Nonclonal Emergence of Colistin Resistance Associated with Mutations in the BasRS Two-Component System in Escherichia coli Bloodstream Isolates.

Infections by multidrug-resistant Gram-negative bacteria are increasingly common, prompting the renewed interest in the use of colistin. Colistin specifically targets Gram-negative bacteria by interacting with the anionic lipid A moieties of lipopolysaccharides, leading to membrane destabilization and cell death. Here, we aimed to uncover the mechanisms of colistin resistance in nine colistin-resistant Escherichia coli strains and one Escherichia albertii strain. These were the only colistin-resistant strains of 1,140 bloodstream Escherichia isolates collected in a tertiary hospital over a 10-year period (2006 to 2015). Core-genome phylogenetic analysis showed that each patient was colonized by a unique strain, suggesting that colistin resistance was acquired independently in each strain. All colistin-resistant strains had lipid A that was modified with phosphoethanolamine. In addition, two E. coli strains had hepta-acylated lipid A species, containing an additional palmitate compared to the canonical hexa-acylated E. coli lipid A. One E. coli strain carried the mobile colistin resistance (mcr) gene mcr-1.1 on an IncX4-type plasmid. Through construction of chromosomal transgene integration mutants, we experimentally determined that mutations in basRS, encoding a two-component signal transduction system, contributed to colistin resistance in four strains. We confirmed these observations by reversing the mutations in basRS to the sequences found in reference strains, resulting in loss of colistin resistance. While the mcr genes have become a widely studied mechanism of colistin resistance in E. coli, sequence variation in basRS is another, potentially more prevalent but relatively underexplored, cause of colistin resistance in this important nosocomial pathogen.IMPORTANCE Multidrug resistance among Gram-negative bacteria has led to the use of colistin as a last-resort drug. The cationic colistin kills Gram-negative bacteria through electrostatic interaction with the anionic lipid A moiety of lipopolysaccharides. Due to increased use in clinical and agricultural settings, colistin resistance has recently started to emerge. In this study, we used a combination of whole-genome sequence analysis and experimental validation to characterize the mechanisms through which Escherichia coli strains from bloodstream infections can develop colistin resistance. We found no evidence of direct transfer of colistin-resistant isolates between patients. The lipid A of all isolates was modified by the addition of phosphoethanolamine. In four isolates, colistin resistance was experimentally verified to be caused by mutations in the basRS genes, encoding a two-component regulatory system. Our data show that chromosomal mutations are an important cause of colistin resistance among clinical E. coli isolates.


RESULTS
Low prevalence of colistin resistance in invasive Escherichia bloodstream isolates. A total of 1,140 bloodstream isolates (collected from January 2006 to December 2015) for which species identification and automated antibiotic susceptibility testing had previously been performed, were available for this study. Twelve isolates were deemed resistant to colistin through routine diagnostic procedures. Two of those isolates were isolated from the same patient, on the same day, and were thus considered duplicates, and only one of these was included in this study. In 10 of the 11 remaining isolates, colistin resistance, defined as an MIC of Ͼ2 g/ml colistin, was confirmed through broth microdilution (Table 1). Strain A783 was a false positive for colistin resistance during automated susceptibility testing in routine diagnostic procedures and was excluded from subsequent analyses, leaving ten isolates for further investigation.
The estimated prevalence of colistin resistance in E. coli strains causing bloodstream infections isolated from January 2006 to December 2015 was thus determined to be 0.88%. Three patients had received colistin in the 3 months before isolation of the colistin-resistant strain ( Table 1). Two of these patients received colistin to treat infections, but all three patients were also administered colistin as part of selective digestive or oropharyngeal decontamination (SDD/SOD), a prophylactic antibiotic treatment widely used in Dutch intensive care units (33). The ten colistin-resistant strains were analyzed further in this study to determine their relatedness and mechanism through which they had developed colistin resistance.
Colistin resistance was independently acquired by each individual bloodstream E. coli isolate. To assess the phylogenetic relationships between the colistinresistant strains, a phylogenetic tree was generated based on the genome assemblies of the colistin-resistant strains and 210 publicly available complete genome sequences (see Table S1 in the supplemental material). Based on a core genome alignment of 874 kbp, we did not observe direct transmission of colistin-resistant strains between patients (Fig. 1A). Three colistin-resistant strains (strains I1121, H2129, and G821) belonged to the globally disseminated ST131 clone, and all three were dispersed throughout the multidrug-resistant clade C of ST131 ( Fig. 1A and B) (3,34). This indicates that the ST131 strains in this study have independently acquired colistin resistance. Strain A2361 clustered among Escherichia albertii (Fig. 1A), although it had been typed as E. coli in routine diagnostic procedures.
