The Plasmid-Mediated Kluyvera-Like arnBCADTEF Operon Confers Colistin (Hetero)Resistance to Escherichia coli

The use of colistin as a last-resort antimicrobial is compromised by the emergence of resistant enterobacteria with acquired determinants like mcr genes, mutations that activate the PmrAB system, and still unknown mechanisms. This work analyzed 74 Escherichia coli isolates from healthy swine, turkey, or bovine, characterizing their colistin resistance determinants.

ABSTRACT

The use of colistin as a last-resort antimicrobial is compromised by the emergence of resistant enterobacteria with acquired determinants like mcr genes, mutations that activate the PmrAB system, or still unknown mechanisms. This work analyzed 74 Escherichia coli isolates from healthy swine, turkey, or bovine, characterizing their colistin resistance determinants. The mcr-1 gene, detected in 69 isolates, was the main determinant found, among which 45% carried this gene on highly mobile plasmids, followed by 4 strains lacking previously known resistance determinants or 2 with mcr-4 (1 in addition to mcr-1), whose phenotypes were not transferred by conjugation. Although a fraction of isolates carrying mcr-1 or mcr-4 genes also presented missense polymorphisms in pmrA or pmrB, constitutive activation of PmrAB was not detected, in contrast to strains with mutations that confer colistin resistance. The expression of mcr genes negatively controls the transcription of the arnBCADTEF operon itself, a downregulation that was also observed in the four isolates lacking known resistance determinants, with three of them sharing the same macrorestriction and plasmid profiles. Genomic sequencing of one of these strains, isolated from a bovine in 2015, revealed an IncFII plasmid of 62.1 kb carrying an extra copy of the arnBCADTEF operon closely related to Kluyvera ascorbata homologs. This element, called pArnT1, was cured by ethidium bromide, and the cells lost resistance to colistin in parallel. Furthermore, a susceptible E. coli strain acquired heteroresistance after transformation with pArnT1 or pBAD24 carrying the Kluyvera-like arnBCADTEF operon, revealing it as a new colistin resistance determinant.

INTRODUCTION

For a long time, it was believed that Gram-negative bacteria could acquire resistance to polymyxins B and E (colistin), two closely related cationic and cyclic antimicrobial peptides, only by spontaneous mutations altering signaling processes that control covalent modifications of lipopolysaccharide (LPS) and reduce its negative charge (1). Polymorphisms that render constitutively active PmrAB or PhoPQ two-component systems (TCSs) lead to the overexpression of two enzymes, EptA (PmrC) and ArnT (PmrK), that transfer phosphoethanolamine (PEtN) and 4-amino-4-deoxy-l-arabinose (l-Ara4N), respectively, to phosphates of the glucosamine disaccharide of lipid A and reduce colistin binding (Fig. 1). Encoded by operons eptA-pmrAB and arnBCADTEF (pmrHFIJKLM), deregulated expression from these genes interferes with cell viability, making colistin-resistant mutants of low prevalence under nonselective conditions and reducing the risk of antimicrobial resistance spread (2). In this context, colistin was used routinely to control intestinal enterobacteria in farm animals, trusting that the unlikely vertical transmission of resistant clones would not compromise its efficacy as an antibiotic of last resort for the control of insidious infections in humans. However, during the last 5 years, colistin resistance has been widely spread among enterobacteria by plasmids carrying different mcr genes originating from plasmid mobilization of eptA orthologs from different species, targeting the encoded transferase activity to the bacterial inner membrane, where its phosphatidylethanolamine substrate is an abundant component of the phospholipid pool (1). Carried by different replicons, mcr genes, among which mcr-1 is the most prevalent, have spread widely among enterobacteria, challenging the clinical utility of polymyxins.

FIG 1.

Expression of colistin resistance determinants in E. coli. The lipid A fraction of LPS and its major modifications by l-Ara4N and PEtN that block the binding of colistin to the outer leaflet of the external membrane (EM) are shown. The TCS PmrAB transduces low-pH and high-Fe³⁺ signals through the plasma membrane (PM) to the cytosol, where it activates the transcription of effector genes. The possible role of the TCS PhoPQ in mediating the activation of PmrAB in E. coli is controversial and might depend on the particular genetic backgrounds (2), whereas the expression of mcr genes downregulates the arnBCADTEF and eptA-pmrAB operons (this work).

In parallel to EptA/Mcr functions, the arnBCADTEF operon and the ugd gene encode the enzymes that synthesize l-Ara4N from UDP-glucose and its transfer to 4′-phosphate lipid A (3) (Fig. 1), involving an eight-step biochemical pathway much more complex than the PEtN addition to lipid A by a single enzyme, probably explaining why the latter is the only known plasmid determinant conferring colistin resistance (1). However, besides polymorphisms of the TCSs, another determinant interfering with signaling is the Enterobacter cloacae ecr gene, whose expression in Escherichia coli upregulated arnBCADTEF and conferred colistin heteroresistance (4).

