IS26-mediated plasmid reshuffling results in convergence of toxin–antitoxin systems but loss of resistance genes in XDR Klebsiella pneumoniae from a chronic infection

Carbapenem-resistant Enterobacterales pose an urgent threat to human health worldwide. Klebsiella pneumoniae sequence type (ST) 14, initially identified in the Middle East and South-Asia and co-harbouring the carbapenemase genes bla OXA-232 and bla NDM-1, is now emerging globally. One such strain was detected in the USA in 2013 from a patient initially treated in India that also carried armA, a 16S rRNA methyltransferase that confers resistance to all clinically relevant aminoglycosides. Genetic and phenotypic changes were observed in 14 serial isolates collected from this chronically infected patient. The index isolate carried five plasmids, including an IncFIB–IncHI1B (harbouring armA and bla NDM-1), an IncFIA (bla CTX-M-15) and a ColE-like (bla OXA-232), and was extensively resistant to antibiotics. Four years later, a subsequent isolate had accumulated 34 variants, including a loss-of-function mutation in romA, resulting in tigecycline non-susceptibility. Importantly, this isolate now only carried two plasmids, including a large mosaic molecule made of fragments, all harbouring distinct toxin–antitoxin systems, from three of the canonical plasmids. Of the original acquired antibiotic resistance genes, this isolate only retained bla CTX-M-15, and as a result susceptibility to the carbapenems and amikacin was restored. Long-read sequencing of a subset of five representative isolates, collected between 2013 and 2017, allowed for the elucidation of the complex plasmid patterns and revealed the role of IS26-mediated plasmid reshuffling in the evolution of this clone. Such investigations of the mechanisms underlying plasmid stability, together with global and local surveillance programmes, are key to a better understanding of plasmid host range and dissemination.


INTRODUCTION
Bacterial antimicrobial resistance has emerged as one of the leading public health challenges of the 21st century [1]. The global increase in carbapenem-resistant Enterobacterales (CRE), best illustrated by the rapid spread of 'high-risk' Klebsiella pneumoniae lineages, is universally regarded as a critical priority by international and national health agencies [2].
At the origin of the emergence of carbapenem-resistant K. pneumoniae (CRKP) lineages is the acquisition, mostly via conjugative plasmids, of carbapenemases, generally from one of three classes: the class A carbapenemases exemplified by the K. pneumoniae carbapenemases (KPCs), the class B metallo-β-lactamases (IMP, NDM and VIM) and the class D OXA-48-type oxacillinases [3]. While sequence type (ST) 258 complex (often carrying an IncFII plasmid harbouring a KPC carbapenemase) has become one of the most successful lineages of CRKP globally [4], those carrying NDM and OXA-48-type enzymes dominate in some regions of the world, including the Middle East and South-Asia [5][6][7]. One consequence of the high prevalence of these two groups of carbapenemases was the emergence of strains carrying both NDM-1 and OXA-48 enzymes (including its variants OXA-181 and OXA-232) [8,9]. While problematic alone, these warrant specific surveillance as a major reservoir of non-KPC carbapenemases in an era where KPC-selective inhibitors are being used as first-line treatment options for CRKP [10].
Lineage K. pneumoniae ST14, which recently became prevalent in some regions of the USA [11], often carries bla OXA-232 and sometimes co-harbours bla NDM-1 [12]. In 2014, the first report of such an extensively drug-resistant (XDR) clone in the USA documented the carriage of bla OXA-232 by a ColE-like plasmid with both bla NDM-1 and the 16S rRNA methyltransferase gene armA on a hybrid IncFIB-IncHI1B plasmid [13,14]. The latter shared similarities with plasmid pNDM-MAR, the representative of a plasmid family first described in a K. pneumoniae from Morocco in 2012 [15]. Since then, members of this family have been detected globally and have played a major role in the spread of bla NDM-1 [16,17], including as occasional co-carriers of biomarker genes previously linked to hypervirulence [18,19].
Contrasting with the rapid spread and increased prevalence of these carbapenemase-carrying plasmids globally, various in vitro and in vivo studies [20][21][22] suggest that their presence has a negative impact on the fitness of K. pneumoniae, possibly explaining their observed instability in the absence of selective pressure [23]. Investigations of the mechanisms underlying plasmid stability, preferentially carried out on relevant clinical specimens, are important to directly improve our understanding of the impediments and opportunities for rapid dissemination. Here, starting from the same ST14 K. pneumoniae strain previously carrying NDM-1 and OXA-232 [13,14], we trace the genetic and phenotypic (AST) changes observed in 14 serial ST14 isolates collected from the same patient over the span of 5 years.

