Whole-Genome Sequencing Identifies In Vivo Acquisition of a bla_CTX-M-27-Carrying IncFII Transmissible Plasmid as the Cause of Ceftriaxone Treatment Failure for an Invasive Salmonella enterica Serovar Typhimurium Infection

Abstract

We report a case of ceftriaxone treatment failure for bacteremia caused by Salmonella enterica subsp. enterica serovar Typhimurium, due to the in vivo acquisition of a bla_CTX-M-27-encoding IncFII group transmissible plasmid. The original β-lactamase-susceptible isolate ST882S was replaced by the resistant isolate ST931R during ceftriaxone treatment. After relapse, treatment was changed to ciprofloxacin, and the patient recovered. Isolate ST931R could transfer resistance to Escherichia coli at 37°C. We used whole-genome sequencing of ST882S and ST931R, the E. coli transconjugant, and isolated plasmid DNA to unequivocally show that ST882S and ST931R had identical chromosomes, both having 206 identical single-nucleotide polymorphisms (SNPs) versus S. Typhimurium 14028s. We assembled a complete circular genome for ST931R, to which ST882S reads mapped with no SNPs. ST882S and ST931R were isogenic except for the presence of three additional plasmids in ST931R. ST931R and the E. coli transconjugant were ceftriaxone resistant due to the presence of a 60.5-kb IS26-flanked, bla_CTX-M-27-encoding IncFII plasmid. Compared to 14082s, ST931R has almost identical Gifsy-1, Gifsy-2, and ST64B prophages, lacks Gifsy-3, and instead carries a unique Fels-2 prophage related to that found in LT2. ST882S and ST931R both had a 94-kb virulence plasmid showing >99% identity with pSLT14028s and a cryptic 3,904-bp replicon; ST931R also has cryptic 93-kb IncI1 and 62-kb IncI2 group plasmids. To the best of our knowledge, in vivo acquisition of extended-spectrum β-lactamase resistance by S. Typhimurium and bla_CTX-M-27 genes in U.S. isolates of Salmonella have not previously been reported.

INTRODUCTION

Greater than 1.4 million cases of nontyphoidal salmonellosis are estimated to occur each year in the United States, with approximately 95% of cases resulting from foodborne transmission (1). Although most infections result in mild to moderate gastroenteritis that usually resolves with or without treatment, concurrent bacteremia is observed in approximately 6% of patients and can be associated with serious disease, especially in infants aged <1 year, the elderly, and those with immunocompromised states such as HIV infection (2–5). Although routine use of antimicrobial therapy is not generally recommended for the treatment of most Salmonella infections, such therapy may be lifesaving in persons with invasive disease.

Treatment of invasive salmonellosis has been compromised in recent years due to the emergence of Salmonella isolates with single or multidrug resistance to a number of first-line agents (6, 7). Antimicrobial resistance in strains of nontyphoidal Salmonella (NTS) has been linked to the use of antimicrobial agents in livestock (8–11). Consequently, agents such as fluoroquinolones and third generation cephalosporins, such as ceftriaxone, have become treatment modalities of choice for therapy against severe Salmonella infections. In the United States before 1996, all reported cases of infection with ceftriaxone-resistant NTS isolates were known or postulated to be acquired abroad (12). In 2000, the first reported case of a domestically acquired Salmonella infection expressing ceftriaxone resistance was reported in the United States (13). Since that time a number of investigations have revealed the ongoing spread of ceftriaxone-resistant NTS isolates in both humans and domesticated animals (14–16).

Resistance to ceftriaxone in NTS strains outside North America has generally been related to acquisition of plasmids containing SHV-, TEM-, or CTX-M-encoding extended-spectrum β-lactamases (ESBLs) (17–19). In contrast, decreased susceptibility to ceftriaxone observed in NTS isolates recovered from both humans and domesticated animals in the United States until recently has been almost exclusively mediated through the production of bla_CMY-2, a Citrobacter freundii-derived ampC β-lactamase that is encoded on a variety of transmissible plasmids (20).

More recently, human NTS isolates harboring plasmids encoding ESBLs of the CTX-M family have been recovered sporadically in the United States, which were thought to be both domestically and nondomestically acquired (21). Consistent with these findings, recovery of bla_CTX-M-expressing NTS in U.S. livestock populations has been subsequently reported (22). The bla_CTX-M-type enzymes are an emerging group of class A ESBLs initially reported in the second half of the 1980s that have rapidly disseminated in certain Gram-negative bacilli, particularly in members of the Enterobacteriaceae family (23). These β-lactamases have become the predominant family of ESBLs in many regions of the world (24). In clinical strains, bla_CTX-M enzymes are encoded most commonly within plasmids that carry additional genes for resistance to other antimicrobials, including aminoglycosides, chloramphenicol, trimethoprim-sulfamethoxazole, fluoroquinolones, tetracycline, and 16S RNA methylase genes, as well as housing other β-lactamase genes, with increasing reports in China and the developing world (25–29), as well as the United States (30, 31).

Whole-genome sequencing (WGS) has recently begun to be used as an epidemiological tool to analyze in vivo acquisition of antibiotic resistance by pathogens, which has most frequently determined the cause to be development of chromosomal mutations or rearrangements (32–36). We report here our use of whole-genome sequencing to analyze in vivo emergence of ceftriaxone resistance in Salmonella enterica subsp. enterica serovar Typhimurium due to acquisition of a plasmid encoding bla_CTX-M-27 during treatment for Salmonella bacteremia.

MATERIALS AND METHODS

Case.

