Mutant RNA polymerase can reduce susceptibility to antibiotics via ppGpp-independent induction of a stringent-like response

Abstract

Background

Mutations in RNA polymerase (RNAP) can reduce susceptibility to ciprofloxacin in Escherichia coli, but the mechanism of transcriptional reprogramming responsible is unknown. Strains carrying ciprofloxacin-resistant (Cip^R) rpoB mutations have reduced growth fitness and their impact on clinical resistance development is unclear.

Objectives

To assess the potential for Cip^RrpoB mutations to contribute to resistance development by estimating the number of distinct alleles. To identify fitness-compensatory mutations that ameliorate the fitness costs of Cip^RrpoB mutations. To understand how Cip^RrpoB mutations reprogramme RNAP.

Methods

E. coli strains carrying five different Cip^RrpoB alleles were evolved with selection for improved fitness and characterized for acquired mutations, relative fitness and MIC_Cip. The effects of dksA mutations and a ppGpp⁰ background on growth and susceptibility phenotypes associated with Cip^RrpoB alleles were determined.

Results

The number of distinct Cip^RrpoB mutations was estimated to be >100. Mutations in RNAP genes and in dksA can compensate for the fitness cost of Cip^RrpoB mutations. Deletion of dksA reduced the MIC_Cip for strains carrying Cip^RrpoB alleles. A ppGpp⁰ phenotype had no effect on drug susceptibility.

Conclusions

Cip^RrpoB mutations induce an ppGpp-independent stringent-like response. Approximately half of the reduction in ciprofloxacin susceptibility is caused by an increased affinity of RNAP to DksA while the other half is independent of DksA. Stringent-like response activating mutations might be the most diverse class of mutations reducing susceptibility to antibiotics.

Introduction

Ciprofloxacin is an important antibiotic with activity against Gram-negative and Gram-positive bacteria.^1,2 It binds to DNA gyrase and topoisomerase IV and inhibits the re-ligation of cleaved DNA.^3,4 The accumulation of DNA breaks leads to bacterial chromosome fragmentation and ultimately to cell death.⁵ Ciprofloxacin resistance mutations are commonly located in genes encoding the drug targets DNA gyrase (gyrA, gyrB) and topoisomerase IV (parC, parE).⁶ Additionally, mutations in genes encoding regulatory proteins of efflux pumps (marR, acrR and soxR) can lead to increased drug efflux.⁶ There are also horizontally acquired genes that reduce susceptibility to ciprofloxacin by protecting the drug target, modifying the drug, or encoding a novel efflux pump.^7–11 In Gram-negatives such as Escherichia coli no individual mutation or acquired gene is sufficient to increase resistance above the clinical breakpoint.^12,13

Recent research has revealed that mutations in transcription- and translation-related genes, including RNA polymerase (RNAP) and tRNA synthetase genes, can reduce susceptibility to ciprofloxacin.^14,15 These mutations lead to global changes in bacterial protein synthesis with a net benefit under ciprofloxacin-selection conditions. Mutations in tRNA synthetase genes were shown to induce the bacterial stringent response by reducing the supply of aminoacylated-tRNA to the ribosome, but the underlying mechanism by which mutations in the RNAP itself reduce susceptibility has not been elucidated. Although their clinical impact is not yet clear, polymorphisms within the RNAP genes are found in about 6% of ciprofloxacin-resistant (Cip^R) clinical E. coli.¹⁴ Some of these mutations are similar to those selected in vitro, but sequencing data alone cannot determine whether they were selected to reduce ciprofloxacin susceptibility. A factor that could potentially limit the role of transcription/translation-related resistance mutations in the clinical setting is that they generally cause a substantial fitness cost. The mutations described so far reduce bacterial fitness by 20%–50% and might be counter-selected.^13–16 In the case of costly rpoB mutations causing rifampicin resistance (Rif^R) fitness-compensatory mutations are frequently selected and help maintain the resistance clinically, but compensatory mutations have not yet been identified for Cip^RrpoB mutations.^17–20

In this study, we investigated parameters that could influence the potential impact of Cip^RrpoB mutations on clinical resistance development in E. coli. Our aims were to: (i) estimate the number of distinct Cip^RrpoB mutations; (ii) identify and characterize mutations that improve growth fitness of strains carrying Cip^RrpoB mutations; and (iii) use the compensatory mutations to identify the underlying mechanism by which the transcriptional pattern of the RNAP is altered to reduce susceptibility to ciprofloxacin.

Materials and methods

Bacterial strains and growth conditions

Strains were derived from the E. coli K12 strain MG1655. Bacteria were grown at 37°C in LB broth or on Luria agar (LA) plates (LB solidified with 1.5% agar, Oxoid). Antibiotics were purchased from Sigma–Aldrich (Stockholm, Sweden). Final concentrations of antibiotics were: ciprofloxacin, 0.008–4 mg/L (see MIC test); rifampicin, 1–32 mg/L (see MIC test); tetracycline, 15 mg/L; chloramphenicol, 60 mg/L; ampicillin, 100 mg/L.

