (MG1655) was generated by gene gorging (Herring 2003). Briefly, a fragment of gyrA encoding the S83L substitution (via a TCG-.TTG mutation) and an I-SceI restriction site was generated by megaprimer PCR (Herring 2003) and cloned into the PCR2.1 vector using TOPO cloning (Invitrogen). This donor plasmid was co-transformed into E. coli K-12 (MG1655) along with the mutagenesis plasmid pACBSR. I-SceI endonuclease and l-red functions encoded on pACBSR were then induced with arabinose. Potential mutants were plated on LB, and replica plated to LB+50 mg/ml kanamycin, LB+25 mg/ml chloramphenicol, or LB+25ng/ml ciprofloxacin to identify clones that had incorporated the S83L substitution and lost the donor and mutagenesis plasmids. Successful mutagenesis was confirmed by Sanger sequencing.

Randomized library construction
The small RNA expression vectors pBad-DsrA, MicF and Spot42 (Sharma et al. 2011; kindly donated by Yohei Yokobayashi, Okinawa Institute of Science and Technology) were used as templates for polymerase chain reaction (PCR) reactions. Randomized artificial sRNAs were constructed by incorporating random sequences in the antisense domains of these sRNAs using PCR primers with degenerate bases ( Figure 1A). The native antisense domains of the sRNA scaffolds were absent from these constructs. Platinum Pfx polymerase (Invitrogen) was used to PCR amplify the whole plasmid with a common reverse primer Sartrev containing 20 degenerate bases and an sRNA specific forward primer consisting of 10 degenerate bases for randomizing the antisense domain. The complete list of oligonucleotides used in this study is provided in Supplemental Material, Table S1.
The PCR amplicon was purified (Bio Basic) and digested using the DpnI enzyme (New England Biolabs) for 2 hr at 37°to cleave the methylated template parental DNA. Digested plasmids were phosphorylated using T4 polynucleotide kinase (New England Biolabs) for 20 min at 37°. After phosphorylation, the linear DNA was self-ligated using T4 DNA ligase (New England Biolabs) and incubated at 25°overnight. The antisense domain is randomized with 10-30 randomized bases. The randomized bases were incorporated upstream of the scaffold regions of DsrA, MicF and Spot42 by PCR amplification of source vectors lacking the antisense regions ("empty" sRNA). (B) The subsequent randomized product was transformed into a gyrA S83L mutant to obtain a randomized artificial library, with each clone bearing a novel sRNA construct.
The consistency of the randomization protocol was first verified in Top10 cells. Subsequently, the randomized sRNA library was introduced into an E. coli gyrA S83L mutant by transforming chemically competent cells with the self-ligated PCR amplicons. Transformations were plated on LB agar supplemented with ampicillin, single colonies were picked, grown overnight at 37°at 150 rpm with 100mg/ml ampicillin, and glycerol stocks were prepared in 96 well plates for further screening. Controls in each of the 96 well plates included gyrA S83L carrying empty sRNA plasmid and untransformed E. coli MG1655. Libraries were constructed separately for each of the three expression vectors, consisting of 5000 transformants in total. 40 plasmids from the randomized library (roughly equal numbers from each plasmid backbone) were extracted and sent for Sanger sequencing to confirm randomization.

Ciprofloxacin sensitivity screen
In order to identify sRNA constructs that increased CIP sensitivity, 24 h growth curves in CIP-containing media were assayed for 5000 gyrA S83L mutant clones bearing randomized constructs. Cultures were inoculated at a 1:100 dilution from glycerol stock cultures and grown overnight at 37°, 150 rpm in LB with 100mg/ml ampicillin for sRNA plasmid maintenance. For 24 h growth curves, the overnight cultures were diluted (1:100) in LB supplemented with 100ng/mL CIP, 0.5mM arabinose, and 100mg/ml ampicillin. Arabinose is required to induce expression of the sRNA construct. 100 ng/mL CIP was chosen as an intermediate concentration between the minimum inhibitory concentration (MIC) of the WT (30 ng/mL) and the S83L mutant (600 ng/mL). The OD 600 of each culture was measured on a Biotek ELx808 plate reader every 37 min for 24 h, incubating at 37°with 30 s of shaking every 5 min. Two growth parameters, lag time and maximum growth rate, were estimated using the program GrowthRates (Hall et al. 2014). For each well, OD 600 at time zero was used as a well-specific blank.
Clones showing growth repression in the primary CIP sensitivity screen were selected for secondary screening in the presence and absence of CIP (100ng/ml) using similar experimental methods as described above. Clones whose growth was repressed in the presence of CIP, but not in LB alone, were selected for further analysis.