By screening for acquired antibiotic resistance genes through ResFinder 3.2, we found that only strain E3090 carried the mcr gene mcr-1.1 (0.086% of all bloodstream isolates) (Fig. 1C). After long-read sequencing and hybrid assembly, the mcr-1.1 gene in this strain appeared to be located as the sole antibiotic resistance gene on a 32.7-kbp IncX4-type plasmid. This mcr-1.1-carrying IncX4-type plasmid from E3090 shares 99% identity to the previously reported mcr-1.1-carrying IncX4-type plasmid pMCR-1_Msc  I1121  16  131  22 April 2015  Yes; inhalation and oral  H2129  8  131  22 July 2014  No  G821  16  131  19 March 2013  No  F2745  4  73  2 November 2012  No  E3090  8  10  12 November 2011  No  E2372  4  59  25 August 2011  No  E650  8  162  11 March 2011  No  D2373  8  6901  20 October 2010  Yes; oral  A2361  8  5268  3 November 2007  No  Z821  4  (GenBank accession MK172815.1) harbored by E. coli isolated from patients in Russia (35), confirming the global dissemination of this plasmid (36). In all strains studied here, a variety of acquired resistance genes was observed (Fig. 1C), reflecting the nonclonal nature of the colistin-resistant strains. The three colistin-resistant ST131 strains possessed different repertoires of acquired resistance genes, further excluding recent transmission between patients of the ST131 strains studied here. Strains F2745 and E2372 carried only one and two resistance genes, respectively, while the E. albertii strain A2361 did not possess any acquired resistance genes. Escherichia isolates exclusively acquire colistin resistance by modification of phosphate groups of lipid A. To determine which modifications to lipid A are affecting colistin resistance in E. coli, we extracted lipid A from the clinical strains and the colistin-susceptible control E. coli strain MG1655 and subjected them to MALDI-TOF mass spectrometry. The lipid A produced by all E. coli strains showed lipid A species with a mass-to-charge ratio (m/z) of 1,797 (Fig. 2a), corresponding to the canonical unmodified E. coli hexa-acylated lipid A (Fig. 2b). Colistin-resistant strains showed additional lipid A species at m/z 1,921, consistent with the addition of phosphoethanolamine (m/z 124) to the hexa-acylated species. Additional species were detected in the lipid A produced by strains E650 and Z821. The species at m/z 2,036 indicated the addition of palmitate (m/z 239) to the hexa-acylated species at m/z 1,797, whereas the species at m/z 2,160 was consistent with the addition of palmitate to the hexa-acylated lipid A species containing phosphoethanolamine (m/z 1,910).
The E. albertii strain A2361 produced lipid A distinct from that by E. coli. The species at m/z 1,825 is likely to represent a hexa-acylated species corresponding to two glucosamines, two phosphates, four 3-OH-C 14 , and two C 14 (Fig. 2c). The species at m/z 1,948 is consistent with the addition of phosphoethanolamine to the hexa-acylated species, with a further addition of palmitate to produce lipid A species at m/z 2,187. Species at m/z 1,868 and m/z 2,107 could correspond to the loss of the second phosphate group, compared to those at m/z 1,948 and m/z 2,187.
Identification of mutations in basRS as candidate mutations involved in colistin resistance. Because chromosomal mutations in basRS, but not in other regulatory systems, were previously suggested to cause colistin resistance in E. coli (24)(25)(26)(27)(28), we next aimed to establish the contribution of the basRS alleles in the colistin-resistant phenotype of these bloodstream isolates. Due to the multidrug-resistant nature of the clinical isolates (Fig. 1C), we were unable to generate targeted mutations in these strains. Therefore, we made chromosomal transgene insertion mutants of the different basRS alleles in the attTn7 site in the BW25113-derived ΔbasRS strain BW27848 using the Tn7 transposon system. By making chromosomal transgenes insertions rather than using an in trans complementation method, we excluded copy number effects by plasmids and the need to use antibiotics to select for the presence of a plasmid used for in trans complementation. Since BW27848 still possesses the gene encoding the phosphoethanolamine transferase EptA, we constructed sequences that consisted of the fused sequences of the promoter region of the eptA-basR-basS operon and the basRS coding sequences in order to prevent eptA gene dosage-dependent effects. We were unable to generate the construct for strain E650, presumably due to the toxicity of the insert.