Bacterial isolates
The index isolate PittNDM01 (hereafter named MRSN 546052 to reflect the exact isolate sequenced in this study) was previously characterized as sequence type 14 (ST-14) and coproducing NDM-1 and OXA-232 [13,14]. For this study, the complete set of 14 serial K. pneumoniae isolates from the same patient were collected throughout multiple visits at the same hospital system in Pittsburgh (PA) ( Table 1). Bacterial species identification and antimicrobial susceptibility testing (AST) were carried out using MicroScan Walkaway (Beckman Coulter) or Vitek 2 (bioMérieux). Isolates were sent to the Multidrug Resistant Organism Repository and Surveillance Network (MRSN) for further phenotypic characterization and genome sequencing. Confirmatory AST was performed in the MRSN College of American Pathologists (CAP)-accredited clinical laboratory using the Vitek 2 (card GN AST 71; bioMérieux, NC, USA). In addition, minimum inhibitory concentrations (MICs) of fosfomycin were determined in triplicate using broth microdilution based on Clinical and Laboratory Standards Institute (CLSI) guidelines.

Impact Statement
Carbapenem-resistant Klebsiella pneumoniae is a significant nosocomial threat. This study provides a rare glimpse into the plasmid evolution of an extensively drug-resistant clone of K. pneumoniae that has persisted within a single host for an extended period of time. The source isolate harboured two carbapenemases on separate plasmids. Five years later, the same lineage of K. pneumoniae had lost both carbapenemases and reverted to a susceptible phenotype. Analysis of closed plasmid sequences revealed the critical role of the insertion sequence IS26 in plasmid reshuffling. Importantly, the presence of toxin-antitoxin (TA) systems did not prevent the loss of plasmid backbones and regions carrying the resistance genes. Instead, IS26-mediated rearrangements allowed for the convergence of selected sequences, all carrying the TA system from their ancestral plasmids, into a single, large, mosaic molecule. Improving our knowledge of the mechanisms of plasmid plasticity and stability is key to better understand the role of these molecules, and genetic determinants within them, in the persistence and spread of associated antibiotic resistance genes.

Whole-genome sequencing and de novo assemblies
Isolates were sequenced on a MiSeq benchtop sequencer (Illumina, Inc., San Diego, CA). DNA was extracted using the DNeasy UltraClean microbial kit (Qiagen, Germantown, MD, USA), and libraries were constructed using the Kapa HyperPlus library preparation kit (Roche Diagnostics, Indianapolis, IN, USA) as previously detailed [24]. Samples were sequenced using MiSeq reagent kit v3 (600 cycles; 2×300 bp) (Illumina). Five isolates were selected for single-molecule real-time (SMRT) sequencing using a PacBio RS II instrument (Pacific Biosciences, CA, USA). MiSeq read data were processed with bbduk to trim Illumina adapters and filter reads by quality prior to assembly. Reads falling below an average of 15 Phred score within a 5 bp sliding window and reads shorter than 100 bp were excluded. Filtered reads were assembled de novo using Newbler 2.9 and minimum thresholds for contig size and coverage were set at 200 bp and 49.5×, respectively. PacBio read data were assembled de novo using a hierarchical genome assembly process (HGAPv3.0). Overlapping contig ends were removed to circularize individual PacBio contigs. Short-read data were mapped to circularized contigs to correct errors.