Target identification
In order to identify potential target mRNAs, sRNA vectors were extracted from CIP-sensitive clones and their sRNA regions were Sanger sequenced. Putative mRNA targets were identified for the randomized sRNA sequences using the online software packages IntaRNA (Mann et al. 2017), Target RNA (Kery et al. 2014) and RNA predator (Eggenhofer et al. 2011). These web based programs predict hybridizations between two RNA molecules, and provide a graphical overview of the sRNA-mRNA binding interactions (Tjaden 2008;Smith et al. 2010;Eggenhofer et al. 2011). The resulting target genes were identified and selected for experimental validation.

Validation for selected targets
Candidate genes whose knockdown may induce CIP sensitization were validated experimentally by constructing double mutants, in which the gyrA S83L mutation was combined with knockout mutations of the computationally predicted target genes. The knockout mutants were selected from the Keio mutant collection (Baba et al. 2006), wherein a kanamycin resistance cassette was used to replace non-essential genes. The gyrA S83L mutation was introduced by oligonucleotide-mediated recombineering (Ellis et al. 2001) into the Keio deletion mutants, to construct each double mutant. A mobilizable derivative of pMA7SacB was used (Lennen et al. 2015), in which an arabinose-inducible promoter (P BAD ) controls expression of the b subunit of l-Red recombinase and the E. coli Dam methylase. Dam methylase induction has been shown to increase mutagenesis efficiencies in E. coli, while reducing offtarget effects, by transiently interfering with the DNA mismatch repair system (Lennen et al. 2015). The plasmid was introduced into recipient strains by conjugation using a donor E. coli strain (WM3064) that is auxotrophic to diaminopimelic acid (DAP) (Dehio and Meyer 1997;Saltikov and Newman 2003). Recipients (Keio mutants) were cultured in LB broth supplemented with 30 mg/ml kanamycin and the donor was cultured in LB supplemented with 0.3 mM DAP and 100 mg/ml ampicillin (for Red plasmid maintenance). Mating mixtures were spotted on LB agar with 0.3 mM DAP and incubated overnight. Exconjugates were selected with 100 mg/ml ampicillin on LB lacking DAP, to prevent growth of donors. l-Red recombination was performed on Keio mutants harboring the Red plasmid as previously described (Lennen et al. 2015) using an oligonucleotide encoding the gyrA (S83L) mutation (59-AACGCAGCGAGAATGGCTGCGCCATGCGG-ACGATCGTGTCATAGACCGCCAAGTCACCATGGGGATGGTA-TTTACCGATTACGTCACCAA-39). A single round of recombination was carried out to minimize off-target effects. Transformants harboring the gyrA mutation were selected by plating on LB supplemented with 30 mg/ml ciprofloxacin and 30 mg/ml kanamycin. The strains were cured of the l Red plasmid by sacB counterselection by streaking on LB agar containing 5% sucrose. The gyrA gene of each putative double n■ Table 1 Randomized sequences of artificial sRNAs constructs. 10-30 randomized bases (bold) were incorporated upstream of the scaffold regions of DsrA, MicF and Spot42 by PCR amplification of source vectors lacking the antisense regions ("empty" vectors). All of the sRNAs start from the vector-derived sequence of 5`ACTCGAG (italics). Example randomized sequences are shown for each construct (2 or 3, as indicated in parentheses)

Scaffold
Randomized region

Randomized sRNA library construction
We generated a library of artificial sRNAs by randomizing the antisense domain sequences of three native E. coli sRNAs: DsrA, MicF, and Spot42. We transformed the sRNA library into an E. coli gyrA S83L mutant and collected a total of 5000 clones, each bearing a plasmid with an artificial sRNA construct ( Figure 1A, B). We arbitrarily isolated 40 clones from the plasmid library and sequenced their sRNA regions to check for the incorporation of randomized sequences. The sequencing results confirmed the presence of unique randomized sequences in Figure 2 Boxplot distributions of CIP-sensitivity phenotypes for 5000 gyrA S83L mutants harboring randomized sRNA plasmid constructs. The growth rate (A) and lag time (B) distributions are depicted for clones cultured in LB with 600 ng/ml CIP. E. coli MG1655 grown without CIP, and gyrA S83L mutants bearing empty sRNA plasmids grown with CIP, were used as controls. The boxplots give the median and first and third quartiles, with whiskers showing either the maximum (minimum) value or 1.5 times the interquartile range of the data, whichever is smaller (larger). The outliers (528 clones) selected for further investigation have substantial reductions in growth rate and/or increased lag phase.
each of the engineered sRNA constructs (example sequences are shown in Table 1).