The colistin MIC determination of the generated basRS chromosomal transgene insertion mutants from strains I1121, H2129, G821, and Z821 had higher colistin MIC values than the BW27848::Tn7-empty strain, with observed MIC values Ն16-fold higher than that of the BW27848::Tn7-empty strain ( Table 2). As expected, the basRS allele of the mcr-1.1-positive strain E3090 did not lead to colistin resistance. We were unable to show the contribution of basRS to colistin resistance in the additional four colistinresistant strains (F2745, E2372, D2373, and A2361) that lacked mcr-1.1.
Mutations in the basRS genes contribute to colistin resistance in E. coli. By construction of the chromosomal transgene insertion mutants, we identified the ability of the basRS sequences of four strains (I1121, H212, G821, and Z821) to cause colistin resistance in BW27848. To identify the mutations in the basRS alleles of these strains that contribute to resistance, we compared the basRS sequences of those strains causing resistance to the phylogenetically most closely related publicly available E. coli genome sequences used in the construction of Fig. 1A. None of these reference strains were reported to be colistin resistant or carried any of the mcr genes. This comparison revealed four distinct mutations: an L10R substitution in BasS in I1121, a G53S substitution in BasR in H2192, the duplication of the HAMP-domain in BasS in G821, and an A159P substitution in BasS in Z821 (Fig. 3). As expected, in the mcr-1.1-positive strain E3090, no mutations in basRS were identified.
We hypothesized that the observed mutations were impacting the normal functioning of the BasRS two-component system. To assess whether the mutations in basRS identified by comparing the basRS sequences of the clinical strains I1121, H2129, G821, and Z82, and their closest match in the set of 178 publicly available E. coli genome sequences (Fig. 3) were causal to the development of colistin resistance, the identified mutations were reversed through site-directed inverse PCR mutagenesis to match the  basRS alleles of the publicly available genome sequences. The MIC values of these mutants returned to levels similar to that of the colistin-susceptible BW27848::Tn7empty strain (Table 2). These experiments support the involvement of basRS sequence variation in colistin resistance in E. coli.

DISCUSSION
In the present study, we set out to characterize the mechanisms through which E. coli bloodstream isolates can develop colistin resistance through a combination of whole-genome sequence analysis and experimental validation. We did not find evidence for transfer of colistin-resistant strains between patients, suggesting that colistin resistance was acquired independently in all cases. In seven patients, colistin-resistant strains were isolated without the patients being previously exposed to the drug. All colistin-resistant strains had LPS that was modified by the addition of phosphoethanolamine to the lipid A moiety of LPS. Resistance in one of the bloodstream isolates could be explained by the acquisition of mcr-1.1. In four other strains, we identified mutations in basRS that contribute to colistin resistance. Although colistin-susceptible strains that were isogenic to the resistant strains were not available, we were able to pinpoint the mutations in basRS leading to resistance in these strains by matching the genomic sequences of our nosocomial isolates with publicly available genomes, none of which were reported to be colistin resistant, and subsequent construction of chromosomally integrated basRS transgene alleles in the ΔbasRS strain BW27848. The mechanisms of colistin resistance in the remaining five strains remain to be characterized.
Some of the mutations we experimentally link to colistin resistance in this study have previously been associated with colistin resistance or the functioning of the BasRS two-component system. In this study, we demonstrated that the amino acid change L10R in BasS (strain I1121) also confers colistin resistance. An amino acid substitution in the same position of BasS (L10P) was previously experimentally proven to cause colistin resistance in E. coli (26). The glycine in position 53 of BasR was previously reported to be altered in colistin-resistant Enterobacteriaceae (37,38), including in E. coli (39). The G53S change specifically, as in isolate H2192, has been experimentally proven to contribute to colistin resistance in Klebsiella (previously Enterobacter) aerogenes (40,41) and Salmonella enterica subsp. enterica serovar Typhimurium (42), and we extend those findings to E. coli here. The previously unidentified duplication of 162 nucleotides in basS (strain G821) leads to the introduction of a second HAMP domain in BasS and confers colistin resistance in the BW27848 background. The HAMP domain is widespread in bacteria and is commonly involved in signal transduction as part of twocomponent systems (43). We hypothesize that the addition of an extra HAMP domain in BasS may change signal transduction in the protein, leading to the constitutive activation of the histidine kinase domain of BasS, increased phosphorylation of BasR, and upregulated expression of eptA, ultimately resulting in the addition of phosphoethanolamine to lipid A. Finally, we demonstrate that the A159P substitution in BasS (observed in strain Z821) contributes to colistin resistance. A mutation leading to an A159V substitution was found in an in vitro evolution study in which E. coli was evolved toward colistin resistance (44) and in clinical colistin-resistant E. coli isolates (45), but experimental confirmation of the role of alterations in A159 in colistin resistance in E. coli was so far lacking. Our data suggest that the basRS alleles of three E coli strains (F2745, E2372, and D2373) and the E. albertii strain A2361 do not confer resistance in the BW25113 E. coli background. Because E. albertii is phylogenetically distinct from E. coli, its basRS allele may not function optimally in an E. coli background, explaining the inability of the transgene insertion complementation in the basRS deletion of BW25113 E. coli strain to cause colistin resistance (46). We are unable to explain the colistin resistance mechanisms of the clinical isolates F2745, E2372, and D2373. It is likely that these strains have become resistant to colistin through other mutations that finally lead to the modification of lipid A by phosphoethanolamine.