Strain constructions

The construction of strains containing rpoB mutations was described previously.¹⁴ Deletions of dksA, relA and spoT were introduced into the WT E. coli chromosome (spoT deletion into a ΔrelA background) by λ-Red recombineering²¹ using the pSIM5 plasmid²² and a tetracycline resistance cassette (tetRA) flanked by FRT sites.²³ Deletions and fluorescence markers were moved between strains using P1 virA-mediated transduction. Excision of tetRA was done by expression of Flp recombinase from a pCP20 plasmid.²³dksA mutations were moved using the Duplication–Insertion Recombineering method²⁴ with a pSIM6 recombineering plasmid²² and a cat-sacB selectable/counter-selectable cassette.²⁵

Estimation of RNAP mutational target size

Independent mutations selected for reduced susceptibility to ciprofloxacin were isolated in rpoB or rpoC (22 once and three mutations 2×, 4× and 6×, respectively) [Table S1 (available as Supplementary data at JAC Online)].^15,26 To simplify calculation, we assumed 22 mutations were isolated once and three mutations isolated twice. The mutational target size N_RNAP was defined as the total number of distinct mutations that will most likely result in the observed distribution (22 singles, three doubles) when 28 random mutations are randomly selected. We simulated evolution experiments by selecting 28 random mutations for various mutational target sizes (60, 70, …, 170). Selection of random mutations was performed using Excel for Mac 15.40 (Microsoft). Each simulation was repeated 500× and the average numbers of single isolations were plotted against the respective mutational target size (Figure S1). N_RNAP was calculated based on a second-order polynomial trend line. This calculation assumes all mutations have an equal chance of being selected, which is most likely not the case. Mutational target size N_RNAP is therefore an underestimate because mutations selected more frequently will skew the distribution to more multiple isolations.

Evolution by serial passage

Independent lineages of each strain were grown overnight with shaking at 37°C in 2 mL of LB. After each cycle of growth, 2 μL of culture was transferred into 2 mL of fresh LB medium to initiate the next cycle. Every 10 cycles (100 generations) cultures were diluted in 0.9% NaCl and approximately 100 cfu were spread onto LA plates and grown overnight at 37°C to visually assess growth improvement. For each lineage, the largest colony was isolated and exponential growth rates of the isolated clones were measured. After 20 cycles (200 generations) isolates from all lineages displayed improved growth rates.

Growth rate measurements

Exponential growth rates were measured using a Bioscreen C machine (Oy Growth curves Ab Ltd). Cultures were grown overnight in LB, diluted 3000-fold in fresh LB (∼10⁶ cfu/mL), then 300 μL incubated at 37°C with continuous shaking in honeycomb microtitre plates. OD (600 nm) was measured at 5 min intervals. Doubling times were calculated from the increase in OD over a sliding window of 10 measurement points. Maximum exponential growth rates were defined as the measurement window with the minimal doubling time. All measurements were performed on three independent cultures.

MIC determination

MICs were determined using broth microdilution in 96-well round-bottomed microtiter plates. Bacteria were suspended in 0.9% NaCl to 0.5 McFarland and 100-fold diluted in 100 μL of LB medium containing rifampicin (1, 2, 4, 8, 10, 12, 14, 16 and 32 mg/L) or ciprofloxacin (0.008, 0.016, 0.032, 0.048, 0.064, 0.125, 0.25, 0.5, 1, 2 and 4 mg/L). MICs were assessed visually after incubation for 18 h at 37°C. All measurements were performed on three independent cultures.

Time–kill assay

Approximately 2 × 10⁶ cfu were transferred from an exponentially growing culture into 2 mL of pre-warmed LB (∼10⁶ cfu/mL) containing ciprofloxacin at 0, 0.25, 0.5, 2 and 4 mg/L. Cultures were grown at 37°C in a shaking water bath and bacterial survival was assessed after 0, 2 and 4 h by plating dilutions on LA plates.