Genome annotation, phylogeny and antimicrobial resistance (AMR) gene detection
The genome of index isolate MRSN 546052 was annotated using Prokka 1.14.6 and variant calling was performed by MrSNP-plus 1.0.3 using the index isolate as a reference. Single-nucleotide polymorphism (SNP)-based trees were generated with RaxML-ng using snippy core genome alignment of all 14 isolates, as detailed previously [18]. AMRFinderPlus v3.9.8 (https://www.ncbi.nlm. nih.gov/pathogens/antimicrobial-resistance/AMRFinder/) and ARIBA v2.14.4 [25] were used to identify resistance alleles in all isolates (Table S1, available in the online version of this article).

Plasmid typing, predicting toxin-antitoxin systems, mapping and construction
Each individual extrachromosomal contig from PacBio long-read sequencing was considered a plasmid and replicon sequences were identified using PlasmidFinder 2.1 [26]. Toxin-antitoxin systems were predicted using the web-based tool TAfinder [27]. Full plasmid sequences were extracted and analysed with a combination of software and online tools. Insertion sequences (ISs) were detected using the ISfinder tool (https://isfinder.biotoul.fr/) [28]. Rearrangements within hybrid plasmids were visualized in Geneious Prime 2020.2.4 and proposed intermediate molecules were constructed with SnapGene Viewer 5.3.1.

Patient history and phylogenetic analysis reveal long-term colonization by multidrug-resistant (MDR) K. pneumoniae
A 69-year-old patient was hospitalized in India in January 2013 for management of subarachnoid haemorrhage. A month later, the patient was transferred to an acute care hospital in Pittsburgh, PA, USA, and subsequently discharged to a long-term care facility. In March 2013, the patient was readmitted to the same US hospital presenting with a fever and a suspected urinary tract infection (UTI). A urine culture grew the index CRKP (MRSN 546052) and was positive by PCR for bla NDM-1, bla OXA-232 and armA.
The patient was administered oral fosfomycin for treatment of the UTI. No additional NDM-targeting antibiotic was administered through 2015, though positive cultures from surveillance swabs revealed the patient was still colonized with CRKP throughout this period ( Table 1). In that time frame, the patient had extensive exposure to antibiotics (most common regimens included piperacillin-tazobactam, ceftolozane-tazobactam, cefepime, ceftriaxone and meropenem) targeting other multidrug-resistant organisms (MDROs). In 2016, the patient had two episodes of line-associated bacteraemia with MDR K. pneumoniae, which was successfully treated with line removal and administration of ceftazidime-avibactam. A year later, the patient presented with a recurring K. pneumoniae UTI and was treated with oral fosfomycin. From these admissions (2013-2017), a total of 14 isolates from various sites were collected and sent to the MRSN for genome sequencing and analysis (Table 1).
Phylogenetic analysis on the core genome revealed that all 14 isolates were highly genetically related (average distance to nearest neighbour was 3.8 SNPs) (Fig. 1). Relatedness was generally correlated with the date of isolation; the more recent isolates (i.e. post-2016) were more related to each other (average of 1.4 SNPs to nearest neighbour) than they were to the 2013 index isolate (average of 35 SNPs). In June 2016, a cluster of five nearly genetically identical isolates were collected from one respiratory and four blood samples within days of each other. The only other blood isolate (MRSN 546131) was collected in a preceding episode of bacteraemia (February 2016) and was less genetically related to the other blood isolates (average of 18 SNPs) than to the three isolates (average of 7 SNPs) collected from urine cultures throughout 2017 (Fig. 1).