In previous works, we analyzed 7 colistin-resistant E. coli isolates from healthy turkey, swine, and bovine, finding mcr-1 in all strains, among which 1 carried an additional mcr-3 gene in the same plasmid and 2 presented mutations conferring colistin resistance in pmrA or pmrB (5–7). These mcr-1 genes were detected in plasmids of different sizes, and among them, 42.9% efficiently transferred colistin resistance by conjugation. In the present work, we extend our screening to 74 additional strains, characterizing their colistin resistance determinants, among which mcr-1 is by far the most prevalent. This larger strain collection contains strains lacking previously known colistin resistance determinants, some of which presented a plasmid carrying a Kluyvera-like arnBCADTEF operon that is associated with resistance to the antibiotic, a phenomenon evoking the spread of mcr genes and representing a second transferable class of colistin resistance determinants.

RESULTS

Colistin resistance determinants of E. coli from farm animals.

Colistin-resistant isolates of E. coli were selected around Spain from feces of healthy animals of three major species targeted for food production: swine, poultry (turkey), and bovine (Table 1; see also Table SI in the supplemental material). The mcr-1 gene was found in 69 out of the 74 isolates, followed by mcr-4 in 2 strains, of which 1 also shared mcr-1. Besides mcr-1 or mcr-4 genes, we identified 15 isolates with missense polymorphisms in the pmrAB operon that were not found in E. coli K-12 or ATCC 25922 or in colistin-susceptible strains recently isolated from farm animals (Table SI) (5). The genotypes of 4 isolates lacked any candidates for resistance determinants.

TABLE 1.

Colistin resistance determinants in E. coli from healthy animals^a

Species of origin (no. of isolates)	Yr	mcr allele(s)	Plasmid size(s) (kb)b	Mobc	No. of strains with genotype (no. of strains with PmrAB polymorphism, mutation[s])d
Turkey (27)	2014	1	33	+	11 (1, A360V)
1	33	−	10 (2, A360V)
1	210–240	−	2
1	240	−	1
1	240–310	+	2 (1, M287I)
1	240–310	−	1
Swine (35)	2015	1	33	+	11 (5, A242T; 1, I318S+S323R)
1	33	−	11 (1, R138H)
1	78	−	1
1	210	+	1
1	210	−	2
1	240	−	1
1	240–310	−	1 (1, A360V)
1	C	−	2
1 + 4	33 + <20	−	1
4	<20	−	1 (1, A360V)
−	3
Bovine (5)	2015	1	33	+	3
1	210	−	1
−	1
Turkey (7)	2016	1	33	+	5 (2, A360V)
1	310–340	−	1
1	C	−	1

^aData are pooled from the information available for every analyzed strain (detailed in Table SI in the supplemental material).

^bPlasmid size measured by PFGE-S1 and hybridization with DIG-labeled mcr-1 or mcr-4 probes. The molecular size reference was XbaI-digested S. Braenderup. C, chromosomal location deduced for the mcr-1 gene.

^cMob, mobilization potential (conjugation positive [+] or negative [−] with the J53 receptor strain [detailed in Table SI in the supplemental material]).

^dTotal number of strains with the genotype and mobilization potential, with the number(s) with PmrAB polymorphisms in parentheses.

Mobilization potential of colistin resistance determinants.

S1 nuclease pulsed-field gel electrophoresis (PFGE-S1) and hybridization to a specific probe showed that 66 out of the 69 mcr-1 genes detected in this work are located in plasmids whose sizes are near 33 kb, 78 kb, or >200 kb (Table 1 and Table SI), fitting almost unambiguously the ranges covered by IncX4, IncI2, or IncHI2 replicons, respectively, that carried more than 90% mcr-1 genes in previously analyzed enterobacteria (8). Besides, the two mcr-4 genes hybridized to a plasmid of <20 kb that would correspond to the low-molecular-weight (8.7-kb) ColE1 replicon previously linked to this resistance determinant (9).

Conjugation assays showed that 47.8% of mcr-1 determinants were carried by mobilizable plasmids, although like the four strains lacking known determinants of colistin resistance, none the three strains that share chromosomally located mcr-1 genes, nor the two ColE10-like plasmids with mcr-4 or the plasmid carrying mcr-1 in the size range of an IncI2 replicon, transferred their phenotype to the recipient strain (Table 1 and Table SI). The efficiency in mobilizing colistin resistance was much higher for IncX4-like (57.7%) than for IncHI2-like (7.2%) plasmids, the two major plasmid categories detected, in 52 and 13 isolates, respectively.

Signaling of colistin resistance upon the expression of the arnBCADTEF and eptA-pmrAB operons.

We measured the mRNA expression of the arnBCADTEF and eptA-pmrAB operons from isolates representing the different determinants identified, mcr-1 and/or mcr-4 and the five polymorphisms found in PmrAB, R138H in PmrA and A360V, A242T, I318S plus S323R (I318S+S323R), and M287I in PmrB, plus all those for which the resistance determinant(s) remains unknown (Fig. 2). Accumulation of mRNA from both operons was detected by probing arnB and eptA, which were strongly downregulated in strains presenting chromosomal mcr-1, IncX4-like plasmids carrying mcr-1, and/or ColE10-like plasmids carrying mcr-4, independently of all PmrAB polymorphisms detected in this work, in addition to the four isolates lacking known resistance determinants. In contrast, the two previously characterized swine isolates ZTA11/01748EC and ZTA13/02182EC that presented mcr-1 and mutations associated with colistin resistance (PmrB-V161G and PmrA-S39I+R81S, respectively) (5, 6, 10) showed arnB upregulation (Fig. 2).

