Genomic and functional analysis of rmp locus variants in Klebsiella pneumoniae

Abstract

Background

Klebsiella pneumoniae is an opportunistic pathogen and a leading cause of healthcare-associated infections in hospitals, which are frequently antimicrobial resistant (AMR). Exacerbating the public health threat posed by K. pneumoniae, some strains also harbour additional hypervirulence determinants typically acquired via mobile genetic elements such as the well-characterised large virulence plasmid KpVP-1. The rmpADC locus is considered a key virulence feature of K. pneumoniae and is associated with upregulated capsule expression and the hypermucoid phenotype, which can enhance virulence by contributing to serum resistance. Typically such strains have been susceptible to all antimicrobials besides ampicillin; however, the recent emergence of AMR hypermucoid strains is concerning.

Methods

Here, we investigate the genetic diversity, evolution, mobilisation and prevalence of rmpADC, in a dataset of 14,000 genomes from isolates of the Klebsiella pneumoniae species complex, and describe the RmST virulence typing scheme for tracking rmpADC variants for the purposes of genomic surveillance. Additionally, we examine the functionality of representatives for variants of rmpADC introduced into a mutant strain lacking its native rmpADC locus.

Results

The rmpADC locus was detected in 7% of the dataset, mostly from genomes of K. pneumoniae and a very small number of K. variicola and K. quasipneumoniae. Sequence variants of rmpADC grouped into five distinct lineages (rmp1, rmp2, rmp2A, rmp3 and rmp4) that corresponded to unique mobile elements, and were differentially distributed across different populations (i.e. clonal groups) of K. pneumoniae. All variants were demonstrated to produce enhanced capsule production and hypermucoviscosity.

Conclusions

These results provide an overview of the diversity and evolution of a prominent K. pneumoniae virulence factor and support the idea that screening for rmpADC in K. pneumoniae isolates and genomes is valuable to monitor the emergence and spread of hypermucoid K. pneumoniae, including AMR strains.

Background

A distinct pathotype of Klebsiella pneumoniae, often referred to as hypervirulent K. pneumoniae (hvKp), poses a significant public health challenge outside of hospital settings where it causes severe and sometimes life-threatening infections [1,2,3]. These are regarded as being distinct from ‘classical Kp’ strains, which typically cause opportunistic infections mostly within healthcare settings, and are often multidrug-resistant (MDR). Community-acquired infections can arise in otherwise healthy and immunocompetent individuals, although there are reportedly associations with comorbidities such as diabetes [4]. Common examples of infections include pyogenic liver abscess, endophthalmitis, pneumonia and meningitis, but they can also present as metastatic, multisite infections. Earlier reports of hypervirulent, community-acquired infections were largely confined to countries in Eastern Asia. Cases are now being more widely reported in other regions including Europe, North America and Australia, although often associated with individuals of East Asian descent [5].

The overwhelming majority of hvKp infections are associated with strains from distinct genetic backgrounds or lineages; these include clonal groups CG23, CG25, CG65, CG66, CG86 and CG380 [6, 7]. Several features are considered hallmark characteristics of these hvKp strains [7]. Most produce a K1 or K2 capsule, encoded by the cps (K) loci KL1 and KL2, respectively, and O1 lipopolysaccharide (OL1 locus). Many hvKp strains also exhibit hypermucoviscosity (HMV), which is defined by a positive string test and/or low OD₆₀₀ measurements in a sedimentation assay, and is associated with presence of the rmpADC locus and/or rmpA2 gene. Lastly, most hvKp also synthesise the siderophores aerobactin (iuc locus) and salmochelin (iro locus), in addition to the intrinsic siderophore enterobactin (ent locus) [7, 8]. The acquired siderophore loci, rmpADC and rmpA2 (which also appears to be part of a locus including homologues of rmpD and rmpC), are typically mobilised by the large K. pneumoniae virulence plasmids (KpVP) that have been stably maintained for over 100 years in some hvKp clones including CG23 and CG86, although iro and rmpADC can also be mobilised via the chromosomal integrative conjugative element ICEKp1 [9,10,11].