A 68-year-old female of Ethiopian descent, who was receiving immunosuppressant medication for rheumatoid arthritis, was admitted to the hospital for stridor due to an enlarged goiter. She was brought to the emergency room by ambulance and intubated for airway protection. She was febrile on admission to 38.9°C (102°F), and two sets of blood cultures were drawn (day 0 [d0]), which revealed no growth. On d2 of her hospitalization, she spiked a fever to 39.1°C (102.4°F), and two sets of blood cultures were obtained, which at 24 h were both positive for Gram-negative rods, later identified as S. Typhimurium. She was started on cefepime at 2 g every 8 h. Two sets of blood cultures obtained the following day (d3) remained negative. The isolate tested susceptible to ampicillin, ceftriaxone, ciprofloxacin, and trimethoprim-sulfamethoxazole by disk diffusion assay. The antibiotic regimen was narrowed to intravenous ceftriaxone given as 1 g daily once the results of antimicrobial sensitivity testing were reported (d5). Her fevers resolved with antibiotic therapy, but on d12 she developed an increase in her peripheral white blood cell count to 15.5 × 10⁹ liter⁻¹ without associated symptoms. Two sets of blood cultures were obtained, and both sets were positive for S. Typhimurium at 24 h. Antimicrobial susceptibility testing on the isolate revealed the emergence of antibiotic resistance to ampicillin and ceftriaxone. Her antibiotic regimen was switched to intravenous ciprofloxacin (400 mg) every 12 h. No endovascular or occult source of infection was found by transthoracic echocardiogram, computed tomography of the abdomen and pelvis, computed tomography with angiogram of the chest or whole-body positron emission tomography-computed tomography. However, the computed tomography of the abdomen and pelvis revealed evidence of wall thickening in the ascending colon suggestive of an infectious colitis, although no stool specimens were obtained for culture throughout her hospitalization. Subsequent blood cultures (d13 to d16) were negative, and she was discharged (d23) on intravenous ciprofloxacin to complete a 6-week course of therapy.

Bacterial strains and susceptibility testing.

The Salmonella isolates were recovered from blood culture specimens incubated on the BD Bactec FX (BD Diagnostic Systems, Sparks, MD) automated system. Positive blood culture specimens were processed at the University of Colorado Hospital Microbiology Laboratory according to standard operating procedures and bacterial isolates identified with the API 20E system (bioMérieux, Marcy l'Etoile, France). Confirmation of Salmonella spp. recovery and serotyping was performed by the Colorado Department of Public Health and Environment Laboratory.

Initial antimicrobial susceptibility testing (for ampicillin, ceftriaxone, ciprofloxacin, nalidixic acid, and trimethoprim-sulfamethoxazole) was performed by disk diffusion on BBL Mueller-Hinton agar plates (Becton Dickinson and Company, Sparks, MD) according to the methods of the Clinical and Laboratory Standards Institute (CLSI) (37). After 16 to 18 h of incubation at 35°C in a non-CO₂ incubator, inhibitory zone sizes were read and interpreted in accordance with the CLSI M100-S23 supplement (37). Antimicrobial MICs were then determined for each bacterial isolate following preparation of NUC51 panels and incubation in the MicroScan WalkAway automated system (Siemens, Munich, Germany) according to the manufacturer's instructions.

Matings.

Conjugal transfer of resistance was determined using 1:10 dilutions of fresh cultures from donor and recipient in prewarmed Luria-Bertani (LB) medium, followed by incubation without shaking for 90 min at 37°C. The numbers of donor, recipient, and transconjugant cells were determined by plating serial dilutions on selective (rifampin [Rif], streptomycin [Str], or ampicillin [Amp]) or differential (X-Gal [5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside]) LB agar and used to calculate the numbers of transconjugants per donor. The recipients were Escherichia coli ER1821R (a Rif^r isolate [38] of ER1821 [restrictionless, lac⁺; New England BioLabs, Ipswich, MA]) or JM83 (Str^r Δlac) (39).

Genomic DNA prep and library construction.

For genome sequencing, DNA was extracted using the UltraClean Microbial DNA isolation kit (Mo Bio, Inc., Carlsbad, CA) and prepared for sequencing using the Nextera XT kit (Illumina, Inc., San Diego, CA). Four DNA libraries were constructed, three from genomic DNA prepared from the ST882S and ST931R isolates, the E. coli Amp^r transconjugant from a cross with ST931R, and a plasmid DNA preparation from the latter strain. Multiplexed sequencing of these libraries was done with a single run on an Illumina MiSeq using paired-end reads (2 × 300 bp; MiSeq reagent kit, v3). After demultiplexing, sequences were analyzed using Geneious v6.1.7 software (BioMatters, Ltd., Auckland, New Zealand). De novo assembly of each data set was performed using a percentage of the total data appropriate for the expected genome size by the Geneious assembler (40) routine using the medium-low sensitivity parameters appropriate for whole-genome-sequencing (WGS) assembly. Using less data both minimizes the amount of memory required by the routine and may also increase the average contig size by reducing the contig fragmentation that can be caused by using too many reads. The routine loads data until the set percentage cutoff has been reached. We and others have previously used Geneious software to successfully assemble complete genomes for Escherichia coli (41) and Bacillus subtilis (42). For analysis of the plasmid content, paired-end reads were first mapped to the S. Typhimurium 14028s genome NC_016856.1 (43), followed by de novo assembly of unused reads into new contigs. Manual trimming and editing of terminally redundant contig ends generated circular plasmid genomes (several additional small contigs corresponded to Salmonella prophage sequences not present in the 14028s genome).

Genome analysis.

Single-nucleotide polymorphism (SNP) analysis was done using the CSIPhyologeny web server with default settings (https://cge.cbs.dtu.dk/services/CSIPhylogeny/) or output from the mapped consensus from within the Geneious program. Another server on the same site, ResFinder (https://cge.cbs.dtu.dk/services/ResFinder/) (44), was used to detect antibiotic resistance determinants in assembled plasmid genomes prior to annotation. Genome-level comparisons were done using progressiveMauve (45), which reorders and maps WGS contig data sets to complete genomes, producing a scalable graphic output from whole-genome down to base-pair resolution. Replicon typing made use of the PubMLST website (http://pubmlst.org/plasmid/) developed by Keith Jolley (46) and sited at the University of Oxford. The development of that website was funded by the Wellcome Trust.