Ciprofloxacin sensitivity screening
To identify clones with increased susceptibility toward CIP, all 5000 gyrA S83L clones bearing sRNA constructs were grown in LB supplemented with 100ng/mL CIP and arabinose to induce sRNA expression. The MIC of the gyrA S83L mutant is 600ng/ml, so we reasoned that incubation at this sub-MIC concentration of CIP would be a good indicator for changes in sensitivity. Controls included wild-type E. coli MG1655 grown in LB without CIP (median growth rate: 0.06 OD 600 /minute and length of lag phase: 41 min), and a gyrA S83L mutant bearing a non-randomized plasmid grown in LB+CIP (median growth rate: 0.05 OD 600 /minute and length of lag phase: 47.8 min) ( Figure 2). We selected a total of 528 clones exhibiting substantial reductions in growth rate and/or increased lag phase (outliers from Figure 2). From the individual sRNA scaffolds, we selected 199 clones from DsrA (median growth rate: 0.04 OD 600 /minute and length of lag phase: 62.1 min), 231 clones from MicF (median growth rate: 0.04 OD 600 /minute and length of lag phase: 64.6 min), and 98 clones from Spot42 (median growth rate: 0.05 OD 600 /minute and length of lag phase: 60 min). We repeated the growth curves of the 528 putatively CIP susceptible clones in the presence and absence of CIP (Figure 3, Figure S1). This secondary screen was performed in order to verify the CIP-sensitive phenotypes and to determine whether or not the growth deficiencies were specific to CIP; a growth deficit in LB alone would indicate a general fitness effect of the sRNA, rather than reversal of CIP resistance. From this secondary screen, we selected 48 clones showing little or no growth inhibition in LB without CIP (Mean growth rate: 0.07 OD 600 /minute and length of lag phase: 48.1 min), but whose growth was inhibited in LB supplemented with CIP (Mean growth rate: 0.02 OD 600 /minute and length of lag phase: 85 min).

Identification of target genes
We sequenced the randomized antisense regions of the sRNA plasmids isolated from the 48 clones exhibiting growth defects specific to the presence of CIP. Candidate target mRNAs were identified using the online tools IntaRNA, Target RNA and RNA predator (Mann et al. 2017;Kery et al. 2014;Eggenhofer et al. 2011). These packages predict target mRNA sites in the E. coli MG1655 transcriptome with the potential for complementary sRNA interactions. We detected complementary hits for 31/48 randomized sequences (an example of complementarity between an artificial sRNA and its predicted target gene (sbmC) is shown in Figure S2). Since these tools identify complementary matches between short sequences, several sequences were matched to multiple targets, resulting in a total of 222 potential gene targets (Table S2). Candidate targets included genes involved in DNA repair and recombination (recC, recD, recJ, sbmC), the SOS response and error prone replication (umuC, xseA), transcriptional regulation (yeeY, mtlR), cell wall division and assembly (zipA, minC), transport and efflux (emrB, tolQ), two-component regulatory systems (cpxA, cpxR, ompR, cheB), and chaperone proteins (hybB, dnaK).

Experimental validation of targets
We carried out validation experiments for 36 genes that encoded potential mRNA targets (Table 2). Since CIP primarily interacts with DNA Gyrase, we focused on putative target genes related to Gyrase and its functions in DNA replication, repair and recombination. Additional genes were chosen because of their roles in efflux transport systems, two-component regulatory systems, transcriptional regulators and as chaperone proteins (Table 2). For each candidate target gene, the corresponding knockout mutant was obtained from the Keio single gene knockout collection (Baba et al. 2006). The gyrA S83L mutation was then transferred into the Keio clone by l-Red mutagenesis, generating a double mutant.
To determine which candidate target genes conferred increased sensitivity to CIP on the gyrA S83L background, we obtained the MIC of each double mutant and the corresponding single mutants (Table S3). Knockout of a pseudogene, cybC, was used as a control for the effect of the kanamycin cassette present in the Keio clones.
The MIC values of a number of double mutants showed reduced susceptibility when compared to the cybC+S83L (MIC = 1000ng/ml) double mutant. Importantly, for 30 genes, the reduction in CIP resistance Figure 3 Effects of sRNA expression on growth rate (OD 600 /minute) and length of lag phase (minutes) with and without CIP. The 48 selected constructs showing repression in growth in the presence of CIP are indicated in blue.
Volume 10 January 2020 | Reversal of Quinolone Resistance | 83 n■  Donahue and Turnbough (1994) was more pronounced on the gyrA S83L background than on the WT background; these genes fall below the 1:1 line on Figure 4. Inactivation of these genes thus leads to substantial reductions in gyrA-mediated CIP resistance. While we did not explicitly account for off-target effects of the l-Red recombineering used to generate the double mutants, offtarget mutations are unlikely to account for the majority of reductions in MIC, since every double mutant was generated independently.
Mutants showing a reduction in CIP MIC on the gyrA S83L background include xseA, tus (double mutant MIC reduction of 16-fold), sbmC (double mutant MIC reduction of eightfold), tolQ and tolC (double mutant MIC reduction of 16-fold).