Growth competition experiments

Growth competition experiments were performed as previously described.²⁷ Briefly, strains for competition experiments were labelled with a galK::SYFP2-kanR cassettes (SYFP2 encodes a yellow fluorescence protein) and competed against the isogenic parental strain without rpoB mutation (CH2133) labelled with a galK::mTagBFP2-kanR cassettes (mTagBFP2 encodes a blue fluorescence protein). For each competition, three independent cultures of each strain were grown for 18 h at 37°C in LB. The fluorescence marker strains to be competed were mixed 1:1 and then 2 μL of the mixtures was used to inoculate 2 mL of LB at indicated ciprofloxacin concentrations. Mixed cultures were grown for 24 h at 37°C. Populations were analysed in the initial 1:1 mixture and following the growth cycle (representing 10 generations of growth difference) with a MACSQuant VYB (Miltenyi Biotec). For each population, 10000 cells were counted and the ratio YFP: BFP was determined. The ratios were used to calculate the selective coefficients for each culture using the equation s=[ln(R(t)/R(0))]/[t].²⁸ Minimal selective concentrations (MSCs) were determined by calculating the x-intercept of a second-order binomial trend curve as previously described.^29,30

PCR and local sequencing

DNA amplification was performed using 2× PCR Mastermix (Thermo Scientific, Waltham, MA, USA), according to the protocol of the manufacturer. Amplification products were purified using SureClean Plus (Bioline, Germany), according to the protocol of the manufacturer, and sequencing of purified PCR products was performed by Macrogen (Amsterdam, the Netherlands). Sequences were analysed with the CLC Main Workbench 8.0.1 (CLCbio, Qiagen, Denmark). Primers to amplify and sequence the dksA gene were dksA_fw: TGTGTGTCTGTCATCTCTTT and dksA_rv: TTTACATTCTGGTCGCGT.

WGS

Genomic DNA was prepared using the MasterPure DNA Purification Kit (Epicentre, Illumina Inc., Madison, WI, USA), according to the manufacturer’s instructions. Samples were prepared for sequencing according to Nextera^® XT DNA Library Preparation Guide (Rev. D) (Illumina Inc.). Sequencing was performed using a MiSeq™ desktop sequencer, according to the manufacturer’s instructions (Illumina Inc.). Sequences were analysed with the CLC Genomic Workbench 11.0.1 (CLCbio, Qiagen, Denmark). Full genotypes of sequenced strains are shown in Table S2.

Structural analysis

Molecular graphics and analyses on an X-ray crystal structure of E. coli RNAP and DksA/ppGpp complex (PDB code 5VSW³¹) were performed with chimera version 1.13.1.³²

Statistical analysis

Statistical analysis of the relative growth rate measurements was performed with the R software version 3.5.0 using two-tailed unpaired t-tests.

Results

Many RNAP mutations reduce susceptibility to ciprofloxacin

Previous work identified six rpoB mutations that reduce susceptibility to ciprofloxacin.¹⁴ Additional RNAP mutations affecting rpoB or rpoC were subsequently identified during selections for ciprofloxacin resistance.^15,26 Currently, we know of 25 distinct RNAP mutations, with 15 mutations in rpoB and 10 in rpoC (Table S1). Their locations in the RNAP structure fall within two clusters previously identified.¹⁴ Surprisingly, most mutations (22 of 25) were isolated only once, indicating that a large number of distinct RNAP mutations is able to reduce susceptibility to ciprofloxacin. Based on the observed distribution of mutations, the total mutational target size was estimated to include more than 100 distinct mutations (see the Materials and methods section). This large mutational target size suggests RNAP mutations are a common class among mutations selected to reduce susceptibility to ciprofloxacin.

Isolation of fitness-compensatory mutations

Five RNAP mutations that reduce susceptibility to ciprofloxacin (rpoB Δ442–445, S455dup, E1272G, A1277V and E1279G) were previously reconstructed into an isogenic background carrying ciprofloxacin target and efflux mutations (gyrA D87G, gyrB S464A, marR S65fs).¹⁴ These strains were chosen as the starting point for this experimental evolution study. By allelic replacement each of the rpoB mutations was shown to reduce growth rate 20%–40% (Table 1 and Table S3), increase MIC_Cip from 0.5 to 1–2 mg/L and MIC_Rif from 12 to 16–32 mg/L (Table 1) and increase bacterial survival in the presence of ciprofloxacin (Figure S2). WGS of each strain and their isogenic parents was made to determine their genotypes relative to MG1655 prior to experimental evolution (Table S2).

Table 1.