Accession number(s).

Chromosome and plasmid genomes have been assigned GenBank accession numbers CP016385 to CP016390.

RESULTS

Antimicrobial susceptibility testing.

Initial testing using Kirby-Bauer disc diffusion assay indicated that ST882S was susceptible to all antimicrobials tested but that ST931R was resistant to ampicillin and ceftriaxone. MICs for 16 antibiotics tested against the susceptible and resistant S. Typhimurium isolates and the E. coli parent and transconjugant strain are listed in Table 1. The emergence of ampicillin and ceftriaxone resistance observed by Kirby-Bauer disk diffusion assay in the ST931R Salmonella isolate was again observed by MIC testing, as MICs to ampicillin and ceftriaxone increased at least 4-fold and 8-fold, respectively, compared to the initial, susceptible ST882S Salmonella isolate. Resistance to other cephalosporins, including cefazolin, cefotaxime, ceftazidime, and cefepime, also developed in this isolate. The addition of clavulanic acid (CLA) to cefotaxime and ceftazidime reversed the resistance observed for these two antibiotics, suggesting the presence of an ESBL-type enzyme. Resistance to members of other classes of antibiotics was not noted, since no discernible MIC changes to ertapenem, gentamicin, levofloxacin, tetracycline, or trimethoprim-sulfamethoxazole were observed between the two clinical Salmonella isolates. Taken together, these findings led us to hypothesize that the ST931R isolate emerged directly from the ST882S isolate by acquisition of ceftriaxone resistance within the patient during antibiotic therapy.

TABLE 1.

Microscan MIC reports for susceptible and resistant Salmonella and E. coli strains

Agenta	Microscan MIC (μg/ml)b
	Salmonella ST882S (susceptible)	Salmonella ST931R (resistant)	E. coli ER1821R (susceptible)	Transconjugant E. coli ER1821R (resistant)
AMP	≤8	≤16	≤8	≤16
AMP-SUL	≤8/4	16/8	≤8/4	16/8
AMX-CLA	≤8/4	≤8/4	≤8/4	≤8/4
Cefazolin	≤8	≤16	≤8	≤16
Ceftriaxone	≤8	≤32	≤8	≤32
Cefotaxime	≤2	≤32	≤2	≤32
Cefotaxime-CLA	≤0.5	≤0.5	≤0.5	≤0.5
Ceftazidime	≤1	≤16	≤1	≤16
Ceftazidime-CLA	≤0.5	≤0.5	≤0.5	≤0.5
Cefepime	≤4	≤16	≤4	≤16
PIP-TZB	≤16	≤16	≤16	≤16
Ertapenem	≤2	≤2	≤2	≤2
Gentamicin	≤4	≤4	≤4	≤4
Levofloxacin	≤2	≤2	≤2	≤2
Tetracycline	≤4	≤4	≤4	≤4
SXT	≤2/38	≤2/38	≤2/38	≤2/38

^aAMP, ampicillin; CLA, clavulanic acid; AMX, amoxicillin; SUL, sulbactam; PIP-TZB, piperacillin-tazobactam; SXT, trimethoprim-sulfamethoxazole.

^bBoldface indicates MIC levels that were reported as resistant, except for AMP-SUL, which was reported as intermediate.

Isolate ST931R encodes a transmissible ESBL resistance determinant.

Log-phase cultures of ST931R (Amp^r) and E. coli ER1821R (Rif^r) transferred the ESBL phenotype (selected as Amp^r) from ST931R to ER1821R at a frequency of 7.0 × 10⁻³ donor/recipient. Semiquantitative analyses (data not shown) showed that the transconjugant could retransmit Amp^r at similar frequencies to E. coli JM83 (Str^r), demonstrating that the transmissible factor itself is conjugation proficient. Detection of plasmid DNA using a commercial miniprep kit was inconclusive, but a modified alkaline lysis protocol using polyethylene glycol precipitation showed that the transconjugant possessed a single large plasmid that we named pESBL931. MIC testing of the E. coli strains showed that the transconjugant's ampicillin and cephalosporin resistance pattern matched that shown by the resistant Salmonella isolate. Together, these data show that the acquisition of a single ESBL-encoding plasmid was necessary and sufficient to explain the emergence of ampicillin and ceftriaxone resistance in the clinical Salmonella isolate.

Genome analysis.

To further analyze the plasmid and confirm the genetic relatedness of the ST882S and ST931R Salmonella isolates, we used next-generation DNA sequencing to assemble the sequences of the isolated plasmid DNA, the ceftriaxone-resistant E. coli transconjugant, and the full genomes (chromosome and plasmidome) of the ST882S and ST931R isolates. The results of sequencing and assembly are summarized in Table 2.

TABLE 2.

Whole-genome sequence assembly data for all DNA libraries

Sample	No. of paired MiSeq reads (% used for de novo analyses)	No. of contigs of >500 bp (no. of plasmid contigs)	Avg fold coveragea	Largest contig (bp)	N50 (bp)	No. of contigs with a size ≥N50
pESBL933 DNA	480,734 (20)	2b (1)	350	69,103	69,103	1
E. coli ER1821R(pESBL931)	1,705,087 (22)	84 (4p)c	20.0 ± 1.1*	269,473	103,211	16
S. Typhimurium ST882S	1,950,898 (20)	74 (2)	18.8 ± 0.6*	365,590	136,132	12
S. Typhimurium ST931R	1,621,490 (24)	84 (5)	15.5 ± 0.2*	271,829	97,525	15

^aThat is, for fractions of the full-read data set used for de novo assembly; coverages for mapping to assembled genomes used the whole data set. *, means for the ten largest nonplasmid contigs ± the standard deviations.