DISCUSSION
We carried out an sRNA screen to identify genes whose knockdown restores quinolone sensitivity in E. coli. Following an initial screen of 5000 sRNA-bearing clones, and secondary screening of over 500 clones, we found 30 genes whose disruption increases sensitivity of a gyrA S83L mutant by twofold or more. Our findings expand knowledge of the genetic interaction network of the essential gene gyrA, and provide potential targets for the development of antibiotic adjuvants to restore sensitivity in quinolone resistant pathogens. Chemical-genetic sensitivity screens have largely used knockout approaches, whereby a library of knockout mutants is screened for sensitivity or resistance to an antibiotic at sub-lethal concentrations (e.g., Tamae et al. 2008;Breidenstein et al. 2008;Fajardo et al. 2008;Gomez and Neyfakh 2006;Liu et al. 2010). Typically, such screens are carried out on a wild-type, antibiotic susceptible background, so they are not well-suited to identifying genes whose knockdown reverses resistance. Nonetheless, Tamae et al. (2008) did show that knockouts of five genes recovered from a sensitivity screen (recC, recA, fis, xseA, tolC; all but fis were also found here) did reduce CIP resistance on a gyrA mutant background. sRNA screens offer a powerful means for identifying genetic-background specific effects, since sRNA libraries can be readily generated in any transformable strain (Sharma et al. 2011(Sharma et al. , 2013Lee et al. 2011). Moreover, inducible sRNA constructs can be used for studying gene functions of essential genes as well as non-essential genes (Rodrigo et al. 2012), which is not possible with knockouts. The randomization approach that we have adopted here (see also Sharma et al. 2011Sharma et al. , 2013 is particularly promising, since it does not require targeted cloning of gene-specific sRNAs. We note a need for additional efforts to test and refine sRNA scaffoldsit is unknown, for example, whether different scaffolds differ in the extent of knockdown achieved for their targets. It is likely that a larger pool of sRNA clones would be needed to saturate the genome. It is not straightforward to calculate how many clones would need to be sampled, since each sRNA may target more than one gene. In the limiting case where each sRNA targets only a single gene, approximately 13600 sRNA clones would be needed to "hit" 95% of the 4539 genes in the E. coli genome (calculated as 1-exp(-13600/4539); gene number from ecocyc.org). If each sRNA knocks down more than one target RNA, however, fewer clones would be needed.
Direct use of sRNAs as therapeutics would require the efficient expression of a synthetic sRNA construct inside the target bacterial cell. This would require a competent delivery system, such as phage or a Figure 4 Fold change in MIC values for 36 knockout mutants on wild-type and gyrA S83L backgrounds. Absolute 1xMIC values for the S83L mutant and WT are 1000 ng/mL and 30 ng/mL, respectively. The 30/36 mutants that fall below the 1:1 line show reductions in CIP resistance on the gyrA S83L background but not on the wild-type background.
conjugative plasmid system. An engineered-phage approach illustrated by Lu and Collins (2009) expressed the LexA and OmpF proteins from phage M13. This resulted in reduction in antibiotic resistance evolution in mice when injected with engineered-phages. Thus, a comparable approach could be a plausible delivery system for combating resistant bacteria. An alternative strategy would be to identify small molecules that inhibit the targets identified here. For example, we showed that inhibition of sbmC reduces CIP resistance by twofold, and SbmC is a known target of the peptide antibiotic Microcin B17 (Baquero et al. 1995;Collin et al. 2011). Thus, sRNA-mediated target discovery may be an efficient option for generating prospects for novel therapeutics.

ACKNOWLEDGMENTS
We thank Dr. Vandana Sharma and Dr. Yohei Yokobayashi (Okinawa Institute of Science and Technology) for providing the small RNA expression vectors. We also thank two anonymous reviewers for their thoughtful comments on an earlier draft of the manuscript.