ESBL plasmids in Klebsiella pneumoniae: diversity, transmission, and contribution to infection burden in the hospital setting

Ethical approval

The primary study was approved by the Alfred Health Human Research Ethics Committee (AHHREC), Project numbers #550/12 (19 February 2013) and #526/13 (10 December 2013). Analysis of clinical isolates 2017-2020 was approved by AHHREC, Project number #371/19 (2 July 2019).

Specimen collection

The primary KpSC collection comprised 440 Klebsiella pneumoniae, Klebsiella variicola and Klebsiella quasipneumoniae isolated at the Alfred Hospital Microbiological Diagnostic Laboratory from patients at hospitals in the Alfred Health network, Melbourne, during a prospective surveillance study (KASPAH) between April 2013 and March 2014 (Supplementary Table 1). KASPAH study protocols are reported elsewhere [13–15]. Briefly, gut carriage (colonising) isolates (n=108) were cultured from rectal screening swabs in either the Alfred Hospital ICU (33% of patients screened) [13] or Caulfield Hospital geriatric wards (31% screened) [14]. Infection isolates (n=332) represent all KpSC clinical isolates identified by standard diagnostic protocols across the study period [15]. In January to March 2014, 3GCR Gram-negative carriage and infection isolates were also collected from Alfred Hospital ICU patients as described previously [16], and those identified as Enterobacteriaceae were included here (n=74). All isolates were subjected to antimicrobial susceptibility testing using Vitek2 GN AST cards (bioMérieux) and interpreted using EUCAST cut-offs (2020).

To determine whether KpSC lineages harbouring a plasmid of interest (Plasmid A) during the primary study period persisted 3-6 years later, we interrogated all 3GCR KpSC clinical isolates identified by the same diagnostic laboratory between March 2017 and December 2020 (n=450, identified as K. pneumoniae or K. variicola by MALDI-TOF (Bruker) and resistant to ceftriaxone (MIC ≥2 mg/mL)).

Whole-genome sequence (WGS) analysis

DNA was extracted and sequenced using the Illumina platforms as previously reported [13–16]. A subset (n=70) of ESBL+ isolates from the KASPAH study were selected for additional sequencing via the Oxford Nanopore Technologies MinION platform as described previously [17]. These included at least one representative of each unique combination of multi-locus sequence type (ST) and ESBL gene. Representatives of bla_CTX-M-15 positive KpSC STs identified in the 2017-2020 follow-up period were also subjected to MinION sequencing. Genomes were assembled using Unicycler v0.4.7 [18] and annotated with Prokka v1.14 [19]. Completed genome assemblies were deposited in GenBank and Illumina read sets in the NCBI Sequence Read Archive (accessions in Supplementary Table 1). STs were identified for each genome using Kleborate v1.1 [20] for KpSC, and mlst v2.19 (https://github.com/tseemann/mlst) for Escherichia. AMR genes were detected using Kleborate v1.1 [20].

Identification and comparison of ESBL/carbapenemase plasmids

For each ESBL or carbapenemase gene identified in multiple genomes, the plasmid sequences harbouring them were compared pairwise using two similarity metrics: (i) Mash similarity (1–[Mash distance], calculated using Mash v2.1.1 [21]), and (ii) gene content Jaccard similarity ([genes in common]/[total genes in either plasmid]). Pairs of plasmids with Mash similarity ≥0.98 and gene content similarity ≥0.8 were designated the same plasmid (thresholds determined empirically, see Supplementary Figure 1). Plasmid replicon markers and MOB genes were detected using Mobtyper v1.4.9 [22], and copy numbers estimated by Unicycler.

For n=418 genomes in the 2017-2020 follow-up collection, we used read mapping to determine the presence of Plasmid A, represented by pINF329. Illumina reads were mapped to pINF329 (accession LR890241) using RedDog (https://github.com/scwatts/reddog-nf) to determine coverage and depth of the plasmid sequence as described previously [23]. Genomes were considered Plasmid A-positive if they had ≥80% mapping coverage and <10 single nucleotide variants (SNVs) compared to the pINF329 reference sequence (n=133 genomes).