In K. pneumoniae, the hypermucoid or HMV phenotype has long been associated with the gene rmpA although the exact mechanisms leading to the phenotype were unknown [12,13,14]. Based on observations and experimental evidence from earlier studies using knockout mutants, it was proposed that the expression of rmpA upregulated expression of cps thereby increasing capsule production and subsequently resulting in HMV. Subsequent work has confirmed that the rmpA LuxR-like transcriptional regulator is part of a larger operon (herein called rmp) together with the rmpD and rmpC genes located downstream [15, 16]. This work has further clarified that enhanced capsule expression and HMV are two discrete features requiring rmpC (regulator of cps) and rmpD (encoding a small protein required for HMV), respectively. While HMV can be attained in the absence of elevated cps expression (i.e. in a rmpC mutant), it was not observed in capsule-defect mutants, suggesting that HMV does rely on the presence of some capsular components but these need not be hyperexpressed. Accordingly, it was recently shown that HMV is driven through elongation of the capsule polysaccharide chain, driven by a direct interaction between RmpD and Wzc (transmembrane protein), and presumptive indirect interaction with Wzy (capsule repeat unit polymerase), both of which are components of the core capsule synthesis and export machinery [17, 18]. The rmpA and rmpA2 genes appear to be frequently subjected to insertions or deletions (indels) within a poly(G) tract that consequently encode a truncated and presumably non-functional product, and this has been suggested as a mechanism by which differential expression of the two genes is achieved [14].

The presence or absence of rmpA in clinical isolates has been investigated in dozens of studies focused on hypervirulent infections, and rmpA has been identified as one of several biomarkers for distinguishing hvKp strains from non-hvKp [19]. However, their detection shows a variable degree of correlation with HMV as measured by string test (51–98%). The functional impact of allelic variation in the rmp locus genes has not yet been explored, although this is likely to be important for explaining or predicting HMV based on rmp locus sequences. Here, we investigate the genetic diversity and distribution of the rmp locus in the K. pneumoniae species complex (KpSC), identify key variant lineages of rmp and their associated mobile genetic elements (MGEs), and demonstrate that representatives of each of the lineages are able to induce HMV and elevated capsule production when introduced into an hvKp isolate lacking its native rmp locus (KPPR1S $\Delta$ rmp).

Methods

Genome sequences and genotyping

The initial screening for rmpADC was conducted on the same 2733 K. pneumoniae species complex (KpSC) genomes included in a 2018 study [10] examining the aerobactin- and salmochelin-encoding loci iuc and iro, which are often co-localised with rmp on the same mobile genetic elements. The SRST2-table-from-assemblies.py Python script (https://github.com/rrwick/SRST2-table-from-assemblies) [20]

was used to screen these assemblies for the presence of existing rmpA alleles from the virulence database on BIGSdb-Kp (37 alleles as of October 2020; https://bigsdb.pasteur.fr/klebsiella/) [21] along with the reference rmpD and rmpC sequences from Walker et al. [15, 16] with BLAST+ v2.2.21 (≥90% coverage and identity), and novel alleles extracted with the --report_new_consensus flag. Unique alleles from 160/2733 genomes with a typeable rmp locus (defined as those in which all genes in the locus could be assigned an allele) were assigned allele numbers, and unique allele combinations were used to define unique ‘rmp sequence types (RmSTs)’ using a multi-locus sequence typing (MLST) approach. These rmpADC alleles and RmST profiles were then incorporated along with rmpA2 alleles into the genotyping pipeline Kleborate v2.0.0 and applied to screen 13,156 publicly available Klebsiella genomes [22]. From this dataset of 13,156 genomes, 944 genomes positive for rmp and/or rmpA2 (including genomes with an incomplete rmp locus) were included for analysis in this study (see Additional file 1: Table S1 for genome accessions, isolate metadata and genotyping information). We also included rmp-positive genomes from three additional datasets: 4/208 isolates collected at an Australian hospital in 2002 [23], in addition to 36/392 isolates from the same hospital in 2020, and 8/276 isolates from the Burden of Antibiotic Resistance in Neonates from Developing Societies (BARNARDS) network [24]. Novel rmpADC alleles and RmST profiles were added to the BIGSdb-Kp database. Clonal groups, and designations of clones as hypervirulent or MDR, were done using previously defined ST-CG assignments [25].