Genotypes and phenotypes of evolved isolates and parental strains

Strain	Relevant genotypea			Evolution (generations)	Relative fitness ± SDd,e	MIC (mg/L)
	CIPb	rpoB c	compensatory			CIP	RIF
CH1464	–	–	–	WT	1.00±0.02	0.016	10
CH2133	CipR	–	–	isogenic parent	0.91±0.02	0.5	12
CH4959	CipR	Δ442–445	–	unevolved	0.63±0.01	2	64
CH8890	CipR	Δ442–445	dksA D71N	100	0.76±0.00***	2	64
CH8891	CipR	Δ442–445	dksA L95P	100	0.76±0.00***	1	32
CH8892	CipR	Δ442–445	dksA D71A	100	0.74±0.00***	1	32
CH3141	CipR	S455dup	–	unevolved	0.52±0.01	2	64
CH8953	CipR	S455dup	dksA nt G-10A, rpoB V588L, marA I18F	200	0.88±0.03***	0.125	10
CH8954	CipR	S455dup	ΔmarA	200	0.64±0.01***	0.5	32
CH8955	CipR	S455dup	marA I18T	200	0.61±0.00***	0.5	32
CH3144	CipR	E1272G	–	unevolved	0.67±0.02	1	32
CH8956	CipR	E1272G	rpoC G1136S, ΔmarA	200	0.81±0.01***	0.125	10
CH8887	CipR	E1272G	rpoC E1200K	100	0.78±0.00***	0.5	12
CH8889	CipR	E1272G	rpoA R191P	100	0.79±0.01***	0.5	14
CH2332	CipR	A1277V	–	unevolved	0.73±0.00	2	16
CH8875	CipR	A1277V	rpoA R191C	100	0.91±0.00***	1	12
CH8876	CipR	A1277V	rpoC H419P	100	0.78±0.00***	1	12
CH8878	CipR	A1277V	rpoB Q618L	100	0.89±0.01***	1	12
CH3073	CipR	E1279G	–	unevolved	0.55±0.00	2	32
CH8879	CipR	E1279G	rpoB R637C	100	0.69±0.00***	2	16
CH8880	CipR	E1279G	rpoC G1136V	100	0.72±0.03***	1	12
CH8881	CipR	E1279G	rpoC L1275Q	100	0.70±0.01***	1	14

CIP, ciprofloxacin; RIF, rifampicin.

^aFull genotypes are shown in Table S2.

^bCip^R: gyrA D87G, gyrB S464A, marR S65fs.

^cMutations in rpoB that were selected for increased ciprofloxacin resistance.

^dFitness ± SD relative to the WT.

^eSignificance compared with the respective unevolved parental strain was calculated using a two-tailed unpaired t-test (

^*** P ≤ 0.001).

Three independent cultures of each strain carrying Cip^RrpoB mutations (15 cultures) were evolved by serial passage selecting for improved fitness. After 100 generations of growth, a single colony per culture was isolated (the largest colony after visual assessment), and growth rate as well as the MIC_Cip and MIC_Rif were determined for the evolved isolates. Isolates from the majority of cultures (11 of 15) displayed an improved growth rate after 100 generations. The remaining cultures were evolved for 200 generations before isolates with improved growth rate were identified (Table 1). The initial fitness costs imposed by the Cip^RrpoB mutations was reduced on average by 52% (23%–100%) compared with the isogenic parental strain without the mutation (CH2133). With one exception (CH8879), the MIC_Cip and MIC_Rif decreased in concert relative to the unevolved parental strains. Compared with the isogenic strain without rpoB mutation (CH2133), the MIC_Cip remained elevated (≥1 mg/L) in 9 out of 15 isolates and the MIC_Rif remained elevated (≥16 mg/L) in 6 out of the 15 isolates (Table 1 and Table S3). Most notable were two isolates (CH8890 and CH8879) where the fitness cost was reduced by 46% and 39%, respectively, without a change in MIC_Cip and two isolates (CH8875 and CH8878) that fully restored fitness while maintaining an elevated MIC_Cip (Table 1 and Table S3).

The genomes of the evolved strains were sequenced revealing that all had acquired at least one mutation. Most of the evolved strains (11 of 15) carried only a single significant genetic change (amino acid substitutions or promoter mutations). Seven of these 11 isolates differed only in a single nucleotide from the unevolved parent strains, in each case leading to an amino acid change in a protein-coding sequence. Three of the 11 isolates differed in two nucleotides from the unevolved parents with only one of the mutations causing an amino acid substitution and the other mutation causing a synonymous codon change (2×), or being located in an intergenic region that is not part of any known promoter features (1×). Finally, the last isolate contained a 46 nucleotides-long deletion in the marR promoter region accompanied by a synonymous codon change (Table S2).

Overall, only five genes (rpoA, rpoB, rpoC, dksA and marA) were mutated in multiple strains and each of the 15 strains had a mutation in at least one of these five genes (Table 1). These data indicate these five genes as potential targets for compensatory mutations that reduce the fitness cost of carrying Cip^RrpoB mutations. Mutations in four genes (rpoA, rpoB, rpoC and dksA) were isolated in strains that carried no other genetic change showing these specific mutations are both necessary and sufficient to compensate the cost of the respective Cip^RrpoB mutations in these strains (Table S2).