FIG 2.

Signaling of colistin resistance determinants upon arnBCADTEF and eptA-pmrAB operon expression. Liquid cultures of the indicated strains were incubated until the mid-exponential phase of growth and processed for RNA extraction, cDNA synthesis, and qPCR to quantify arnB and eptA transcript accumulation by using recA as an endogenous calibrator and the E. coli ATCC 25922 strain for the normalization of gene expression, the level of which is indicated by a discontinuous line. Strains and genotypes (mcr genes and PmrAB polymorphisms [see Table SI in the supplemental material) are indicated. 1, ZTA15/02267EB1 (chromosomal mcr-1); 2, ZTA15/00389EB1 (mcr-1); 3, ZTA15/01072COL1 (mcr-1 plus PmrB-A242T); 4, ZTA15/01844EB1 (mcr-1 plus PmrB-A242T); 5, ZTA15/02333EB2 (mcr-1 plus PmrB-A242T); 6, ZTA15/01825EB1 (mcr-1 plus PmrB-A242T); 7, ZTA15/00233EB1 (mcr-1 plus PmrB-A242T); 8, ZTA15/00750EB1 (mcr-4 plus PmrB-A360V); 9, ZTA15/00630EB1 (mcr-1 plus PmrB-A360V); 10, ZTA15/01313EB1 (mcr-1 plus PmrB-I318S+S323R); 11, ZTA14/00808EB (mcr-1 plus PmrB-M287I); 12, ZTA15/01196EB1 (mcr-1 plus PmrA-R138H); 13, ZTA15/00420EB1 (mcr-1 plus mcr-4); 14, ZTA11/01748EC (mcr-1 plus PmrB-V161G [6]); 15, ZTA13/02182EC (mcr-1 plus PmrA-S39I+R81S [6]); 16, ZTA15/00213-1EB1 (unknown); 17, ZTA15/01360EB1 (unknown); 18, ZTA15/00235EB1 (unknown); 19, ZTA15/00702EC (unknown). PmrAB* indicates strains with PmrAB polymorphisms detected in this work, whereas PmrAB^C indicates strains with mutations already known to confer colistin resistance (10). For each strain, the averages from two biological replicates (independently grown), each with three technical replicates (triplicate PCRs), and their standard deviations are shown.

Finding of a Kluyvera-like arnBCADTEF operon in colistin-resistant E. coli.

PFGE analysis revealed that ZTA15/00213-1EB1, ZTA15/01360EB1, and ZTA15/00235EB1, three out of the four remaining isolates without a known determinant for colistin resistance, are closely related and share two plasmids, roughly comigrating with the 55-kb and 105-kb bands of the size standard, whereas the remaining isolate, ZTA15/00702EC, presented a different genetic background (Fig. 3).

FIG 3.

PFGE analysis of strains lacking previously known colistin resistance determinants. (A) XbaI macrorestriction and genomic profiling; (B) nuclease S1 and plasmid profiling; (C) Southern blotting of PFGE-S1 hybridized to a DIG-labeled probe of the Kluyvera-like arnB sequence PCR amplified from strain ZTA15/00213-1EB1. M, molecular weight standard; lane 1, ZTA15/00213-1EB1; lane 2, ZTA15/01360EB1; lane 3, ZTA15/00235EB1; lane 4, ZTA15/00702EC.

Genome sequencing was performed for isolates ZTA15/00213-1EB1 and ZTA15/00702EC, representing the two strains with different PFGE types identified, which were isolated from bovine and porcine, respectively (Table SI). In addition to the search already performed by PCR of mcr genes and sequencing of PmrAB coding sequences, both genomes were screened for the mcr-1 to mcr-10 genes (11) and polymorphisms in the phoPQ, mgrB, or pmrD gene (2), without finding any candidate that might confer colistin resistance, although other antibiotic resistance determinants were detected (Table SIII). However, searching the genes involved in LPS modification and looking for their possible structural alterations, we found a nonnative arnBCADTEF operon of 7,382 bp in isolate ZTA15/00213-1EB1, in addition to its E. coli-like counterpart of 7,351 bp (Table SIV). The foreign arnBCADTEF operon gives the best match (97.7%) by BLASTN to the Kluyvera ascorbata OT2 genome found in the RefSeq genome database (refseq_genomes) available at the NCBI. The identity between arnBCADTEF DNA sequences from K. ascorbata and the Kluyvera-like operon found in ZTA15/00213-1EB1 is above 96% along their seven coding sequences, whereas it is even higher (above 99%) among E. coli-like sequences from its own operon and that of E. coli K-12 or ZTA15/00702EC, the other isolate lacking previously known colistin resistance determinants (Table SIV). The identities between Kluyvera-like and E. coli-like arnBCADTEF sequences decreased to between 53 and 74%, still closely related but far below the hybridization threshold that determines the specificity of primers (see above) and the DNA probe (see below) used in this work. Besides that, the seven coding sequences of the E. coli-like arnBCADTEF operons found in the genomes of the two E. coli isolates are complete and apparently functional (Table SIV).

pArnT1, an IncFII plasmid carrying a Kluyvera-like arnBCADTEF operon.