Phylogenetic analyses

For each unique RmST, an alignment of the concatenated rmpA, rmpD and rmpC sequences was generated with MUSCLE v3.8.31 and used as an input for maximum likelihood phylogenetic inference with RAxML v8.2.9 [26] run five times with the generalised time-reversible GTR+gamma model. Similarly, phylogenies were also generated for rmpA, rmpD and rmpC genes individually using the same approach described for RmST. Rmp lineages were defined based on monophyletic groups of RmSTs mobilised by a unique mobile element (see below). Phylogenetic trees for genomes belonging to hypervirulent clonal groups CG23, CG65 and CG86 were generated using a core genome distance approach implemented in Pathogenwatch (pathogen.watch) [27]. In brief, the genomes for each CG of interest were uploaded to Pathogenwatch, matched/aligned against the K. pneumoniae reference core library (2,172,367 bp) to generate a matrix of pairwise distances that were then used to build a neighbour-joining tree.

Comparison of rmp genetic contexts

All contigs containing rmp were manually inspected in Bandage v0.8.1 [28] to determine whether the locus was located on the chromosome or on a previously described virulence plasmid (KpVP-1 reference pSGH10 with accession CP025081.1 for contigs with rmp lineage 1, i.e. rmp1, or KpVP-2 reference Kp52.145 plasmid II with accession FO834905.1 for rmp2 contigs; both are reference sequences that have been widely used for lab and/or genomic studies). For any assemblies where the rmp-containing contig did not match to any of the lineages described above, the contig sequence was screened against the NCBI non-redundant nucleotide database and top hits (i.e. BLASTn results with the highest E value corresponding to hits with the highest coverage and identity) noted. A single annotated representative (complete sequence where possible) was selected for each rmp lineage to compare the overall genetic structures of the neighbouring regions of rmp (i.e. up to 15 kbp upstream and downstream of rmp). Annotations were performed with Prokka v1.14.6 [29] and comparisons between the annotated genes were visualised with clinker v0.0.21 (github.com/gamcil/clinker) [30].

Bacterial strains, plasmids and growth conditions

The K. pneumoniae strains and plasmids used in this study are listed in Table 1. The strains from which representative rmp was amplified were as follows: SGH10 for rmp1, 52.145 for rmp2, NCTC 13669 for rmp2A, KPPR1S for rmp3 and NCTC 1936 for rmp4. Strains were grown at 37 °C in LB medium (10 g tryptone, 5 g yeast extract, 10 g NaCl; BD Difco, Ref# 244630). Saturated overnight cultures were diluted to an OD₆₀₀ of 0.2 and grown for 5.5 h. Antibiotics were used where appropriate: kanamycin (Kan), 50 μg/ml; rifampin (Rif), 30 μg/ml; spectinomycin (Sp), 50 μg/ml. All plasmids were introduced into K. pneumoniae by electroporation as previously described [15]. For expression of genes cloned into the pMWO-078 vector, 100 ng/ml anhydrous tetracycline (aTc) was added to the medium at the time of subculture.