Phylogenetic analyses

We generated core-genome phylogenies for Plasmid A, and chromosomes of K. pneumoniae ST323 (n=40), ST29 (n=17), ST2856 (n=10) and K. variicola ST347 (n=54). Illumina reads were mapped to completed reference sequences generated from study isolates (pINF329 for Plasmid A, accession LR890241; INF018 for ST323, accession LR890493; INF250 for ST29, accession LR890374; KPN342 for ST2856, accession CP089384; INF345 for ST347, accession LR890399) using RedDog to identify SNVs as described above. For additional context, chromosome SNV alignments were supplemented with genomes of the same ST from public isolate collections [20]. Phylogenies were inferred using RAxML v8.2.9 [24] with a general time-reversible substitution model. See Supplementary Methods for further details.

ESBL plasmid diversity

To determine the genetic context of ESBL genes, we generated complete closed reference sequences for n=67/104 ESBL+ genomes (64%), including at least one for each unique combination of ESBL gene and species/ST, representing n=48/77 of all ESBL episodes (62%). Most of the completed KpSC genomes (n=48/52, 92%) carried a plasmid-borne ESBL gene (including n=14 with an additional chromosomal copy); the rest (n=4/52, 8%) carried a chromosomal ESBL gene only (bla_CTX-M-15). Similarly, amongst Escherichia genomes, 87% (n=13/15) carried a plasmid-borne gene (including n=2 with an additional chromosomal copy) and 13% (n=2/15) carried a chromosomal ESBL gene only (bla_CTX-M-15).

Using pairwise similarity scores to compare all 61 completed ESBL plasmid sequences (see Methods) we identified 25 distinct plasmids. Those found in only a single genome were labelled with their host isolate, and those found in ≥2 genomes were labelled Plasmid A, Plasmid B etc. Each ESBL plasmid was restricted to either KpSC (n=12) or Escherichia (n=13) (Table 1). The majority (n=17, 68%) were IncF type plasmids, ranging in size from 45–245 kbp with estimated copy number 1.3–3 per cell, and accounting for 70% of the bla_CTX-M genes that were identified (Table 1, Supplementary Table 2). Other notable ESBL plasmids included two IncN plasmids carrying bla_CTX-M-62 (both copy number ∼5x; one 59 kbp, the other 102 kbp and also harbouring an IncR replicon marker), and a small (10 kbp) high copy number (16x) plasmid carrying bla_CTX-M-14 in E. coli ST48 (Table 1, Supplementary Table 2).

Ten of 12 KpSC ESBL plasmids carried additional AMR genes (median 8.5 AMR genes per plasmid); however, only half (n=6/13) of the Escherichia ESBL+ plasmids carried additional AMR genes (median 6 per plasmid) (Table 1, Supplementary Table 2). Nearly all (n=45/48, 93%) of the completed ESBL plasmid-carrying KpSC genomes carried at least one additional plasmid (range 0-9 plasmids), and 70% carried at least three (Figure 2b). The majority of additional plasmids did not carry AMR genes, except for two STs carrying carbapenemase plasmids: K. pneumoniae ST340 with bla_IMP-4 (n=4 isolates) and K. pneumoniae ST231 with bla_OXA-48 (n=8 isolates) (Supplementary Table 2, Supplementary Text). All of the completed ESBL plasmid-carrying E. coli genomes carried at least one additional plasmid (range 1-8) (Figure 2b), the majority of which lacked AMR genes (n=8/13, 62%).