The Kluyvera-like arnBCADTEF operon from the ZTA15/00213-1EB1 genome has been located in a 62.1-kb plasmid spanning 74 putative coding sequences (Fig. 4), carrying mobilization functions like tra and trb genes and replication control like repA1, a signature of IncFII replicons (12). In addition, it presents four antimicrobial resistance genes, bla_TEM-1, ermB, qacE, and sul1, for a class-A β-lactamase, a 23S RNA methylase, a small multidrug resistance protein, and a dihydropteroate synthase, respectively. Between positions 27630 and 27876 and 20 bp upstream of the arnB start codon, there is a 5′-truncated ISEcp1 element with its right inverted repeat (IRR) toward the 5′ end of the Kluyvera-like arnBCADTEF operon, which presents a 5-bp direct repeat (TTATA) at both ends (positions 27877 to 27881 and 35546 to 35550). This ISEcp1 element, containing promoter sequences TTGAAA (−35) and TACAAT (−10) near its right border, has been shown to be involved in the expression and mobilization of plasmid-borne extended-spectrum β-lactamases encoded by the bla_CTX-M gene family (13). Most of the remaining 55 kb of DNA surrounding the Kluyvera-like operon is closely related to fragments from two previously described replicons, pBH100alpha and p65 (Fig. 4). This plasmid is named pArnT1 here.

FIG 4.

Structure of pArnT1. The plasmid sequence is covered over almost its entire length, with discontinuous regions sharing high identity with E. coli plasmids pBH100alpha (GenBank accession number CP025254.1) and p65 (accession number MT077888.1), in addition to the arnBCADTEF operon sequence from K. ascorbata OT2 (accession number NZ_RHFN01000006.1). TBLASTN identified 74 coding sequences, which are represented by boxes above or below the black line for sense or antisense strands, respectively. Black boxes correspond to truncated sequences, and gray boxes correspond to coding sequences for hypothetical proteins.

The presence of pArnT1 in colistin-resistant strains of E. coli was evidenced by PFGE-S1 and Southern hybridization to a Kluyvera-like arnB probe. Isolates ZTA15/00213-1EB1, ZTA15/00235EB1, and ZTA15/01360EB1 sharing PFGE profiles (Fig. 3A and B) carry pArnT1, the plasmid migrating above 55 kb and specifically hybridizing to the Kluyvera-like arnB probe (Fig. 3C).

pArnT1 stability and Kluyvera-like arnBCADTEF expression are associated with colistin resistance.

The colistin resistance phenotype showed stability differences among the three strains that share pArnT1 (Fig. 5). Whereas isolates ZTA15/00213-1EB1 and ZTA15/00235EB1 steadily preserved colistin resistance and pArnT1 after (up to four complete cycles of) growth in media lacking colistin, ZTA15/01360EB1 lost both traits without selection, and only 1 clone out of 5 grown on colistin media retained the arnBCADTEF Kluyvera-like sequence (Fig. 5A). Accordingly, mRNA expression from the plasmid-carried arnBCADTEF operon was detected in isolates ZTA15/00213-1EB1 and ZTA15/00235EB1 but not in ZTA15/01360EB1 (Fig. 5B), although the growth on colistin from the ZTA15/01360EB1 strain revealed a determinant of resistance (PmrB-V161G) (2) that was not originally present and that might have been selected during in vitro growth (Table SI). Thus, the natural loss of pArnT1 from the ZTA15/01360EB1 isolate was linked to a reduction of colistin resistance, although PmrAB polymorphisms might be acquired upon selection. In addition, plasmid curing of pArnT1 from ZTA15/00213-1EB1 and ZTA15/00235EB1 was performed by culturing bacteria in the presence of subinhibitory concentrations of ethidium bromide. Although one-tenth of the clones of each strain retained the phenotype of resistance to colistin and associated pArnT1, most of the cells cultured without selection lost in parallel the plasmid and phenotype (Fig. 5C), reducing their MIC to ≤0.5 mg/liter of colistin (see below) (Fig. 6).

FIG 5.