^bThe second contig coverage only matched IS5 sequences from E. coli chromosome five times.

^cThe “p” denotes partial incomplete or misassembled contigs that nevertheless fully cover the pESBL931 plasmid.

Chromosomal relatedness of isolates ST882S and ST931R.

To examine the relatedness of the Salmonella isolates, we used CSIPhylogeny, a single-nucleotide polymorphism (SNP) web server, to map one 300-bp read set each from the ST882S and ST931R data sets to 14028s, the first fully virulent S. Typhimurium genome sequenced to completion (NC_016856.1 [43]). This yielded 206 identical SNPs for both data sets at a reference genome coverage of 97.5%. Mapping both the ST882S and ST931R paired reads to the S. Typhimurium 14028s genome using the variant/SNP call routine in Geneious reported 260 identical SNPs for both ST882S and ST931R read sets (198 SNPs in common with an additional 62 SNPs not called by CSIPhylogeny, using the default parameters). Together, these two complementary methods show that the ST882S and ST931R chromosomal genomes cannot be distinguished from one another and support our hypothesis that the ST931R isolate arose from the ST882S isolate by acquisition of the pESBL931 plasmid encoding resistance to ceftriaxone.

Further proof for this conclusion came from a complete genome sequence for the ST931R chromosome that we assembled from the consensus sequence called by mapping the paired-read data set to the 14028s genome and replacing discordant regions (mostly insertion sequences [IS] or prophage insertions or deletions) with corresponding regions from contigs with perfect flanking homology from a de novo assembly. Illumina library inserts are generally too short to unequivocally assemble across regions of identity larger than 1 kb, such as some insertion sequences and rRNA operons. These made up the majority of the termini of the contigs in the de novo assembly. Mapping to a closely related complete genome can provide a framework from which to assemble a complete genome. We expand on this in Text S1 in the supplemental material. Subsequent minor manual editing of single polymorphisms (mostly in prophage or duplicated genes or rRNA loci) in the consensus versus 14082s produced a single circular chromosome of 4,852,920 bp with 61-fold average coverage (using the complete data set). The S data set remapped to this R genome with 95-fold average coverage, no gaps, and zero SNPs. A de novo assembly of unused ST882S reads from this mapping (83,870 paired and 9,273 mostly low quality unpaired) produced only two significant-sized contigs that corresponded to the virulence plasmid and p3904, and many other small low-coverage (<27 reads) and low-quality contigs, mostly of 4 reads or less. Only 1,222 unused paired reads of low quality remained, and low-quality unused unpaired reads increased to 9,585. In addition, at the genome level, progressiveMauve completely aligned the 74 contigs (>500 bp) from a de novo assembly of ST882S paired reads to the ST931R consensus genome, with four ambiguities and four unaligned contigs. Four contigs had incorrectly joined noncontiguous segments at IS200 homologies. Two unaligned contigs corresponded to the virulence plasmid and p3904, and two contigs (1.0 and 2.5 kb) corresponded to fragments of chimeric rRNA loci. This indicates that no additional genetic determinants were present in ST882S and confirmed that the S and R chromosomes, virulence plasmids and p3904 genomes can reasonably be said to be identical. Notably, both ST882S and ST931R have the identical SNP, 290950C, in the rrnH locus that is not found by a BLAST search in any other Salmonella genome (nonredundant/nt refseq or WGS). Full coverage for ST882S reads over the whole of the ST931R genome and lack of any substantial high-quality novel contigs from the ST882S unused reads after mapping additionally indicates that we did not miss any gross genome rearrangements by using 14028s as a guide genome.

While ST882S and ST931R are highly similar to 14082s, their prophage profiles differ, having a unique Fels-2 prophage not present in 14028s, while ST882S/ST931R lack the Gifsy-3 prophage found in 14028s. These prophages are named after the institute (Fels [47]) or its location (Gif sur Yvette [48]) of the discovering laboratories. We could find no complete closed genome for an S. Typhimurium isolate with an identical prophage profile, but, we did find a contig data set (based on the almost complete identity of their virulence plasmids, noted in the next section) for an isolate, CVM N42450, that appears to be the closest available genomic match to ST882S and ST931R. We analyzed SNPs between the genomes of ST931R and the contig set for isolate CVM N42450 (JYZN01) using CSIPhylogeny. This reported only 46 SNPs between these two isolates, compared to the 206 SNPs seen versus 14082s; 19 of these SNPs are shared by both comparators. Thus, the genome of ST931R is more closely related to that of CVM N42450, an S. Typhimurium strain isolated from chicken breast in California in 2012, than it is to 14028s. A progressiveMauve alignment of the 126 CVM N42450 contigs to the ST931R genome showed complete congruity except for what appears to be a misassembly of contig 27 (at the repeat region flanking the clustered regularly interspaced short palindromic repeat (CRISPR)-cas locus; marked with an asterisk in Fig. 1). Figure 1 aligns the four best-characterized complete genomes with the highest symmetric identity scores in the NCBI genome neighbor report for CVM N42450 (http://www.ncbi.nlm.nih.gov/genome/neighbors/152?genome_assembly_id=260322#) with the ST931R and CVM N42450 genomes. Prophage loci are marked with colored boxes. All pathogenic isolates have an ST64B-related prophage and all except CVM N42450 and ST882S/ST931R (which have completely syntenic prophage loci) differ at one or more other prophage loci. Although a more detailed analysis of the ST882S/ST931R genome is beyond the scope of this study, a brief overview of the ST882S/ST931R genome is given in Text S1 the supplemental material. The presence of a complete virulence plasmid (see next section) and the ST64B prophage is consistent with the invasive phenotype (49, 50) and, having been isolated from patient blood cultures, the ST882S/ST931R isolates clearly have a full complement of virulence genes necessary to produce an invasive infection.