Plasmid transmission network

We used the completed genomes to investigate the evidence for nosocomial transmission of ESBL plasmids between distinct KpSC strain backgrounds (defined by ST), by constructing a network in which nodes represent ESBL+ plasmids identified in KpSC (Figure 3). The vast majority of unique ESBL plasmids were detected in either a single genome (n=13/25) or a small number of genomes of a single ST (n=11/25, 2-8 genomes each) (Table 1, Figure 3). However, Plasmid A (which carried bla_CTX-M-15) was identified in 41 isolates belonging to four K. pneumoniae STs (ST323, ST29, ST5822, ST221) and K. variicola (ST347, see Figure 3). Plasmid A was highly conserved between isolates, with median pairwise gene content similarity 0.93 (range 0.84-1) and median Mash distance 0.998 (range 0.996-1; see Figure 3, Supplementary Table 2). Median 0 SNVs were detected amongst Plasmid A sequences (19 identical, range 0-54 SNVs), regardless of host chromosomal ST, with only occasional differences in plasmid gene content (Supplementary Figure 2). Taken together, these data support transmission of Plasmid A between multiple K. pneumoniae lineages and one K. variicola lineage.

• Download figure
• Open in new tab

Figure 3: ESBL plasmid cluster network for all completed KpSC ESBL+ genomes.

a, bla_CTX-M-15 plasmids; b, bla_CTX-M-14 plasmids; c, other detected ESBLs. Each node represents a completed genome (from one representative isolate per patient), coloured by the bacterial host cell ST as per legend. Dashed outline surrounding dot indicates if the ESBL is carried on both a plasmid and the chromosome. Nodes are connected if they share a similar plasmid sequence, defined as mash similarity ≥0.98 and gene content similarity ≥0.8 (see Supplementary Figure 1); line colour and width indicate plasmid gene content similarity (as per legend). Clusters comprising ≥2 genomes are labelled with the median jaccard gene similarity score. ‘+bla_OXA-48 (Plasmid M)’ and ‘+bla_IMP-4 (Plasmid N)’ indicate STs that carry an additional carbapenemase plasmid (details in Supplementary Table 2).

Applying a similar analysis to the carbapenemase plasmids revealed an IncL plasmid (Plasmid M carrying bla_OXA-48, Supplementary Table 2) in two different host strains (E. coli ST38 and K. pneumoniae ST231) from the same patient. The sample collection dates were consistent with transfer of Plasmid M from K. pneumoniae ST231 to E. coli ST38 in this patient’s gut (Supplementary Figure 3, Supplementary Text).

Dissemination of Plasmid A and contribution to ESBL burden

Overall Plasmid A was present in 51% (n=29/57) of KpSC ESBL+ episodes (53% of ESBL+ infections) during the study period. The remaining 49% (n=28/57) of KpSC ESBL+ episodes were attributed to plasmids/strains isolated from one patient only (Figure 5a). K. pneumoniae ST323 was the most common host strain for Plasmid A (n=22). Plasmid A-positive ST323 were detected throughout the entire study period and accounted for 29% of ESBL+ KpSC episodes (25.5% of infections, 50% of carriage) (Figure 4). Plasmid A-positive ST323 isolates from distinct patients differed from each other by 0-55 chromosomal SNVs (median 22, see Supplementary Figure 4), consistent with clonal transmission of this strain within the hospital network as previously reported [14].

• Download figure
• Open in new tab

Figure 4: Timeline of ESBL+ infection and carriage episodes.

Each episode is represented by a single isolate, grouped in rows by the isolate ST. Colours indicate the ESBL enzyme encoded in the genome (as per legend), and shapes indicate whether the episode was carriage (circle) or infection (triangle). Presence of the ESBL plasmid most commonly associated with the ST is indicated by filled shapes (with plasmid name listed under ST on the y-axis). ESBL genes on the chromosome rather than plasmid are indicated by shapes with no fill and a coloured outline. Grey box indicates that no data was available for this period for the Escherichia STs. Kp – K. pneumoniae; Kv – K. variicola; Ec – E. coli; Em – E. marmotae.