Association of the Kluyvera-like arnBCADTEF operon with colistin resistance. (A) The three E. coli isolates sharing the pArnT1 plasmid, ZTA15/00213-1EB1, ZTA15/00235EB1, and ZTA15/01360EB1, were grown overnight with or without antibiotic selection (colistin at 2 mg/liter). Five clones were selected from every strain and condition; their growth in colistin was tested, and after DNA purification, the Kluyvera-like arnB sequence was detected by PCR. Identical results were obtained for isolates ZTA15/00213-1EB1 and ZTA15/00235EB1 (not shown). (B) RNA was isolated from one clone selected on colistin-selective medium from every strain, and after cDNA synthesis, the DNA fragment spanning the arnB gene was amplified by RT-PCR with the primers indicated in Table SII in the supplemental material. Lane 1, ZTA15/00213-1EB1; lane 2, ZTA15/01360EB1; lane 3, ZTA15/00235EB1; lane 4, ZTA15/00702EC; lane 5, E. coli ATCC 25922; lane 6, negative control with water instead of cDNA. (C) The plasmid was cured by growing cells overnight with ethidium bromide (BrEt) (100 mg/liter). Ten colonies were selected on nonselective medium from every strain, their growth on colistin was tested, and after DNA purification, the Kluyvera-like arnB sequence was detected by PCR.

FIG 6.

Heterologous expression of colistin resistance determinants. XL1-Blue MRF′ cells transformed by the indicated plasmids, including pBAD (pBAD24 empty vector), and the ZTA15/00213-1EB isolate and its cured strain ZTA15/00213-1EB^c1 were grown overnight on liquid cultures (Mueller-Hinton broth) supplemented with 100 mg/liter ampicillin, diluted 1/10 with fresh medium, and induced (+) or not (−) with 0.2% arabinose. When the cultures reached an OD₆₀₀ of 1.0, cells were diluted to a 0.5 McFarland standard (OD₆₀₀ of ∼0.1), and 10-μl aliquots were spotted on Mueller-Hinton agar plates containing 0.02% arabinose and colistin at the indicated concentrations and incubated at 37°C overnight. A representative result from two independent experiments is given.

The Kluyvera-like arnBCADTEF operon confers colistin heteroresistance to XL1-Blue MRF′.

The colistin resistance phenotypes of the ZTA15/00213-1EB1 and ZTA15/00235EB1 isolates, presenting stably maintained pArnT1, could not be mobilized by conjugation (see above). To address the heterologous expression of the Kluyvera-like arnBCADTEF operon in a colistin-susceptible background, plasmid DNA from the ZTA15/00213-1EB1 isolate was transferred by electroporation to E. coli XL1-Blue MRF′. Using plasmid-encoded beta-lactamase as a selectable marker on ampicillin-supplemented medium, PCR detected the Kluyvera-like operon in half of the transformed cells (not shown), which therefore carry pArnT1. In parallel, E. coli XL1-Blue MRF′ was transformed with the construct pBAD24::arnBCADTEF, where the expression of the Kluyvera-like operon cloned from isolate ZTA15/00213-1EB1 is under the strict control of arabinose. Both plasmids transformed the colistin susceptibility of XL1-Blue MRF′ to heteroresistance, evidenced as heterogeneous and persistent growth along a range of antibiotic doses spanning at least a 16-fold-higher MIC than the maximal noninhibitory concentration (from 4 to ≤0.25 mg/liter of colistin) (Fig. 6). Furthermore, arabinose is required for heteroresistance by gene expression from pBAD24, showing that the unique determinant producing this phenotype is the Kluyvera-like operon, in contrast to colistin-susceptible XL1-Blue MRF′ carrying native pBAD24; ZTA15/00213-1EB1^c1, a clone where pArnT1 was cured; or strains canonically resistant to colistin, like ZTA15/00213-1EB1 from which pArnT1 was isolated or the previously described pBAD24::mcr-1 construct in XL1-Blue MRF′ (14).

DISCUSSION

The two acquired mechanisms conferring colistin resistance to enterobacteria, chromosomal mutations that constitutively activate PmrAB and plasmid-mobilized mcr genes, promote opposite signaling on mRNA expression from the arnBCADTEF and eptA-pmrAB operons, encoding key enzymes for LPS modification. This work evidences that polymyxin-resistant enterobacteria lacking previously known determinants show mcr-like downregulation of target genes, evoking structural modification of their lipid A. In this category of colistin-resistant E. coli, we identified pArnT1, a new plasmid carrying a Kluyvera-like arnBCADTEF operon whose association between element functioning and antibiotic susceptibility was found.

The analysis of E. coli from feces of healthy animals performed in this work (Table 1; see also Table SI in the supplemental material) showed that mcr-1 was the main determinant for colistin resistance in this environment at the time of screening (2014 to 2016). mcr-1 was carried by different plasmids, among which IncX4 predominated (70%), accurately identified as a 33-kb element by PFGE-S1 plus specific hybridization. These IncX4-like elements were highly mobilizable (44%), in contrast to the other mcr gene identified in this work, mcr-4, which was found linked to a low-molecular-weight ColE1-like plasmid and was not transferable in conjugation assays (Table 1 and Table SI), like the megaplasmid with the mcr-3 gene described previously (7).

Among the 10 mcr genes described so far in Gram-negative bacteria, mcr-1 was the first plasmidic colistin resistance determinant identified (15), and, not casually, it is the most common and widespread and largely predominates in E. coli (16). The health status of the animals from which E. coli is isolated does not seem to contribute to the prevalence of mcr-1, being as high (90%) in China as in Spain from healthy poultry and pigs (17) (Table 1) or in France from sick pigs (18), among which roughly 50% of the colistin resistance genes were mobilized by conjugation, similarly to this work (Table 1). However, local outbreaks may explain why E. coli from cases of postweaning diarrhea in Spain presented a higher prevalence of mcr-4 than of mcr-1 or mcr-5, the other two determinants frequently found (19).