FIG 1.

progressiveMauve alignment of the ST931R genome with representative S. Typhimurium genomes. Genomes are aligned on the empty ST64B locus of LT2 (bottom panel). Locally colinear blocks (conserved segments that appear to be internally free from genome rearrangements) are shown in different colors; their boundaries generally coincide with prophage loci. Almost all variability between genomes is due to these prophage loci and occurs in the central 2 Mb. Each panel corresponds to a single genome, with the strain name at bottom left and base pair numbering across the top. Red ticks on the CVM N42450 genome show contig boundaries (all others are complete circular genomes). Colored boxes identify homologous prophages (linked between genomes by solid lines), with red boxes showing unique prophages. Smaller black boxes show the CRISPR-cas-associated repeat that appears misassembled in CVM N42450 (marked with an asterisk). To the right is shown the percent identities (symmetrical and gapped) of each comparator genome to CVM N42450 taken from the genome neighbor profile. Citations for published genomes are SL1344 (FQ312003.1) (106), UK-1 (NC_016863.1) (107), 14082s (NC_016856.1) (43), and LT2 (NC_003197.1) (51).

Plasmidomes of ST882SS and ST931R.

Plasmid-related contigs were identified in all data sets based on the presence of terminally repeated sequences and increased data coverage (indicative of increased copy number) relative to the genomic contigs. Both the ST882S and the ST931R genomic data independently assembled contigs that produced identical 93,855-bp and 3,904-bp circular plasmids. The larger plasmid was >99.9% identical to the 93,832-bp 14028s virulence plasmid CP001362 (51), with only five SNPs (two intergenic, two silent, and an F₂₀₀L substitution in SpvD), a 24-bp expansion of a repeat sequence (11 versus 7 repeats of CCTGTT) and a 1-bp deletion in a putative transglycosylase gene. A BLAST search of the WGS database with the DNA sequence, including the 1-bp deletion, found a single complete match to a contig from the genome assembly of a 2012 S. Typhimurium isolate, CVM N42450, from California (WGS NZ_JYZN01). The ST882S/ST931R virulence plasmid is 100% identical to eight contigs (99.8% coverage) for CVM N42450, covering all of the above SNPs and the deletion; unfortunately, coverage for the repeat sequence is lacking. The smaller plasmid was identical to pEC147-3 (JX238454), a cryptic plasmid discovered in a plasmidome analysis of a set of ESBL-producing Escherichia coli isolates from urinary tract infections (52).

The ST931R genome data set identified three other plasmid genomes not present in the ST882S genome data set. First we identified pESBL931, a 68,117-bp IncFII (2:A−:B−) group plasmid (alleles determined from pubmlst.org/plasmid). This sequence was analyzed with ResFinder and a single β-lactamase resistance gene, identical to an ESBL-type bla_CTX-M-27 determinant (53), was identified. In addition, two IncI plasmid replicons were assembled, a 93,202-bp IncI1 (3:4:17:3:3) plasmid and a 60,496-bp IncI2 plasmid. The pESBL931 plasmid DNA data set assembled a single large contig producing a 68,117-bp circular genome identical to that from the ST931R genome assembly. The total genomic data set for E. coli carrying pESBL931 was first mapped to an E. coli K-12 genome (MG1655) and then, by de novo assembly of the unused reads, we assembled a 68,117-bp circular plasmid identical to the independent assemblies of pESBL931. From this data set, we have also assembled and published a complete genome for the E. coli recipient strain ER1821R (38). All data sets from Amp^r samples assembled identical 68,117-bp pESBL931 plasmids. A graphical depiction of pESBL931 is shown in Fig. 2.

FIG 2.

Physical map of pESBL931 with comparison of its bla_CTX-M-27 resistance module to other similar bla_CTX-M modules and plasmids. The lower portion of the figure shows a circular map of the 68,117-bp plasmid pESBL931, with distinct regions of the plasmid as boxes on the circle (red, bla_CTX-M-27 module; light blue, toxin/antitoxin [pemIK and hok/mok] and stability [stbBA] loci; white, tra, conjugal transfer region; light green, rep, plasmid replication genes). Open reading frames are shown on two tracks inside the circle as solid red arrows/triangles pointing in the direction of transcription (sense strand, followed by antisense strand), except that bla_CTX-M-27 is shown in yellow, and putative transposases are shown in black with ΔiroN in gray. The inner track shows the deviation from the average %GC. Tracks were generated by CGView server (108). The upper portion expands (blue lines) and compares the pemIK-stbBA region (these genes flank the resistance module) of pESBL931, with putative ancestral plasmid 2009C-3133 (CP013026) encoding a complete yjcA gene. The plasmid name, size of the region expanded, and the full plasmid size are given on the left. R100 has a simple insertion (not shown to scale) of an IS1-flanked (red boxes) resistance module into yjcA; all other plasmids have deleted the 5′ portion of and 1,375 bp distal to yjcA, shown as a black box. Other IS elements are shown as different colored boxes (blue, IS26; dark green, IS903B; orange, ISEcp1; white, fragments of Tn2 or Tn1722). Open yellow arrows indicate resistance genes; black-filled arrows indicate transposase genes of IS elements. Unlabeled genes or elements correspond to similarly labeled elements in pESBL931. Open red arrows denote plasmid backbone genes. Text to the right shows the status of two plasmid backbone loci that may contain an intron in ycjA (note, this gene is distinct from yjcA) or a complete Tn10 or IS10 scar into parB; wt, wild type (no insertion); wt_var, no insert but variant gene sequence. Blue arrows point to these loci on the circular map.