Evaluation of putative fitness-compensatory mutations

Four mutations affected marA of which one deleted the entire marA gene and another deleted the marA promoter (Table 1). Inactivation mutations of MarA have recently been shown to be selected to reduce the fitness cost imposed by mutations in marR, which cause overexpression of marA.³³ Therefore, the selection of marA mutations in this study is most likely due to the presence of a marR S65fs mutation in genetic background rather than the specific cost imposed by the Cip^RrpoB mutations. Second-site mutations in RNAP genes (rpoA, rpoB and rpoC) were previously shown to reduce the fitness costs imposed by rpoB mutations that increase resistance to rifampicin.^19,34,35 More than half (6 of 10) of the mutations identified in this study are identical (3×) or alter the same amino acid (3×) as mutations that compensate the fitness cost of Rif^RrpoB mutations. These mutations have previously been characterized and are likely to represent a general mechanism to compensate the fitness cost imposed by rpoB mutations (Rif^R and Cip^R) rather than the specific cost caused by Cip^RrpoB mutations.^19,34,35 Therefore, further analysis was focused on the novel class of mutations selected in dksA. Faster growing isolates can be selected even for WT E. coli MG1655, usually as a result of increased uptake of amino acids or sugars from the growth medium.³⁶ To ensure that mutations in dksA are not generally selected to improve growth even in the absence of Cip^RrpoB mutations, we evolved three cultures of the isogenic parental strain (CH2133) for 200 generations and sequenced the dksA genes of the resulting evolved isolates. None of the three isolates carried a mutation in dksA, indicating that the identified mutations are specific for strains with Cip^RrpoB alleles. Twenty additional lineages of the strain containing rpoB Δ442–445 (the strain where three of the four dksA mutations were selected during the evolution experiment) were evolved for 100 generations and the dksA gene of a single isolate per lineage was sequenced to identify additional independent compensatory dksA mutations. Mutations in dksA were identified in 18 of the 20 isolates, increasing the number of independent dksA mutations to 22 (Figure 1 and Table S4). The majority (20 of 22) of mutations were amino acid substitutions affecting four different residues in DksA: D71 (15×), L95 (1×), A132 (1×) and D137 (3×). These mutations in dksA are localized in the coiled-coil tip (D71) that interacts with the catalytic site of the RNAP, or in amino acids that are part of the ppGpp-binding pocket (L95, A132 and D137) (Figure 1b). The remaining two mutations were located upstream of the coding sequence in the dksA regulatory region,³⁷ one mutation (dksA nt G–10A) in the ribosomal-binding site (RBS) and one (dksA nt T–69G) in the extended –10 element (ext –10) (Figure 1c).

Figure 1.

Overview of dksA mutations. (a) Structure of RNAP (grey) in complex with RpoD (yellow), DksA (turquoise) and ppGpp (red) (PDB code 5VSW³¹). The RpoB Δ442–445 mutation is indicated in green and DksA mutations are shown in purple. (b) Close-up view of DksA (turquoise) and ppGpp (red). DksA side chains that interact with ppGpp are shown in black and mutated residues in purple. (c) Overview over the dksA promoter.³⁷ The dksA start codon is indicated in red, the ribosome- and RNAP-binding sites are shown below the sequence, and the transcriptional start site is indicated by a black arrow over the sequence. Promoter mutations isolated during the evolution experiment are indicated above the sequence.

Mutations in dksA increase fitness of strains carrying each of the five rpoB mutations

During the evolution experiment, dksA mutations were only isolated in strains carrying the rpoB mutations located in cluster I (rpoB Δ442–445 and S455dup). The absence of dksA mutations selected in isolates with rpoB mutations in cluster II (rpoB E1272G, A1277V and E1279G) could be due to chance (e.g. compensatory mutations in RNAP genes are more frequent or have a larger effect than dksA mutations), or indicate that the underlying mechanism of the rpoB mutations in cluster I differs from those in cluster II. In the latter hypothesis, dksA mutations might not have an effect on the rpoB mutations in cluster II. Three dksA mutations (dksA nt T–69G, nt G–10A and L95P) representing the three different mutation types (ext –10, RBS and protein alteration) were moved into each of the five unevolved parental strains carrying Cip^RrpoB mutations, the isogenic parental strain with WT rpoB, and a ciprofloxacin-susceptible WT strain (Table 2). The dksA mutations were either neutral or imposed a small fitness cost in the parental strain and the WT. In contrast, all three dksA mutations had a positive effect on the fitness of strains carrying each of the five different Cip^RrpoB mutations. The improvement in fitness for strains carrying Cip^RrpoB mutations in cluster I (mean improvement 24%) was slightly larger than for cluster II (mean improvement 16%) (Table 2, Table S5 and Figure S3). These data show that the fitness cost imposed by each of five Cip^RrpoB mutations can be ameliorated by mutations affecting dksA. Accordingly, differences in the frequency of dksA mutations selected in the different Cip^RrpoB backgrounds do not indicate mechanistic differences between the various Cip^RrpoB mutations.

Table 2.