Variable carriage of AMR genes in serial isolates correlates with phenotypic resistance
The index isolate MRSN 546052 was phenotypically characterized as XDR with resistance to carbapenems, cephalosporins, β-lactam/β-lactamase inhibitor combinations, and all aminoglycosides [14]. Nine of the serial isolates had an identical susceptibility profile, including all blood isolates collected in 2016 (Fig. 1). In contrast, four isolates (blood isolate MRSN 546131 and phylogenetic neighbours MRSN 546195, MRSN 546201 and MRSN 546210) were susceptible to carbapenems and amikacin but  Table 1 are abbreviated to the last three digits. Phenotypic susceptibilities to select antibiotics are shown (squares) as well as the presence/absence of relevant AMR genes (box coloured so as to designate the corresponding carrying plasmid). For the five isolates with complete genomes, the presence of plasmids (named according to their size in kb and pertinent AMR genes carried) is indicated (coloured circles). Hybrid plasmids are depicted as pie charts with each segment from a canonical plasmid featured as a coloured slice (proportional to its size).
retained resistance to cephalosporins and other aminoglycosides. Interestingly, and unlike all other serial isolates, the three most recently collected urine isolates were non-susceptible to tigecycline (Fig. 1).
With the exception of tigecycline, the acquired AMR genes carried by the isolates correlated well with antibiotic susceptibility (Fig. 1). Index isolate MRSN 546052 carried the carbapenemase genes bla NDM-1 and bla OXA-232 , the 16S methyltransferase gene armA and the ESBL gene bla CTX-M- 15 . In contrast to bla CTX-M-15 , which was conserved in all isolates, only two carried both carbapenemase genes and, after June 2013, all successive isolates lacked bla OXA-232 . Blood isolate MRSN 546131, which was the first to show susceptibility to carbapenems and amikacin, lacked bla NDM-1 and armA, and both genes were also missing in the three urine isolates sharing the same susceptibility profile and the highest genetic relatedness (Fig. 1). In addition to bla NDM-1 and armA, these four isolates simultaneously lost other resistance markers (Table S1), suggesting that a large region of DNA, most likely of plasmid origin, was missing.
Overall, none of the four other isolates analysed shared genetically identical molecules to the plasmids (hereby referred as canonical) in MRSN 546052, with the exception of p2.1, which was conserved in all. In the case of pMR6.1-OXA232, complete plasmid loss was observed in all four isolates analysed and explained the lack of bla OXA-232 in the majority of isolates from this patient (Fig. 1).  (Figs 1 and 2c). Finally, MRSN 546195 carried the least plasmid content, both in number of plasmids and total kilobases, as fragments (each carrying their distinct TA systems) of all pMR283.4-NDM, pMR103.7 and pMR70.8-CTXM canonical molecules converged into a single IncFIA-IncFIB mosaic plasmid named pMR206.6-CTXM (Figs 1 and 3). In the process, bla CTX-M-15 was retained but bla NDM-1 and armA were lost.

IS26 drives plasmid rearrangements
Mechanistically, the inter-and intra-plasmids rearrangements that generated the various hybrid plasmids in isolates MRSN 546113,-122, -140 and -195 were likely mediated primarily by IS26 transposition events. In all cases, the proposed routes for the genesis of each hybrid plasmids involved pMR284.7-NDM (first described in MRSN 546113, Fig. 2a) Fig. 2c), either subsequent or antecedent to the recombination event, resulting in inversion of the region containing bla NDM-1 and duplication of IS26.
Finally, in MRSN 546195 (susceptible to carbapenems and amikacin) the proposed genesis of the IncFIA-IncFIB hybrid plasmid pMR206.6-CTXM first involved a trans-oriented intramolecular transposition event in pMR284.7-NDM, where a ~63.5 kb segment was inverted by an IS26-mediated attack (step 1, Fig. 3). A duplication of the attacking IS26 and resulting TSD [29] were at the origin of a newly formed composite transposon, highlighted in the proposed intermediate molecule I1 (Fig. 3). At step 2, this transposon appears to have inserted into the canonical IncFIB pMR103.7 (with the remaining backbone of IncFIB-IncHI1B pMR284.7-NDM carrying bla NDM-1 , armA and TA system RelE/PHD being lost), where a subsequent cis-oriented intramolecular attack by IS26 resulted in the loss of a ~36 kb region encoding a type IV secretion system and the associated TSD (step 3 and