Relative plasmid copy number can be estimated by mapping the full read data set back to the circular contigs and comparing the ratio of read coverages for each plasmid contig versus the complete chromosome. This assumes plasmids and chromosomes behave similarly at each step in the process (e.g., DNA isolation, library preparation or amplification steps). Average coverage is remarkably consistent; for example, the top 10 non-plasmid contigs for the ST931R de novo assembly (over 2 million bases) gave an average coverage of 15.5 ± 0.2 (Table 2), with an average standard deviation of 5.6 ± 0.2 (36% ± 1%). For the ST931R genomes, the calculated plasmid copy numbers (rounded to the nearest whole number to emend false precision) versus 61-fold coverage for the genome were 3× (virulence plasmid and p931IncI1), 5× (p3904), 7× (pESBL931), and 9× (p931IncI2). Relative copy numbers for the virulence plasmid and p3904 versus the ST882S genome were similar at 2× and 4×, respectively, at 95-fold genome coverage. Interestingly, the copy number for pESBL931 in E. coli was 6-fold lower than in ST931R, at only 1× to 2× (135-fold versus 90-fold genome coverage), although this difference was not reflected in the MIC levels, which were the same for both strains (Table 1).

Together, these genome assemblies suggest that ST931R arose from ST882S by the concurrent acquisition of all three plasmids in vivo, concomitant with acquisition of ceftriaxone resistance. No other antibiotic resistance determinants were found to be encoded by any of the other plasmids, a finding consistent with the MIC data for the S and R isolates, although the IncI1 plasmid encodes a novel microcin highly similar to mccPDI (54). mccPDI is responsible for the proximity-dependent inhibition phenotype (55) expressed by a group of cattle-adapted E. coli and hypothesized, by virtue of being able to inhibit a wide variety of E. coli strains, to give these strains a selective advantage in vivo. A brief characterization of these plasmids is presented in Text S1 in the supplemental material.

Analysis of pESBL931 and context of the bla_CTX-M-27 gene.

The bla_CTX-M-27 locus encoded by pESBL931 is flanked by direct repeats of IS26 in reverse orientation (i.e., tnpA encoded on the reverse strand relative to bla_CTX-M-27). Internal to the IS26 elements is an ISEcp1 fragment 5′ of the bla_CTX-M-27 allele that likely provides a high-strength promoter (56), a complete IS903D element 3′ of bla_CTX-M-27, followed by an amino-terminally truncated iroN reading frame, Tn2-related sequences, and a second copy of IS26. This organization is also found in many bla_CTX-M-24 and bla_CTX-M-65 loci (29) (Fig. 2). This arrangement and insertion site are identical to those found in a group of ESBL-encoding plasmids from China, typified by plasmids pHN7A8 (JN232517) (29) and pXZ (JF927996) (57). However, the bla_CTX-M loci in these plasmids encode different CTX-M alleles and also have additional genes encoding resistance to fosfomycin and aminoglycosides, also flanked by direct repeats of IS26 in the same reverse configuration (Fig. 2).

All of these plasmids are related to R100 from Shigella flexneri, the type plasmid for the IncFII group (NC_002134.1). pESBL931 is most similar to pXZ (57) (JF927996) and pHN3A11 (JX997935) (58), plasmids found in ESBL-producing E. coli strains in pets and food animals from China. This IncFII plasmid backbone is a common vector for multiple drug resistance cassettes worldwide, often with identical insertions at this locus. The R100 plasmid, also known as NR1, was isolated in Japan in the 1950s and is the archetypical drug resistance R plasmid (59). It possesses an IS1-flanked multidrug resistance (MDR) determinant (known as Tn2670) inserted into the yjcA gene, and a full copy of Tn10 at another location (ydeB-yefA) on the plasmid. yjcA is a nonessential gene located between the plasmid addiction system pemIK and plasmid stability/partition stbBA loci (Fig. 2).

Figure 2 also shows this region from a plasmid from a Shigatoxinogenic E. coli isolate 2009C-3133 (CP013026) that appears to have complete uninterrupted locus. An additional plasmid pEC743_3 (NZ_CP015072.1) has a minimal IS1-IS26-based deletion/insertion at this locus but lacks any antibiotic resistance genes. The pESBL931 bla_CTX-M-27 locus is a recombinational insertion of an IS26-flanked cassette at this site, and the related MDR plasmid loci in pHN7A8/pXZ can be obtained from this by recombinational insertion of another IS26-based cassette encoding the fos, bla_TEM1b, and rmtB genes. The backbones of all these plasmids are highly related but differ from R100 at two other sites as noted by He et al. (29). Some may have a 2.3-kb group II intron near ycjA (unfortunately easily confused with yjcA, the site of the bla_CXT-M-27 locus). The second site has either a wild-type gene or an IS10 inverted repeat-related scar (Fig. 2), where R100 has an intact Tn10 element. A more detailed analysis of these rearrangements is given in Text S1 in the supplemental material.

DISCUSSION

In vivo acquisition of pESBL931.