Effects of dksA mutations in different Cip^RrpoB backgrounds

Strain	Relevant genotype			Relative fitness ± SDb,c	CIP MIC (mg/L)
	CIPa	rpoB	other
CH1464	–	–	–	1.00±0.02	0.016
CH9168	–	–	dksA nt T–69G	1.01±0.00n.s.	0.016
CH9169	–	–	dksA nt G–10A	0.96±0.01*	0.016
CH9285	–	–	dksA L95P	1.00±0.02n.s.	0.016
CH2133	CipR	–	–	0.91±0.02	0.5
CH9171	CipR	–	dksA nt T–69G	0.92±0.01n.s.	0.5
CH9172	CipR	–	dksA nt G–10A	0.87±0.04n.s.	0.5
CH9286	CipR	–	dksA L95P	0.88±0.01n.s.	0.5
CH4959	CipR	Δ442–445	–	0.63±0.01	2
CH9174	CipR	Δ442–445	dksA nt T–69G	0.73±0.00***	1
CH9175	CipR	Δ442–445	dksA nt G–10A	0.68±0.01**	1
CH9287	CipR	Δ442–445	dksA L95P	0.76±0.01***	1
CH3141	CipR	S455dup	–	0.52±0.01	2
CH9177	CipR	S455dup	dksA nt T–69G	0.55±0.00**	1
CH9178	CipR	S455dup	dksA nt G–10A	0.59±0.00***	1
CH9288	CipR	S455dup	dksA L95P	0.58±0.01***	1
CH3144	CipR	E1272G	–	0.67±0.02	1
CH9180	CipR	E1272G	dksA nt T–69G	0.71±0.01*	1
CH9181	CipR	E1272G	dksA nt G–10A	0.72±0.00*	1
CH9289	CipR	E1272G	dksA L95P	0.72±0.01*	1
CH2332	CipR	A1277V	–	0.73±0.00	2
CH9183	CipR	A1277V	dksA nt T–69G	0.76±0.01**	2
CH9184	CipR	A1277V	dksA nt G–10A	0.77±0.00***	2
CH9290	CipR	A1277V	dksA L95P	0.78±0.01***	1
CH3073	CipR	E1279G	–	0.55±0.00	2
CH9186	CipR	E1279G	dksA nt T–69G	0.57±0.03n.s.	2
CH9187	CipR	E1279G	dksA nt G–10A	0.58±0.01**	2
CH9291	CipR	E1279G	dksA L95P	0.56±0.01n.s.	2

CIP, ciprofloxacin.

^aCip^R: gyrA D87G, gyrB S464A, marR S65fs.

^bFitness ± SD relative to the WT.

^cSignificance compared with the respective isogenic parental strain (dksA^WT) was calculated using a two-tailed unpaired t-test (

^n.s, non-significant;

^*, P ≤ 0.05;

^**, P ≤ 0.01;

^***, P ≤ 0.001).

The decrease of ciprofloxacin susceptibility caused by Cip^R rpoB mutations is partly dependent on DksA but independent of ppGpp

The identification of fitness-compensatory mutations in dksA indicates that the mechanism responsible for transcriptional reprogramming is related to the stringent response. Under starvation conditions, the cellular concentration of the alarmone ppGpp is increased by RelA and/or SpoT. Binding of ppGpp to the RNAP increases its affinity for the transcription factor DksA and the resulting formation of an RNAP–DksA–ppGpp complex leads to a shift in the cellular transcriptional pattern referred to as the stringent response (Figure 2a).³⁸ The significance of selecting compensatory mutations in dksA is that it indicates that the Cip^RrpoB mutations induce a stringent-like response in the absence of a starvation signal, as has been described for similar mutations in the RNAP genes.³⁹ Theoretically, the Cip^RrpoB mutations could affect transcriptional reprogramming in different ways: (i) render RNAP hypersensitive to ppGpp, leading to RNAP–DksA–ppGpp complex formation at basal ppGpp concentrations; (ii) create a conformational change in RNAP causing ppGpp-independent binding of DksA; or (iii) change the conformation of the RNAP such that a stringent-like response is activated independently of ppGpp and DskA (Figure 2b). To distinguish between these possibilities, we measured in each strain the effects on ciprofloxacin susceptibility of the absence of DksA (ΔdksA) or ppGpp (ppGpp⁰, ΔrelA/ΔspoT) (Table 3). Removal of DksA or ppGpp had no effect on the MIC_Cip for either the WT strain or the parental strain (gyrA D87G, gyrB S464A, marR S65fs). Similarly, the MIC_Cip for strains carrying each of the five Cip^RrpoB alleles was unchanged in a ppGpp⁰ background indicating that induction of the stringent-like response by these mutations is fully ppGpp independent. In contrast, deletion of dksA reduced the MIC_Cip for strains carrying four of the five rpoB alleles (from 2 to 1 mg/L) corresponding to the loss of half of the resistance caused by the Cip^RrpoB alleles (MIC_Cip for the parental strain is 0.5 mg/L) (Table 3). The only rpoB allele unaffected by the dksA deletion (rpoB E1272G) also caused a smaller increase in MIC_Cip (from 0.5 to 1 mg/L) than the other alleles. A MIC measurement may be too crude to test the effect of the dksA deletion on this particular rpoB allele. Therefore, we performed competition experiments in the presence of various concentrations of ciprofloxacin to test whether deletion of the dksA gene also affects the rpoB E1272G allele. A strain that carries the ciprofloxacin resistance target and efflux mutations (gyrA D87G, gyrB S464A, marR S65fs) was competed against isogenic strains with: (i) the rpoB E1272G allele, and (ii) the rpoB E1272G allele in combination with the dksA deletion (Figure S4). The MSC of ciprofloxacin (MSC_Cip) was determined for each competition and displayed a 1.6-fold increase when the dksA gene was deleted (rpoB E1279G: 0.36 mg/L; rpoB E1272G, ΔdksA: 0.57 mg/L). A higher MSC generally indicates an increased drug susceptibility. This result indicates that also the rpoB E1272G allele displays an increased ciprofloxacin susceptibility when the dksA gene is deleted. Taken together, these results show that approximately half of the increase in MIC_Cip associated with the Cip^RrpoB alleles is caused by binding of DksA to RNAP (ppGpp-independent activation of a stringent-like response) while the other half is caused by a change in the RNAP transcription pattern (ppGpp- and DksA-independent activation of a stringent-like response).