Besides mcr genes, chromosomal mutations and another as-yet-unknown acquired determinant(s) are systematically detected among colistin-resistant strains, although their prevalence is variable depending on the environment from which E. coli is isolated. Thus, colistin-resistant E. coli from human inpatients in the Paris, France, region presented low rates of carriage of mcr genes (5%, with mcr-1 uniquely found), a high rate of missense polymorphisms in pmrA or pmrB coding sequences (69%), and a significant fraction (26%) lacking known resistance determinants (20). In contrast, a low fraction (4%) of isolates remained without known determinants from diseased animals analyzed in Spain (19, 21), fitting the prevalence found in this work (Table 1). In our screening, we sampled bacteria from healthy animals, and we found 15 isolates presenting chromosomal mutations in PmrAB that cannot be detected in colistin-susceptible strains, a first clue of their possible involvement in colistin resistance (5). However, neither of these resulted in the upregulation of eptA and/or arnB expression, in contrast to previously known mutations that confer colistin resistance (10) that activate PmrAB constitutively, even in the presence of mcr determinants (Fig. 2). Thus, although mcr expression might negatively signal mRNA accumulation from both operons (14) (Fig. 2), PmrAB mutations seem to dominate in strains that present both resistance determinants, which might suggest that the PmrAB polymorphisms detected in this work would not present functional relevance.

A significant fraction of the isolates analyzed in this work (5.4%) lacked previously known acquired determinants for colistin resistance, and most of them carry pArnT1, a new 62.1-kb plasmid belonging to the IncFII replicon type. This element is a chimera among fragments from two replicons previously described, pBH11alpha (GenBank accession number CP025254.1) and p65 (GenBank accession number MT077888.1), and a genomic fragment closely related (97.7% identity) to the arnBCADTEF operon from the opportunistic human pathogen K. ascorbata. This sequence is located downstream of ISEcp1, an element that could control its transcription and may have participated in its mobilization in a way similar to the role played in the spread of plasmid-borne bla_CTX-M genes, also from Kluyvera genomes (13). Since, to our knowledge, plasmid mobilization of the arnBCADTEF operon has never been found previously, its occurrence in E. coli isolates sharing low susceptibility to colistin, lacking previously known resistance determinants, and presenting downregulated expression from their own native arnBCADTEF operon strongly suggests a role for pArnT1 in colistin resistance. Furthermore, we have shown by plasmid curing that colistin resistance and pArnT1 are linked, although this element lacks mobilization potential by conjugation, in contrast to the roughly 50% of plasmids carrying mcr-1 genes detected in this work. Indeed, pArnT1 lacks mobilization functions characteristic of IncFII replicons and presents truncated coding sequences for TraB and TraC (Fig. 4), probably explaining the reduced spread of this element, until now found to be limited to the vertical transmission of clonally related isolates (Fig. 3A).

The Kluyvera-like arnBCADTEF operon found in pArnT1 is apparently functional and, like its E. coli orthologs (Table SIV in the supplemental material), might be related to LPS modification of enterobacteria, although little is known about the colistin resistance mechanisms of K. ascorbata. In contrast to mcr genes, several of which (mcr-1, -2, -3, -4, and -5) have been proven to confer colistin resistance upon cloning in E. coli and transcription from expression vectors, there has not been any similar approach focused on the complete arnBCADTEF operon, despite the parallel roles of their encoded functions in LPS modification (Fig. 1). The expression in E. coli of its own arnT coding sequence from the same pBAD vector used to express mcr genes increased colistin resistance only marginally (1). However, the arnT coding sequence was expressed lacking the other six gene functions carried by the arnBCADTEF operon plus the unlinked ugd gene, altogether providing l-Ara4N-undecaprenyl phosphate, the substrate for ArnT activity (Fig. 1).

In this work, we successfully expressed the Kluyvera-like arnBCADTEF operon carried by E. coli colistin-resistant isolates in the susceptible XL1-Blue MRF′ strain, where it was shown to confer heteroresistance to the antibiotic. This noncanonical behavior might be the consequence of the emergence (due to genetic, epigenetic, or environmental factors) of a heterogeneous phenotype due to the presence of mixed cells with different levels of susceptibility, which produces partial inhibition of growth when the resistance of the susceptible subpopulation is overcome, whereas resistant subpopulation growth continues until a concentration is reached that is at least 8-fold higher in a standard MIC assay by 2-fold increases of the antibiotic concentration (22). The fact that the Kluyvera-like operon from pArnT1 or pBAD24 confers a similar phenotype by transformation into the XL1-Blue MRF′ strain suggests that no other function encoded by the 62.1-kb plasmid is required for colistin resistance and that heteroresistance might be mediated by a singularity of the genetic background where arnBCADTEF is expressed. Future works are required to understand the biochemical and physiological consequences derived from the functioning of these elements in different cell lineages.