We report a case of ceftriaxone treatment failure for a Salmonella enterica serovar Typhimurium isolate due to the in vivo acquisition of a bla_CTX-M-27-encoding IncFII:2 transmissible plasmid, during antibiotic therapy, from an unidentified donor organism. In vivo acquisition of antibiotic resistance is rarely documented; in the last 20 years, we found only three reports that identified putative in vivo acquisition of ESBL resistance due to conjugative plasmid transfer during therapy with cefotaxime or ceftriaxone, in three other Salmonella serovars: Enteritidis in 1995 (60), Anatum in 2003 (61), and Oranienburg in 2016 (62). However, as is the case in this study, none of these reports could identify the source of resistance, although the first two reports speculated that the occurrence of similar resistance patterns or resistance alleles in other isolates in the clinical environment indicated a likely source for the plasmid donor. A few other reports document varied antibiotic resistance acquisition during Salmonella infections from nosocomial (63) or inpatient isolates of other Gram-negative pathogens (64–66). Most drug-resistant Salmonella isolates are suggested to arise from foods of animal origin, potentially from the use of antimicrobial agents in livestock (13, 67, 68). However, the development of antibiotic resistance within Salmonella strains as part of human fecal carriage has also been observed. In a community outbreak of a susceptible S. Typhimurium strain, ca. 15% of persons receiving ampicillin therapy had ampicillin-resistant isolates identified subsequently in their stool specimens (69). As in animals, fecal carriage in humans of commensal microbiota that harbor resistance determinants, including bla_CTX-M genetic elements, has been observed. Thus, normal human microbiota likely is an additional pool of genetic resistance determinants from which resistant Salmonella isolates can emerge. In support of this model, several studies have documented the possible in vivo transfer of plasmid-mediated antibiotic resistance between Salmonella isolates and resident human commensal microbes (64, 70). Although providing a precedent for in vivo acquisition of antibiotic resistance, we cannot definitively identify a source of the pESBL931 plasmid.

Our patient reported consuming raw beef 5 days prior to the onset of Salmonella bacteremia, the only identifiable epidemiologic risk factor for Salmonella acquisition in this patient. The closest genome match to ST882S/ST931R, CVM N14540, is a S. Typhimurium that has recently been found elsewhere in the food chain. Raw beef, if contaminated at some point in the food chain or preparation, could also be a potential source of an ESBL donor. ESBL-positive E. coli has increasingly been found in the food chain (22, 71), and the biosample information for the pESBL931-related bla_CTX-M-27 encoding plasmid pTC_N37410PS (CP007652) reports the source as a 2012 isolate from U.S. cattle. An alternative and equally plausible source for the ESBL donor is the patient's own gut flora. The patient's case history details evidence of wall thickening in the ascending colon, suggestive of an infectious colitis, a likely primary source of infection for S. Typhimurium. We can envisage enteric ST882S encountering an ESBL donor in the gut milieu, followed by a secondary invasive infection of ST931R during antibiotic treatment. Carriage by healthy (72) and gastrointestinal-symptomatic (73) adults of ESBL-positive E. coli and bla_CTX-M-27 in particular (74) has been documented, although specific plasmids were not identified. Unfortunately, the diagnostic regime for our patient did not require stool cultures, which may in retrospect have been used to identify an ESBL donor organism.

ESBL plasmids CTX-M and Salmonella.

Currently, pESBL931 is the only annotated and complete bla_CTX-M-27 plasmid in GenBank (accessed 8/31/2016) from an organism other than E. coli. A recent entry (KX008967) appears to be a complete but unannotated multidrug resistant plasmid from Shigella sonnei with a bla_CTX-M-27 identical DNA sequence. There is only one other complete and annotated bla_CTX-M-27 plasmid sequence, from an E. coli isolated from a patient in Thailand in 2012 (pEC732-2, NZ_CP015140 [unpublished data]). pEC732-2 has a different plasmid backbone and a divergent bla_CTX-M-27 locus, with an identical IS1R insertion site 3′ of pemK, but a truncated ISEcp1 preceding the bla_CTX-M-27 gene. This is followed by a partial IS903 sequence linked to an unrelated 18-kb IS26-flanked MDR region in a different and hybrid IncF I/II replicon. A plasmid very similar to pESBL931 (pTC_N37410PS, CP007652 [unpublished data]; a 2012 E. coli isolate from Texas cattle) has an incomplete but syntenic bla_CTX-M-27 locus (from pemK to stbA) with 34N′s replacing approximately 1.8 kb of pESBL931 sequence (covering almost all of IS1R-Tn2-IS26) 3′ of the bla_CTX-M-27 gene, a 60-bp deletion 5′ of ISEcp1, and four small insertions (36, 1, 1 and 22 bp) in the distal IS26 element that frameshift the tnpA gene.

Information in general on flanking sequences for most bla_CTX-M-27 loci is similarly sparse. Of the 32 sequences for bla_CTX-M-27 in GenBank as of June 2016, only 5 have significant flanking sequence data; all available flanking sequences have an ISEcp1 fragment (with variable spacing) 5′ of the bla_CTX-M-27 allele and part or all of an IS903 element 3′ of bla_CTX-M-27. Most are also followed by an amino-terminally truncated iroN reading frame (an organization also found in many CTX-M-24 and 65 loci), followed by a second IS26 (Fig. 2). The bla-IS903D-ΔiroN association was first found in a 1999 Klebsiella isolate bla_CTX-M-19 locus associated with a complete ISEcp1 element and which appeared to be inserted within a Tn1721 element (75). Although the pemIK-IS1R-IS26-IsEcp1-bla_CTX-M-IS903-ΔiroN gene order is common to several CTX-M loci (e.g., M14, M24, and M65), the order of the genes 3′ of bla_CTX-M-27-IS903D (i.e., ΔiroN-′Tn2′-IS26) appears restricted to and can be considered a marker of bla_CTX-M-27 loci. In a PCR-based study of local Japanese E. coli ESBL isolates, Matsumura (76) noted six distinct arrangements to the bla_CTX-M-27 flanking sequences, almost all (174/176) flanked by inverted repeats of IS26 in contrast to the locus in pESBL931 that is flanked by direct repeats. Details of the type of plasmid or insertion location were not presented. This arrangement may be peculiar to the local environment; the only other instance of a bla_CTX-M27 locus flanked by inverted repeats of IS26 is a contig beginning and ending with partial IS26 elements (and thus no genomic context; NZ_JVGK01000036.1) from a urine isolate of E. coli, collected and sequenced as part of a longitudinal study of all isolates from an intensive care unit in Washington, USA (77).