Figure 2.

Schematic view of stringent response activation. (a) Upon starvation conditions RelA and/or SpoT cause the accumulation of ppGpp leading to the binding of ppGpp and DksA to the RNAP. The RNAP–DksA–ppGpp complex leads to a shift in the cellular transcriptional pattern. (b) Activation of a stringent-like response at basal ppGpp concentrations. Mutations in rpoB could (i) render the RNAP hypersensitive to ppGpp binding, (ii) lead to ppGpp-independent binding of DksA, or (iii) cause ppGpp- and DskA-independent activation of a stringent-like response.

Table 3.

Effects of Cip^RrpoB alleles in ΔdksA and ppGpp⁰ backgrounds

Strain	Relevant genotype			CIP MIC (mg/L)
	CIPa	rpoB	other
CH1464	–	–	–	0.016
CH9157	–	–	ΔdksA	0.016
CH9217	–	–	ΔrelA, ΔspoT	0.016
CH2147	–	Δ442–445	–	0.048
CH9748	–	Δ442–445	ΔdksA	0.016
CH2379	–	E1279G	–	0.048
CH9750	–	E1279G	ΔdksA	0.032
CH2133	CipR	–	–	0.5
CH9201	CipR	–	ΔdksA	0.5
CH9230	CipR	–	ΔrelA, ΔspoT	0.5
CH4959	CipR	Δ442–445	–	2
CH9203	CipR	Δ442–445	ΔdksA	1
CH9231	CipR	Δ442–445	ΔrelA, ΔspoT	2
CH3141	CipR	S455dup	–	2
CH9205	CipR	S455dup	ΔdksA	1
CH9232	CipR	S455dup	ΔrelA, ΔspoT	2
CH3144	CipR	E1272G	–	1
CH9207	CipR	E1272G	ΔdksA	1
CH9233	CipR	E1272G	ΔrelA, ΔspoT	1
CH2332	CipR	A1277V	–	2
CH9209	CipR	A1277V	ΔdksA	1
CH9234	CipR	A1277V	ΔrelA, ΔspoT	2
CH3073	CipR	E1279G	–	2
CH9211	CipR	E1279G	ΔdksA	1
CH9235	CipR	E1279G	ΔrelA, ΔspoT	2

CIP, ciprofloxacin.

^aCip^R: gyrA D87G, gyrB S464A, marR S65fs.

^bFitness ± SD relative to the WT.

Activation of the stringent-like response is independent from the strain background

It has previously been shown that DNA gyrase mutations can alter the global patterns of gene expression due to changes in the DNA topology.⁴⁰ Therefore, it is possible that the induction of the stringent-like response by the Cip^RrpoB mutations is dependent on the ciprofloxacin resistance mutations present in the strain background (gyrA D87G, gyrB S464A, marR S65fs). To address this possibility, we selected one Cip^RrpoB mutation from each cluster (cluster I: rpoB Δ442–445, cluster II: rpoB E1279G) and measured the MIC_Cip for strains that carry the Cip^RrpoB mutations but no other ciprofloxacin resistance mutations. The presence of either rpoB mutations led to a 3-fold increase of the MIC_Cip (from 0.016 to 0.048 mg/L), which is comparable to the 4-fold MIC_Cip increase seen in strains with the ciprofloxacin resistance mutations in the strain background (Table 3). As expected, deletion of the dksA gene increased susceptibility to ciprofloxacin to 0.016–0.032 mg/L (Table 3). We conclude that the activation of the stringent-like response by the rpoB mutations is independent of ciprofloxacin resistance mutations in the strain background. This agrees with the finding that RNAP mutations are selected in combination with many different mutations in gyrA (D82N, ΔS83, S83L, D87G, D87Y), gyrB (S464A, V467E, ΔC476), parC (G78D, S80R) and parE (Q428E).²⁶