Within-patient evolution and emergence of fosfomycin and tigecycline resistance
Whole-genome variant detection and annotations were performed to provide a more comprehensive overview of the evolution of this clone of K. pneumoniae. Variants were classified as singleton (observed once) or shared (observed more than once) across the 14 isolates. For each isolate, the total number of singleton or shared SNPs (i.e. excluding indels) was plotted as a function of days, since the index culture and linear regression showed an accumulation of shared variants at ~11 SNPs per year (Fig. 4a), similar to rates previously described for K. pneumoniae [21,30]. Seventy shared variants were identified in addition to 35 singleton variants ( Fig. 4b and Table S2). The distribution of the predicted variant impact was similar between shared and singletons with the majority (~40 %) being missense mutations and 12-15 % being predicted loss-of-function (LOF) mutations. Specifically, a total of 15 genes carried either a frameshift or a stop codon mutation in ≥1 isolates (Fig. 4c). Predicted LOF due to mutations in the genes coding for a subunit of the dimethyl sulfoxide reductase (dmsB), a formate-nitrine transporter (focA), a putative lipoprotein (00897), a d-lactate dehydrogenase (dld) and a putative pseudocatalase (00951) were acquired early (first detected in June 2015) and shared by all isolates thereafter. While the impact of these variants remains unclear, a one-nucleotide substitution in romA (part of the romA-ramA locus), resulting in a premature stop codon, was observed in the three most recent isolates and was likely responsible for their acquired phenotypic resistance to tigecycline via upregulation of an efflux-mediated pathway, as described previously [31] (Figs 1 and 4c). Similarly, five predicted LOF mutations were uniquely found in isolate MRSN 546113 (collected from a rectal swab in April 2015), including a frameshift in glpT that encodes the main transporter responsible for the uptake of fosfomycin (Fig. 4c) [32]. While this mutation possibly emerged as a result of the oral fosfomycin regimen prescribed on at least four different occasions between 2013 and 2017 (Table 1), no noticeable phenotypic impact was observed in vitro, as all 14 isolates were highly resistant to this antibiotic (MIC >256 µg ml −1 ) due to the presence of a conserved chromosomal fosA allele.  Fig. 1. Antimicrobial resistance genes (red), insertion sequences (grey), plasmid replicon ori sites (green) and putative transposases (white) are indicated, as well as genes at the boundary of recombination or insertion events (yellow). Segments involved in insertion, inversion and/or recombination are shaded (grey arrow), and directionality of event is shown (black arrow).