Table 1 Strains and plasmids used in this work

Primers used for the construction of expression vectors are listed in Additional file 2: Table S2. All rmp loci were amplified from genomic DNA by PCR and then cloned into pMWO-078 [32] by Gibson assembly (NEB). The rmp expression vectors were then transformed into KPPR1S ∆rmp. KPPR1S (ST493; K2/O1) is a well-characterised streptomycin and rifampicin-resistant derivative of ATCC 43816 and has been used in previous studies to characterise rmpD, rmpC and HMV [15,16,17]. Wildtype KPPR1S, ∆wcaJ and ∆rmp strains transformed with pMWO-078 were also included as control strains for measuring HMV. The wcaJ mutant does not produce capsule and is HMV negative, while the rmp mutant produces capsule but is HMV negative as it does not produce RmpD.

Assessment of hypermucoviscosity

Saturated overnight cultures of strains containing rmp plasmids were subcultured in fresh LB for 5.5 h at 37 °C with 100 ng/ml anhydrotetracycline (ATc) to induce rmp expression. Cultures were normalised to 1 OD₆₀₀/ml and centrifuged at 1000 × g for 5 min. Mucoviscosity of cultures was determined by normalising OD₆₀₀ of the culture supernatant to the starting culture as previously described, and an OD₆₀₀ = 0.25 considered hypermucoviscous [33].

Uronic acid measurement

Uronic acid (UA) was measured following an established protocol [34,35,36] from cultures grown as described for assessing HMV. Briefly, UA was extracted from 500 μl of culture with 1% zwittergent in 100 mM citric acid at pH 2, precipitated with 1.2 ml 100% ethanol, and resuspended in 1.2 ml 12.5 mM sodium tetraborate dissolved in sulphuric acid. Twenty microlitres of 0.15% 3-phenylphenol in 0.5% NaOH was added and absorbance at 520 nm was measured. UA amounts were determined from a standard curve generated with glucuronolactone (from 1500 to 50 µM).

Results

Prevalence of rmp in the K. pneumoniae species complex

Screening for the rmp locus in publicly available KpSC genomes (n = 15 species/subspecies) identified its presence in 992/13,993 genomes across only three KpSC species: K. pneumoniae (980/11,967, 8.2%), K. quasipneumoniae subsp. similipneumoniae (Kqs; 5/522, 1.0%) and K. variicola subsp. variicola (Kv; 7/626, 1.1%) (see Additional file 2: Table S3). The majority of these included complete rmp loci with intact coding sequences for rmpA, rmpD and rmpC (i.e. presumably functional variants, detected in 73% of rmp+ genomes); however, some loci were incomplete (i.e. deletion variants missing at least one gene within the locus or arising from assembly fragmentation) or had nonsense mutations resulting in premature stop codons truncating one or more of the encoded proteins (i.e. truncation variants) (see Additional file 2: Table S3). Fifteen genomes were identified with multiple rmp loci (two rmp each); the majority of these carried at least one functional rmp locus (73%; eight genomes with two functional rmp loci and three genomes with one functional rmp) while the remaining four genomes each had two non-functional rmp loci arising from deletion and/or truncation variants. BLASTn search of all non-redundant bacteria in NCBI with default parameters identified complete rmp loci in four non-Klebsiella genomes, in all cases rmp was present in plasmids sequenced from Escherichia coli transconjugants that had been mated with a K. pneumoniae strain carrying KpVP-1-like virulence plasmids (GenBank accessions MN200130.1, MN182750.1, MZ475697.1, CP068571.1). Incomplete rmp loci (44% coverage) were detected in an additional five plasmids that were also sequenced from E. coli transconjugants.

Genetic diversity of the rmp locus

Each of the genes comprising the rmp locus showed some degree of genetic diversity, with 85 rmpA alleles, 77 rmpD alleles and 42 rmpC alleles observed across rmp+ genomes. For the 825 genomes carrying an intact ‘typeable’ rmp locus (838 loci total accounting for genomes with two rmp loci), allelic variants could be assigned to all three genes, and resulted in 170 unique combinations which were each assigned a unique RmST (rmp sequence type). The rmp loci from the remaining 167 genomes were designated ‘non-typeable’ due to the locus being incomplete (i.e. missing at least one gene or comprising a fragmented gene to which an allele could not be designated).