Discussion

Here, we have shown that there is a class of mutations in RNAP that reduce susceptibility to ciprofloxacin by ppGpp-independent activation of a stringent-like response. Approximately half of the effect involves ppGpp-independent binding of DksA to the RNAP, but the other half is independent of both ppGpp and DksA. These results concur with those from a previous study showing that mutations in aminoacyl-tRNA synthetases could decrease antibiotic susceptibility by activating the stringent response.¹⁵ Activation of the stringent response was shown to alter the expression of multiple genes related to antibiotic resistance (e.g. mdtK, acrZ and ompF) thus leading to a net benefit in the presence of multiple antibiotics including fluoroquinolones, rifampicin, chloramphenicol, β-lactams and trimethoprim.^14,15 Recent selections for reduced susceptibility to ciprofloxacin that we have performed have revealed a large number of genes where mutations could potentially exert their effects by activating a stringent-like response in E. coli (Figure 3, Table S1 and Table S6).^14,15 Examples include mutations that delete tRNA genes and mutations affecting tRNA-modification enzymes or aminoacyl-tRNA synthetases that could plausibly reduce the flow of correctly charged aminoacyl-tRNAs into the ribosome, leading to induction of the stringent response. This mechanism has been shown for mutations in tRNA synthetase genes encoding LeuS and AspS.¹⁵ Similarly, mutations identified in ribosomal proteins, ribosome-modification enzymes and translations elongation factors could also potentially lead to increased binding of uncharged tRNAs to the ribosome and directly or indirectly increase the production of ppGpp. Finally, mutations in RNAP genes or transcription factors could activate a stringent-like response in a ppGpp-independent fashion as shown for the Cip^RrpoB mutations in this study (Figure 3, Table S1 and Table S6). We estimate that there are at least 100 distinct mutations in the RNAP genes that could cause this phenotype, which is in the same range as the number of RNAP mutations that can give rise to rifampicin resistance.²⁰ The mutation rate for rifampicin resistance is around 10⁻⁸–10⁻⁹ per cell per generation,^41,42 indicating that the mutation rate for stringent-like response activating Cip^RrpoB mutations probably has a similarly high value. Adding the additional potential mutational targets affecting transcription or translation (discussed above) that could also induce a stringent-like response (Figure 3, Table S1 and Table S6) will most likely increase the mutation rate by one to two orders of magnitude, moving the class into the same order of mutation rate that is observed for mutations that inactivate efflux pump regulators.¹³ These data suggest that genes where mutations could modify bacterial gene expression patterns via activation of a stringent-like response constitute a very large and diverse genetic target that could contribute to the evolution of resistance to multiple antibiotics.

Figure 3.

Overview of mutations that could cause an activation of a stringent-like response. Deletions of tRNA genes and mutations in tRNA-modification enzymes could plausibly reduce the flow of correctly charged aminoacyl-tRNAs into the ribosome. Mutated ribosomal proteins, ribosome-modification enzymes and elongation factors could alter the interaction of the ribosomes with uncharged tRNAs and/or RelA. Alterations in RNAP genes and transcription factors could lead to an ppGpp-independent activation of a stringent-like response. A detailed list over all mutations can be found in Tables S1 and S4.

The clinical impact of this new class of resistance mutations is hard to estimate. Due to its diversity it is difficult to determine if specific mutations in clinical isolates were selected to increase antibiotic resistance. Additionally, mutations in RNAP and aminoacyl-tRNA synthetase genes have been associated with a reduction of bacterial fitness in the range of 20%–50%, which could potentially reduce their clinical impact.^14,15 Here, we showed for Cip^RrpoB mutations that bacterial fitness can be restored by compensatory mutations while maintaining an elevated resistance level (Table 1) as has previously been shown for rpoB mutations in Rif^RSalmonella^19,34,35 and Mycobacterium tuberculosis^{17,18,43–48} isolates. These results indicate that it is possible to activate a stringent-like response that decreases bacterial susceptibility to various antibiotics without incurring the fitness cost associated with a full activation of the stringent response.

Funding

This work was supported by grants to D.H. from Vetenskapsrådet (The Swedish Research Council) (grant number 2017–03593) and from the Scandinavian Society for Antimicrobial Chemotherapy (grant numbers SLS-693211, SLS-876451). The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Transparency declarations

None to declare.

Supplementary data

Figures S1 to S4 and Tables S1 to S6 are available as Supplementary data at JAC Online.