DISCUSSION
In this study, longitudinal sampling of a chronically infected patient showed that a ST14 CRKP clone, initially harbouring bla NDM-1 and bla OXA-232 on separate plasmids, rapidly lost both carbapenemases through complete plasmid loss and IS26-mediated plasmid reshuffling, eventually restoring carbapenem susceptibility. This echoes previous observations that plasmids carrying AMR genes generally impose a fitness cost on their bacterial hosts in the absence of selective pressure [21,33,34]. Specifically, the plasticity of K. pneumoniae plasmids has been extensively described using global collections or serially passaged isolates in vitro [35,36].
Recently, additional reports on the evolution of plasmids within the host are emerging [20][21][22]37]. For example, in 2017, an analysis of 15 K. pneumoniae isolates from six patients revealed the loss of bla KPC in one patient, and an overall significant reduction in plasmids due to recombination events in isolates from multiple patients [20]. Other studies, generally focused on serial K. pneumoniae isolates from patients with chronic infections and prolonged length of stay, have also reported decreased copy number or complete loss of plasmids in the absence of antibiotic pressure [21,34].
Acknowledging this large body of evidence for the biological cost of carrying plasmids, the 'plasmid paradox' was theorized to represent the evolutionary dilemma posed by the persistence, over the long term, of conjugative plasmids instead of the fixation of the select beneficial genes into the bacterial chromosome [38,39]. In such a context, the repeated identification of K. pneumoniae isolates carrying plasmid-borne bla NDM-1 and bla OXA-232 carbapenemases [5][6][7]14] appears even more paradoxical. Interestingly, a recent study showed that the simultaneous presence of a large bla NDM-1 (IncFIB/IncHI1B) and a smaller bla OXA-232 (ColE-like) plasmid, sharing a high level of genetic relatedness with pMR283.4-NDM and pMR6.1-OXA232 analysed here, increased the fitness and virulence of a transconjugant Escherichia coli host in vitro, while a single plasmid did not [40]. This is reminiscent of observations from San Millan et al. who showed that co-infection of a large and a small plasmid could invoke positive epistasis, minimizing the cost associated with carrying multiple plasmids [41] and could contribute to their relatively high prevalence in nature.
An important element in the persistence and stability of plasmids is toxin-antitoxin systems [42,43] and our observation of a rapid, complete loss of plasmid pMR6.1-OXA232 (the only molecule lacking a TA system) in these ST14 isolates further illustrates that. Interestingly, TA systems harboured by all the other AMR-carrying plasmids did not prevent the loss of bla NDM-1 and armA in isolate MRSN 546195. Instead, all TA systems (with the exception of RelE/PHD, which is likely explained by cross-reactivity with the antitoxin from HigB/HigA [44]) were conserved via the convergence of fragments of their respective canonical plasmids into the mosaic pMR206.6-CTXM molecule.
The detailed molecular analysis of circularized plasmid sequences from serial isolates collected in this study revealed that IS26 is a major contributor to the genetic reshuffling of carbapenemase-carrying plasmids. The IS26 family is common in carbapenemresistant Enterobacterales [45] and has been implicated in plasmid reorganization [29] through well-characterized replicative and/or conservative transposition mechanisms [29,46,47]. ISs are well known to introduce adaptive traits into plasmids and a recent study even characterized IS26-flanked pseudo-composite transposons as the likely primary contributor to the genetic reshuffling of bla NDM [16,48]. However, the work by Porse et al. also showed that IS26 is a driving force of plasmid persistence in novel hosts [35]. Specifically, they showed that IS26-mediated deletions of costly regions, such as type 4 secretion systems (T4SSs), from a plasmid backbone could improve plasmid stability and effectively expand its host range [49]. Interestingly, in our study, genes coding for a T4SS were also lost in the intermolecular and intramolecular shuffling, leading to the emergence of plasmid pMR206.6-CTXM. Consistent with our observations, such restructuring events are thought to come at the expense of conditionally useful components (e.g. the loss of armA and bla NDM-1 ) that might ultimately limit plasmid dissemination.
A few considerations for this study should be noted. First, our observations are derived from selective culture (i.e. single colonies) of serial rectal surveillance swabs and other clinical specimens, which risks underestimating diversity [50], including the possible co-carriage of carbapenem-resistant and carbapenem-susceptible ST14 strains within or between the different body sites. In fact, this could explain an apparent resurgence of CRKP cultures in January 2018 in this patient (unfortunately isolates were not available for sequencing), after all urine isolates collected in 2017 lacked bla NDM-1 and bla OXA-232 . Second, although exposure to tigecycline was not documented, the patient's medical history while outside of the hospital (e.g. while at a LTC facility) is unknown. It cannot be excluded that additional antibiotic regimens were prescribed and favoured the emergence of tigecycline resistance via the inactivation of romA, a mutation known to arise in response to treatment [51,52]. Third, while congruent with observations reported in other studies [8,29,34,35,53], the proposed routes for plasmid streamlining observed in these isolates from a single patient likely only represent a partial picture of the many ways IS elements, and IS26 in particular, drive plasmid evolution by maintaining genetic plasticity.
As such, further investigations of the mechanisms underlying plasmid persistence and the role of transposable elements herein are needed to understand and prevent our current epidemic of multidrug resistance. As demonstrated here, the elucidation of the complex repetitive plasmid patterns is now achievable using long-read sequencing. The advancement of these technologies opens up many avenues from the assessment of plasmid maintenance, spread and diversity, to the integration of such efforts into global and local surveillance programmes.