• Download figure
• Open in new tab

Figure 5: Presence of Plasmid A over time.

a, Bar height indicates total number of ESBL episodes across the initial year-long study period, by month. Bars are coloured by Plasmid A positive ST episodes as per legend, and grey for non-Plasmid A ESBL episodes. b, Timeline of Plasmid A positive episodes during the initial year-long study period. Each point represents an episode, with colour indicating ST (as per legend) and shape indicating whether the episode was infection (circle) or carriage (triangle). c, Bar height indicates the total number of 3GCR KpSC episodes per month from March 2017 – December 2020. Bars are coloured by ST as per legend if the episode was positive for Plasmid A.

We hypothesise that Plasmid A was originally hosted in ST323 and subsequently transmitted into four other KpSC STs. This is supported by identification of an outgroup ST323 genome (EuSCAPE_IT380, accession ERR1217451, isolated in Italy in 2013 [25]) carrying Plasmid A (2 plasmid SNVs, 59 chromosomal SNVs, Supplementary Figure 4). Additionally, two ST29 isolates collected in months 4 and 7 did not carry Plasmid A (INF206 and INF122, Figure 4), and were evolutionarily distant from the 11 Plasmid A-positive ST29 genomes recovered in months 8-12 (INF206=10,341 core genome SNVs, INF122=106 SNVs; Supplementary Figure 4) [14]. Similarly, a Plasmid A-negative K. variicola ST347 isolate collected in month 8 (INF238, Figure 4) was distant to subsequent ST347 isolates (35,020 core genome SNVs; Supplementary Figure 4). Analysis of the Plasmid A backbone sequence showed that the variants present in ST29, ST5822, ST221 and ST347 genomes were clustered in a single sub-clade of the Plasmid A phylogeny, whereas ST323 variants were more diverse and present both within and outside of this sub-clade (Supplementary Figure 2).

We detected only a single episode each of ST221 and ST5822 with Plasmid A, but the data supported nosocomial transmission of the ST29 and ST347 strains following acquisition of Plasmid A (median chromosomal SNVs between isolates from different patients = 1 for both STs, range 0-5). Assuming all four STs acquired Plasmid A from ST323, these plasmid transmission events and subsequent onward transmission of the resulting ESBL+ strains would account for 23% (n=13/57) of KpSC all ESBL+ episodes (28% of ESBL+ infections) during the one-year study period, including 54% of those in the last 3 months (Figure 5a,b).

Long-term persistence of Plasmid A in diverse strain backgrounds

We screened for Plasmid A in additional WGS data from 3GCR KpSC clinical isolates collected 3-6 years later from the same hospital laboratory (n=418, representing 92% of all 3GCR KpSC clinical isolates during March 2017 – Dec 2020). These 418 isolates represented 365 infection episodes (57% urinary tract, 13% bacteraemia, 12% respiratory, 5.8% wound/soft tissue). Plasmid A was identified in 35% of all sequenced 3GCR isolates and 21.2% of all 3GCR infection episodes (24%, 21%, 42% and 8.5% of 3GCR urinary, bacteraemia, wound/soft tissue, and respiratory episodes, respectively). We observed the same three expanded STs seen in the original study (n=11 ST323, n=3 ST29, n=69 ST347), as well as two additional STs (n=11 ST2856, n=1 ST5823) (Figure 5c). Phylogenetic analysis of the host chromosomes was consistent with persistence of the plasmid during long-term local clonal expansion of the ST323, ST29 and ST347 host strains (Supplementary Figure 4). Inspection of completed genome sequences for one genome per ST (KPN029, KPN110, KPN392, KPN342, KPN692) confirmed that the more recent isolates carried Plasmid A, with few structural changes (Supplementary Figure 5a). Plasmid A was most prevalent from March 2017 to March 2018, accounting for 52% of 3GCR episodes (n=62/120, Figure 5c). From April 2018 onwards, Plasmid A episodes were rare (<10%), except for a cluster of infections between September 2018 and May 2019 during which time Plasmid A-positive ST347 accounted for 27.5% of 3GCR infections (Figure 5c).