The observed modification of lipid A with phosphoethanolamine in all isolates underlines the crucial role of phosphoethanolamine transferases in the ability of Escherichia to become resistant to polymyxins (14). The lipid A of three of the colistinresistant strains was also modified with palmitate, but the contribution of lipid A palmitoylation to colistin resistance in clinical E. coli strains is currently unknown. We did not observe modifications of lipid A by 4-amino-4-deoxy-L-arabinose in the colistinresistant isolates. While this modification was shown to contribute to polymyxin B resistance under low Mg 2ϩ conditions in a laboratory isolate of E. coli (20), it may be rare in clinical E. coli isolates. Indeed, Sato et al. also exclusively found phosphoethanolamine-modified lipid A in colistin-resistant clinical E. coli isolates (24). The reliance of Escherichia on the modification of lipid A by phosphoethanolamine to acquire colistin resistance, suggests that the inhibition of this class of enzymes by blocking the conserved catalytic site (31) could be a target for future drug development and opens the possibility of combination therapy with colistin and an inhibitor of phosphoethanolamine transferase (47). With the increasing clinical issues posed by infections with multidrug-resistant Gram-negative bacteria, there is an urgent need to better understand resistance mechanisms to last-resort antibiotics such as colistin.
While the discovery of the mcr genes has generated considerable interest in transferable colistin resistance genes, our data suggest that chromosomal mutations remain an important cause of colistin resistance among clinical isolates in the genus Escherichia.

MATERIALS AND METHODS
Ethical statement. Approval to obtain data from patient records was granted by the Medical Ethics Review Committee of the University Medical Center Utrecht, in Utrecht, The Netherlands (project numbers 16/641 and 18/472).
Colistin-resistant E. coli strains were isolated as part of routine diagnostic procedures. This aspect of the study did not require consent or ethical approval by an institutional review board.
Bacterial strains, growth conditions, and chemicals. Colistin-resistant E. coli strains from bloodstream infections were obtained retrospectively from the strain collection of the clinical microbiology laboratory of the University Medical Center Utrecht in Utrecht, The Netherlands. In initial routine diagnostic procedures, blood cultures were plated on tryptic soy agar (TSA) plates with 5% sheep blood. Strains collected up to 2011 were identified and their antibiogram was determined using the BD Phoenix automated identification and susceptibility testing system (Becton, Dickinson, Vianen, The Netherlands). From 2011 onwards, species determination was performed by MALDI-TOF on a Bruker microflex system (Leiderdorp, The Netherlands). E. coli strain BW25113 and the BW25113-derived ΔbasRS strain BW27848 from the Keio collection were obtained from the Coli Genetic Stock Center (48,49). Strains were grown in lysogeny broth (LB; Oxoid, Landsmeer, The Netherlands) at 37°C with agitation at 300 rpm unless otherwise noted, with exception of strains containing pGRG36, which were grown at 30°C (50). When appropriate, kanamycin (50 mg/liter; Sigma-Aldrich, Zwijndrecht, The Netherlands) and ampicillin (100 mg/liter; Sigma-Aldrich) were used. Colistin sulfate was obtained from Duchefa Biochemie (Haarlem, The Netherlands). L-(ϩ)-Arabinose was obtained from Sigma-Aldrich. Plasmids were purified using the GeneJET Plasmid Miniprep kit (Thermo Fisher Scientific, Landsmeer, The Netherlands). PCR products were purified from gels using GeneJET Gel Extraction and DNA Cleanup Micro kits (Thermo Fisher Scientific).
Determination of MIC. MICs to colistin were determined as previously described (51) For strain E3090, we performed long-read sequencing using the MinION platform (Oxford Nanopore Technologies) to fully resolve the mcr-1.1 plasmid. MinION library preparation for barcoded twodimensional (2D) long-read sequencing was performed using the SQK-LSK208 kit (Oxford Nanopore Technologies, Oxford, United Kingdom), according to the manufacturer's instructions, with G-tube (Covaris, Woburn, MA, USA) shearing of 1 g chromosomal DNA for 2 ϫ 120 s at 1,500 ϫ g. Sequencing was performed on the MinION sequencer (Oxford Nanopore Technologies) using 2D barcoded sequencing through a SpotON Flow Cell Mk I (R9.4; Oxford Nanopore Technologies).