The presented evidence suggests that enterobacteria have gained a new colistin resistance determinant by plasmid mobilization of genes for the modification of lipid A with l-Ara4N, in addition to mcr genes that perform it by PEtN addition. pArnT1 or similar elements should be screened in enterobacteria, mainly in those lacking previously known colistin resistance determinants, to know their prevalence in animal and human environments, evaluating the risk of this new challenge from a One Health perspective.

MATERIALS AND METHODS

Strain origins.

In the context of the Spanish Surveillance Network of Antimicrobial Resistance in Bacteria of Veterinary Origin (VAV Network, VISAVET) (23), E. coli strains isolated from turkey feces in 2014 and 2016 and from swine or bovine feces in 2015 were included in the current study (see Table SI in the supplemental material). Samples were pooled in situ from 10 different animals in the case of turkey and from 2 different animals in the case of swine and bovine. All the samples were collected at a slaughterhouse in the frame of samplings carried out in Spain, randomly covering the entire country and performed according to the rules of the European Union authorities (29).

Determination of colistin resistance.

The MIC of colistin (milligrams per liter) was determined by the 2-fold broth microdilution reference method according to ISO 20776-1:2006. The threshold of antimicrobial resistance adopted in this work was based on the epidemiological cutoff (ECOFF) values recommended by EUCAST (http://www.eucast.org). E. coli strain ATCC 25922 was used for quality control of the technique.

DNA synthesis, purification, and Sanger sequencing.

DNA samples for PCR were obtained by boiling LB cultures overnight. Primers for PCRs were synthesized by Stabvida (Caparica, Portugal). PCR was performed with Dream Taq DNA polymerase (Thermo Fisher) according to the manufacturer’s protocol, with cycling started by an initial denaturation step for 5 min at 98°C, followed by 30 cycles of 30 s at 98°C, 30 s at the annealing temperature (Table SII), and 2 min at 72°C (except for quantitative PCR [qPCR] and full-length arnBCADTEF amplification [see below]), and ended by a final elongation step for 10 min at 72°C. PCR products were purified (MEGAquick-spin plus fragment DNA purification kit; iNtRON Biotechnology, Seongnam-Si, South Korea). Sanger sequencing of DNA fragments was performed in the facilities of the University of Extremadura, at the Servicio de Técnicas Aplicadas a la Biociencia (STAB).

Screening colistin resistance determinants.

Among the 10 plasmid-mediated colistin resistance determinants (mcr genes) (11), the first 4 genes, mcr-1 to mcr-4, were screened by PCR using primers and conditions described previously (Table SII and references therein). In addition, further screening was performed with a set of primers designed from the sequence alignment of the mcr-1 to mcr-8 genes (see reference 1 and references therein) to specifically detect mcr-1, -2, and -6; mcr-5; mcr-3 and -7; or mcr-4 and -8 in separate PCRs, whose functionality was evidenced by using DNA from control strains carrying mcr-1, -2, -3, and -4 or mcr-5. The recently described mcr-9 and mcr-10 genes were voluntarily omitted from this PCR screening since, in contrast to all previously known mcr genes, their role in colistin resistance has yet to be demonstrated and could be marginal at best (11, 16).

The sequences from pmrAB genes were screened by PCR and sequencing (see above) in search of missense polymorphisms of encoded proteins, excluding those variants found in colistin-susceptible strains like E. coli ATCC 25922 and K-12 or those recently isolated from farm animals (5).

Quantitative PCR.

Gene expression analyses were performed with cells growing in Mueller-Hinton II liquid cultures. After growth overnight, cultures were renewed by diluting them 1/10 with fresh medium and incubated at 37°C with gentle shaking (200 rpm). When cell cultures reached an optical density at 600 nm (OD₆₀₀) of 0.3 to 0.5, they were quickly cooled on ice, centrifuged, and processed for RNA extraction (Aurum total RNA minikit; Bio-Rad), reverse transcription (RT) (PrimeScript RT reagent kit; TaKaRa), and the removal of genomic DNA (Turbo DNA-free kit; Ambion), according to the manufacturers’ protocols. SYBR green real-time quantitative assays were performed by using SYBR Premix ExTaq II (Tli RNase H Plus; TaKaRa) and an Applied Biosystems Step One PCR system. Oligo Primer Analysis software v.7 was utilized to design primer sequences with optimal amplification efficiencies (Table SII). The normalized relative quantities (NRQs) of transcripts were obtained by the 2^−ΔΔCT calculation method, with the expression of the recA gene used as the calibration reference and E. coli strain ATCC 25922 grown under the same conditions used as the reference for normalization (24).

Pulsed-field gel electrophoresis.