Maximum likelihood phylogenetic analysis of the concatenated rmpA, rmpD and rmpC sequences belonging to each unique RmST revealed the grouping of sequences into five distinct lineages, which we labelled rmp1, rmp2, rmp2A, rmp3 and rmp4 (Fig. 1). These lineages were labelled as such to match with their associated iro and/or iuc lineages (i.e. rmp1 is associated with iuc1 and iro1). The nucleotide divergence between lineages ranged from 0.7 to 11% (mean 4.7%), decreasing to 0–5.9% within lineages (mean 0.4%). No rmp gene alleles were shared between lineages (see individual gene phylogenies in Additional file 2: Fig. S1), suggesting an absence of recombination between the lineages. The rmp3 lineage was notably divergent from the other rmp lineages, with a mean divergence of 8.4% compared to 1.8% between all other lineages (Table 2). Further, the rmpD alleles of the rmp3 lineage were longer than those of other lineages, measuring 176–177 bp in length compared to the 151–162 bp alleles observed in rmpD from the other lineages. No associations between rmp lineages with geography or host were observed.

Fig. 1

Maximum-likelihood phylogenetic tree of rmp sequence types (RmSTs). Lineages are labelled and tips are coloured by the associated mobile genetic element according to the legend. Column labels are as follows for a particular RmST: presence or absence of truncations in the rmpA, rmpD and rmpC, detection within a hypervirulent (blue) or MDR (red) clone (definitions of clones from Wyres et al. [25]) or non-K. pneumoniae species (Kv, Klebsiella variicola; Kqs, Klebsiella quasipneumoniae subsp. similipneumoniae). Clones included in ‘Other MDR’: CG14, CG17, CG29, CG101, CG147, CG152 and CG307. The number of genomes from which each RmST was detected is shown in the bar graph on the right-hand side

Table 2 Nucleotide divergence within (bolded) and between rmp lineages. Values given are range (mean) and expressed as percentages

Diversity of rmp-associated mobile genetic elements

Each of the rmp lineages were associated with unique plasmids or chromosomal contexts (see Fig. 1), many of which have been previously characterised in a study examining the aerobactin- and salmochelin-encoding loci iuc and iro [10], which are typically co-localised with rmpA2 and rmp respectively on the same mobile elements. The dominant K. pneumoniae virulence plasmid, KpVP-1 (associated with iuc1 and iro1 loci) was associated with the rmp1 lineage and accounted for the majority of rmp carriage (785 genomes, 79.1%). Next most common were the iuc2A virulence plasmids (associated with iuc2A) [10], carrying rmp2A lineage (77 genomes, 7.8%); followed by ICEKp1 carrying rmp3 (58 genomes, 5.8%); and KpVP-2 virulence plasmids with rmp2 (associated with iuc2 and iro2, 45 genomes, 4.6%). Two additional genetic contexts were observed in this study: (i) the rmp4 lineage was chromosomally encoded and associated exclusively with K. pneumoniae CG67 genomes (six genomes, identified as ST67 or single-locus variants thereof, also known as K. pneumoniae subspecies rhinoscleromatis [37]) and (ii) presence of rmp3 into the yersiniabactin-encoding ybt4-type plasmids (three genomes; note partial deletion of rmp locus in two of these genomes). For the remaining genomes, one was associated with a KpVP-2/iuc2A hybrid plasmid as previously described [10], nine carried multiple rmp loci (due to presence of both KpVP-1 and ICEKp1), and 17 were unresolved due to assembly issues. KpVP-1 was detected in 785 genomes, where it typically carried both rmp and rmpA2 (669/785). However, variants of KpVP-1 were common (i.e. coverage of KpVP-1 reference ranged from 26.7% coverage in rmp1+ genomes): we observed 116/785 genomes with rmp only without rmpA2 (20 and 32 of these lacking an intact iro1 and iuc1, respectively). An additional 180 genomes carried rmpA2 only without rmp (849 rmpA2+ genomes in total). Nine genomes appeared to have chromosomal integration of KpVP-1 (n = 7 ST11/LV, n = 1 ST23 and n = 1 ST383; see Additional file 1: Table S1).