MinION sequence read data in FastQ format was extracted from Metrichor base-called raw FAST5 read-files using Poretools (53). A hybrid assembly for strain E3090 was generated with trimmed Illumina short-read data and Oxford Nanopore Technologies MinION long-read data by using SPAdes v3.6.2 with the same settings as the Illumina short-read assemblies and specifying the long-read data with the -nanopore flag.
Core genome phylogenetic analysis and determination of mutations in candidate colistin resistance determinants. Genome assemblies generated in this study with Illumina data were aligned with 178 complete E. coli genomes and 32 E. albertii genomes that were available from NCBI databases on 24 June 2016 (see Table S1 in the supplemental material) using ParSNP v1.2 (57). MEGA6 was used to midpoint root and visualize the phylogenetic tree (58). We identified whether nonsynonymous mutations were present in basRS by pairwise comparison of the gene sequences of colistin-resistant isolates to their closest matching publicly available genome from the phylogenetic tree using BLAST (59). Mutations that were identified in the genome sequences were confirmed through PCR (oligonucleotide primer sequences are provided in Table S2) and subsequent Sanger sequencing of the PCR product by Macrogen (Amsterdam, The Netherlands).
Isolation and analysis of lipid A. Isolation of lipid A molecules and subsequent analysis by negative-ion MALDI-TOF mass spectrometry was performed as previously described (19,60,61). Briefly, Escherichia strains were grown in LB (Oxoid), and the lipid A was purified from stationary cultures using the ammonium hydroxide-isobutyric acid method described earlier (62). Mass spectrometry analyses were performed on a Bruker autoflex speed TOF/TOF mass spectrometer in negative reflective mode with delayed extraction using as matrix equal volumes of dihydroxybenzoic acid matrix (Sigma-Aldrich) dissolved in (1:2) acetonitrile-0.1% trifluoroacetic acid. The ion-accelerating voltage was set at 20 kV. Each spectrum was an average of 300 shots. A peptide calibration standard (Bruker) was used to calibrate the MALDI-TOF. Further calibration for lipid A analysis was performed externally using lipid A extracted from E. coli strain MG1655 grown in LB medium at 37°C.
Construction of chromosomal basRS transgene insertions. Chromosomal transgene insertions of basRS were constructed in BW27848 by utilizing the Tn7 transposon system on the pGRG36 plasmid (50). The promoter of the eptA-basRS operon was fused to the basRS coding sequence by separate PCRs for the promoter region and the basRS amplicon, with high fidelity Phusion Green Hot Start II DNA polymerase (Thermo Fisher Scientific) using strain-specific primers (Table S2) (oligonucleotides were obtained from Integrated DNA Technologies, Leuven, Belgium). The promoter and the basRS amplicon were subsequently fused by overlap PCR. Fused PCR products were cloned into pCR-Blunt II-TOPO using the Zero Blunt TOPO PCR Cloning kit (Thermo Fisher Scientific) and subsequently subcloned into pGRG36 (50). Electrocompetent BW25113 and BW27848 E. coli cells were prepared as described previously (63) and transformed using the following settings: voltage, 1,800 V; capacitance, 25 F; resistance, 200 ⍀; with a 0.2-cm cuvette using the Gene Pulser Xcell electroporation system (Bio-Rad Laboratories, Veenendaal, The Netherlands). Transformants were grown at 30°C. After confirming integration of the Tn7 transposon at the attTn7 site by PCR (primers listed in Table S2) and Sanger sequencing (Macrogen), the pGRG36 plasmid was cleared by culturing at 37°C.
Inverse PCR site-directed mutagenesis was performed on amplicons cloned in pCR-Blunt II-TOPO to reverse the mutations that were identified in colistin-resistant strains to the sequences of basR or basS in the closest matching publicly available genome (64). After gel purification of the amplified fragments, (hemi)methylated fragments were digested using DpnI (New England BioLabs [NEB], Ipswich, MA, USA). Subsequently, the vector was recircularized using the Rapid DNA Ligation kit (Thermo Fisher Scientific) after phosphorylation using T4 polynucleotide kinase (NEB). The constructs were then transformed into chemically competent DH5␣ E. coli cells (Invitrogen, Landsmeer, The Netherlands). Mutated sequences were subsequently subcloned to pGRG36 as described above.
Data availability. Sequence data have been deposited in the European Nucleotide Archive (accession number PRJEB27030).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.