Determination of the genetic relationships among E. coli isolates was performed by macrorestriction with XbaI followed by PFGE (Chef DRII; Bio-Rad), according to the PulseNet protocol, with pulses oscillating from 2.16 to 63.8 s for 21.5 h (25) and Salmonella enterica serovar Braenderup used as the molecular weight standard, and the gel was stained with ethidium bromide. Plasmid composition analysis was performed by PFGE under the same conditions as the ones described above, after incubation of plugs with S1 nuclease (Thermo Fisher) according to the manufacturer’s recommendations. For plasmid hybridization, the PFGE-S1 gel was transferred to a nylon membrane and hybridized to digoxigenin (DIG)-labeled probes PCR amplified from mcr-1, mcr-4, or Kluyvera-like arnB genes carried by E. coli isolates HSP38 (26), ZTA15/00750EB1, and ZTA15/00213-1EB1 (this work), respectively, by using specific primers (Table SII), according to the manufacturer’s instructions (Sigma-Aldrich).

Genome analysis.

DNA was extracted by using the Qiagen DNeasy blood and tissue kit, and sequencing libraries were prepared using the Nextera XT kit and sequenced on a MiSeq platform (Illumina) using v3 reagents with 2 by 300 cycles. Isolates ZTA15/00213-1EB1 and ZTA15/00702EC produced 493,612 and 473,653 reads, respectively, which were assembled using SPAdes v 3.9.0. The draft genomes were 5,076,949 and 5,092,360 bp long and composed of 452 and 449 contigs with N₅₀ values of 28,806 and 20,026, respectively, corresponding to 20× coverage in both cases. The pArnT1 sequence (62,100 bp) was deduced from the initial draft under GenBank accession number JACEFD000000000 by assembling the contigs NODE_75 (24,898 bp), NODE_304 (595 bp), and NODE_51 (34,939 bp), in that order. The joining through NODE_304 was confirmed by PCR with primer pair pArnT1a (Table SII), whereas closing the plasmid structure was performed by PCR between NODE_75 and NODE_51 with primer pair pArnT1b (Table SII) and sequencing the resulting DNA fragment.

Mobilization potential of colistin resistance determinants.

Conjugations were performed by mating donor cells with the E. coli J53 strain on LB medium during 20 h at 37°C and selecting transconjugants in 200 mg/liter sodium azide and 2 mg/liter colistin, and the efficiency was calculated by dividing the number of conjugants by that of donor cells, sharing resistance to both compounds or just to colistin, respectively.

For transformation, plasmids were directly used from ligation reactions (see below), or in the case of pArnT1, it was purified from the E. coli ZTA15/00213-1EB1 strain by using a QIAamp DNA minikit (Qiagen) according to the manufacturer’s protocol and transformed into E. coli XL1-Blue MRF′ cells (Agilent) by electroporation according to classical methods (27) and using ampicillin (100 mg/liter) to select for β-lactamase-producing clones.

Plasmid stability and curing.

The stability of plasmids carrying colistin resistance determinants was determined by incubating cells through successive growth cycles under nonselective conditions. Every cycle was initiated by diluting (10⁻²) a stationary-phase liquid culture with fresh LB medium and incubating the culture at 37°C with gentle shaking (200 rpm) for 16 to 20 h.

To challenge the stability of the plasmids, their curing was performed by using ethidium bromide in a gradient of 0.05, 0.1, and 0.2 g/liter and a 10⁻³ dilution of a stationary-phase liquid culture with fresh LB medium; incubating the culture at 37°C with gentle shaking (200 rpm) for 16 to 20 h; and, after that, inoculating the mixture in parallel nonselective (LB) and colistin-selective (LB supplemented with 2 mg/liter colistin) semisolid media. The curing efficiency was determined by comparing the numbers of growing colonies, and as expected (28), the optimal efficiency (90%) occurred by using 0.1 g/liter of ethidium bromide, the maximal noninhibitory concentration, since no visible growth was observed with 0.2 g/liter in liquid cultures.

Cloning the Kluyvera-like arnBCADTEF operon in pBAD24.

Specific primers (Table SII) were designed to amplify the full coding sequence of the Kluyvera-like arnBCADTEF operon from the ZTA15/00213-1EB1 strain. PCR was performed with Phusion green high-fidelity DNA polymerase (Thermo Fisher) according to the manufacturer’s protocol. PCR conditions for amplification included a primary denaturation step for 5 min at 98°C, followed by 30 cycles of 30 s at 98°C and 4 min at 68°C for annealing and elongation steps and a final elongation step for 10 min at 72°C. The DNA sample for the amplification of PCR products was obtained by using a QIAamp DNA minikit (Qiagen). After purification, the PCR fragment was digested with enzymes EcoRI and SalI (New England BioLabs), ligated (T4 DNA ligase; Thermo Fisher) to the arabinose-inducible pBAD24 expression vector (Life Science Market), and transformed into the XL1-Blue MRF′ strain (see above) by the selection of ampicillin (100 mg/liter)-resistant strains. XL1-Blue MRF′ cells carrying the intact pBAD24 vector or pBAD24::mcr-1 (14) were used as control strains for gene expression and antimicrobial resistance studies.

Data availability.

The sequences of isolates ZTA15/00213-1EB1 and ZTA15/00702EC from these whole-genome shotgun projects have been deposited at GenBank under the accession numbers JACEFD000000000 and JACEFE000000000, respectively. The versions described in this paper are accession numbers JACEFD010000000 and JACEFE010000000, respectively.