ESBL, CTX-M, and bla_CTX-M-27 in Salmonella.

In the United States, most third-generation cephalosporin resistance among human Salmonella isolates has historically been due to a plasmid-mediated bla_CMY-2 (AmpC-type) β-lactamases (14, 78, 79), and while NTS with ESBL determinants (including bla_CTX-M) are increasingly common in other parts of the world (19, 80–82), they still remain rare in the United States (83). bla_CTX-M determinants in NTS infections are now being reported in the United States (84) and have increased significantly in veterinary and food isolates in the rest of the world in recent years (85–87). bla_CTX-M-encoding NTS were first isolated in the late 1980s in Tunisia (88) and in 1990 in Argentina and concurrently in Europe (89). The first U.S. occurrence of bla_CTX-M of any type in NTS in humans was found in S. Typhimurium in 2003 (90), and three more cases were reported in 2007 (21). In 2005, the first publication specifically reporting bla_CTX-M-27 in NTS (serovar Livingston) appeared, occurring in a nosocomial outbreak among neonates in Tunisia in 2002; these isolates all showed high-level resistance to ceftriaxone (91). Two recent reports from China found bla_CTX-M-27 plasmids (all with varied but non-FII Inc group replicons) in serovar Indiana isolates in ducks from 2009 to 2010, reported in 2014 (28) and in Indiana and Typhimurium serovars from pigs and chickens in 2014, reported in 2016 (92), which now identifies bla_CTX-M-27 as the most common ESBL determinant in food-producing animals in China. A single case of an Indiana serovar with bla_CTX-M-27 from a 2011 neonate infection (from 21 ESBL isolates of human and 133 isolates of animal origin) was also reported in 2016 from China (93). The bla_CTX-M-27 allele was first identified in 2003 from a 2000 French clinical isolate of E. coli (pathotype not stated) (53). It differs from bla_CTX-M-14 at a single residue (D240G), a substitution that is also found in bla_CTX-M-15 and bla_CTX-M-16 loci. Compared to bla_CTX-M-14, bla_CTX-M-27 has a decreased K_m for ceftazidime, producing a corresponding increase in MIC (53). F-group plasmids with bla_CTX-M-27 alleles are much more common in E. coli isolates (94), and bla_CTX-M-27 is now found frequently in the ST131 group of E. coli (74, 76, 95, 96), although these studies also note that the context of the loci are limited or infrequently reported.

Context of the other plasmids unique to ST931R and shared with ST882S.

Plasmids related to the other two IncI group plasmids found in ST931R have received little study since the shufflon characteristic of this group of plasmids were first reported in 1987 for IncI1 (97) and 1990 for IncI2 (98). Twenty-five years later, with the widespread use of whole-genome sequencing, coverage of this group of plasmids has exploded through recognition of their role in spread of antibiotic resistance, particularly ESBL β-lactamases—with over 200 PubMed articles relating to IncI1 plasmids in the last decade; IncI2 plasmids have only recently been discovered as vectors for ESBL and other resistance determinants (99), with more than half of relevant Pubmed articles published since 2013. Both sets of plasmids were historically associated with E. coli but are increasingly found in NTS isolates (80, 100). Alarmingly, ESBL-encoding IncI2 plasmids are now spreading the colistin resistance determinant mcr-1 in China (101) and Europe (102). The first U.S. case of colistin resistance due to mcr-1 occurred with an E. coli carrying a novel MDR bla_CTX-M-55 IncF plasmid earlier this year (103). Although both Salmonella IncI plasmids were cryptic, the ST931R IncI1 plasmid is most similar to pTF2 (KJ563250), a bla_CTX-M-1-encoding plasmid in a Danish E. coli isolate from an unpublished study (99% identity with 89% coverage). The ST931R IncI2 plasmid is most similar (99% identity with 90% coverage) to an unpublished cryptic plasmid (CP011294) from a European S. enterica subsp. diarizonae 60:r:z causing diarrhea and sepsis with fatal outcome. A plasmid identical to the 3,904-bp plasmid found in both ST882S and ST931R was originally discovered in an ESBL-positive (IncF bla_TEM-1 and IncI1 bla_CTX-M-9) urinary tract E. coli isolate (52); almost identical plasmids have been found in an E. coli isolate causing septicemia (104) and in an MDR NTS serovar Heidelberg (105) isolate from turkey meat in Canada (as the largest of four small plasmids, along with an MDR-IncHI2 plasmid). These matches support the idea that the original ST882S Salmonella isolate had already encountered plasmid-bearing E. coli prior to the acquisition of pESBL931 and the IncI plasmids.

Conclusion.

This study is the first to use whole-genome sequencing to unequivocally determine the in vivo acquisition of plasmid-determined antibiotic resistance by an originally susceptible Salmonella as the cause of relapse. This is an infrequently documented and unusual but clinically significant observation. In this case, multiple plasmids were transferred to the pathogen, probably simultaneously. Since the particular ESBL determinant involved, bla_CTX-M-27, is uncommon in Salmonella in the United States and is most often found in E. coli isolates, the most likely source was an E. coli donor, possibly from a food source or the patient's own intestinal flora. This study highlights the importance of plasmid biology in the epidemiology of enteric infections and the spread of antibiotic resistance in human pathogens.

Funding Statement

A portion of this work, including some of the efforts of Michael G. Jobling, was funded by HHS | NIH | National Institute of Allergy and Infectious Diseases (NIAID) (AI-31940). Conjugation experiments were performed by M.G.J. while supported by NIH grant AI-31940 to Randall Holmes. The remainder of his contribution and those of all other authors received no specific contribution from any funding agency in the public, commercial, or not-for-profit sectors.