The genetic context of rmp in representatives of each of the key rmp lineages is shown in Fig. 2. The rmp locus was located adjacent to the iro locus and peg-344 gene (which has been mis-labelled as pagO in some studies) in most cases, with the exception of the ybt4 plasmids (which lack iro and peg-344 genes) and iuc2A plasmids (which carry iroB only). The iro locus is typically intact in KpVP-1, KpVP-2 and ICEKp1, but is partially disrupted by insertion sequences (IS) in the ST67 (rhinoscleromatis) chromosome and iuc2A plasmids. IS flank the rmp/iro region in all contexts; however, the specific IS vary and it is not clear which have played a role in mobilising rmp and/or iro. Notably, all contexts include IS3 on one or both ends of the rmp/iro region. The closely related rmp1, rmp2, rmp2A and rmp4 lineages all have IS3 in a conserved position adjacent to rmpA; this also appears to be conserved in the divergent lineage rmp3, suggesting it was likely present in the common ancestor of all rmp lineages and may have played a role historically in the mobilisation of rmp/iro between genetic backgrounds (see Fig. 2). However, all rmp/iro regions have additional IS near or within the locus, some of which may have contributed to further mobilisation and/or degradation of the distinct lineages over time. While the assembly of the rmp region in the ybt4 plasmid was incomplete, an IS3 fragment was detected next to rmp.

Fig. 2

Comparison of upstream and downstream regions of rmp associated with different mobile variants. Arrows represent coding sequences and those corresponding to genes of interest are labelled and coloured by functionality as per the legend. Other labelled loci include arsCBR (arsenic resistance), virB1-virB11 (virB-type 4 secretion system) and ybtSX (part of the yersiniabactin locus). Shading corresponds to regions of similarity (amino acid sequence identity ≥ 30%) as identified by clinker

Hypermucoviscosity of rmp variants in a KPPR1S background

To determine if there were functional differences between the level of HMV conferred by the rmp lineages, we ectopically expressed variants of the rmp locus cloned into pMWO-078 in an HMV-negative rmp deletion mutant KPPR1S ∆rmp. Wildtype KPPR1S is a ST493 K2/O1 isolate containing rmp3 (ICEKp1), which is the most divergent rmp lineage (Fig. 1); KPPR1S ∆rmp therefore provides a good background to assess potential functional differences between rmp variants. HMV was assessed using a sedimentation assay. The ∆wcaJ and ∆rmp vector controls fully sedimented, whereas the KPPR1S vector control did not sediment well (Fig. 3A). Ectopical expression of the native locus, rmp3, conferred elevated HMV in ∆rmp compared to the KPPR1S vector control. This elevated phenotype is likely due to the overexpression of rmp, and this complemented strain therefore serves as a reference to compare the effects of expressing the other rmp loci. Expression of the rmp genes from rmp1, rmp2 and rmp2A also conferred elevated HMV levels in ∆rmp (Fig. 3A), and there were no statistically significant differences in sedimentation resistance between these strains versus that expressing rmp3. However, expression of rmp4 conferred a significantly lower level of HMV compared to the other lineages.

Fig. 3

Hypermucoviscosity and capsule production of ectopically expressed rmp loci from the five different lineages. A Mucoviscosity assay and B uronic acid assay of KPPR1S, ΔwcaJ and Δrmp strains with vector (pMWO-078) or lineage-specific pRmp (see Table 1). Data from two technical replicates and three biological replicates were obtained following a 5.5-h induction of KPPR1S expression plasmid-borne rmp genes as described in Methods. One-way ANOVA with Tukey’s post-test was used to determine significance. *, P < 0.05; ***, P < 0.001; ****, P < 0.0001; comparisons not indicated were not significant

Capsule production of rmp variants in a KPPR1S background

Elevated capsule production is an rmpC-dependent phenotype in KPPR1S, independent of HMV, whereby RmpC upregulates capsule expression via the promoters upstream of the galF and manC genes in the K locus, and KPPR1S ∆rmp mutants display decreased capsule production [15, 16]. To further investigate any functional differences between rmp variants, we assayed the same transformants described above for uronic acid (UA) production, which is used as an indicator of the amount of capsule [34]. KPPR1S, ∆wcaJ (i.e. capsule-negative) and ∆rmp strains were included as controls. Expression of rmp from each of the lineages increased capsule production (Fig. 3B). Expression of rmp1, rmp2A and rmp4 in the ∆rmp mutant increased capsule production to the same level observed when expressing the native rmp3 locus (Fig. 3B). Expression of rmp2 produced slightly but significantly higher uronic acid levels compared to rmp3 (P < 0.05). Collectively, these data suggest that there are subtle but significant functional differences between the rmp lineages.

Distribution of rmp in the K. pneumoniae species complex

The rmp locus was detected in 129 unique K. pneumoniae STs, 3 Kqs STs and 5 Kv STs, representing 143 unique combinations of STs and rmp lineages. As expected, the prevalence of rmp is typically quite high within known hvKp clones (≥80% prevalence within any given ST assigned to a hypervirulent clonal group with the exception of CG25, which has recently been shown to comprise of two distinct lineages [38]; mean 83.6%) (Fig. 4). Rmp was also common in CG67 (rhinoscleromatis, 100% rmp4; 25% truncated) and CG91 (ozaenae, 79% rmp2A incomplete or truncated); both lineages have previously been defined as subspecies due to their distinct pathotypes [37]. However, relatively high rmp prevalence was also observed in some ‘generalist’ (i.e. non-hvKp and non-MDR) K. pneumoniae clones (for generalist clones with n ≥ 5 genomes: 1.6–100% rmp prevalence, mean rmp prevalence = 40.2%, median rmp prevalence = 15.1% [IQR = 5.6–86.4%]). Conversely, the MDR clones together with CG36 and CG45, which are also clones that are commonly detected in healthcare settings, had relatively lower frequencies of rmp (mean prevalence = 11.4%).

Fig. 4

Distribution of rmp across K. pneumoniae sequence types with ≥5 rmp+ genomes. Rows indicated K. pneumoniae sequence types (STs; as labelled), and grouped and labelled by clonal group (CG) where applicable; those corresponding to hypervirulent, pathotype subspecies or multidrug-resistant (MDR) clones are labelled accordingly. Other clones are considered ‘generalist’ clones. The bubble plot shows the number of rmp+ genomes assigned to a particular rmp lineage, coloured by the type of mobile element. The first bar plot shows prevalence of rmp (i.e. black: present, grey: absent) and the second shows functional status of the locus (i.e. black: intact and functional, red: non-functional due to deletions or truncations)

KpVP-1-rmp1 was the most widely disseminated and was not only detected in the hypervirulent clones from which they were initially characterised but also in MDR clones. Numbers of KpVP-1-rmp1 were notably high in ST11 (211 genomes) and ST15 (50 genomes), reflecting recently reported MDR-hypervirulent convergent variants of these well-known MDR clones [39]. KpVP-1 also accounted for the majority of rmp acquisitions in the other KpSC species where the genetic context could be resolved. In comparison, the other mobile elements appeared in relatively fewer STs, but also appeared to be fixed in most of these clones that likely serve as native hosts to the respective mobile elements (see Fig. 4). Of the fifteen genomes that carried multiple rmp loci, ten were ST23 genomes that carried ICEKp1 (rmp3) on the chromosome and the KpVP-1 (rmp1) virulence plasmid.

Amongst rmp+ genomes with a confident K locus call (917/992), 38 different K loci were detected (Fig. 5). Half (i.e. 19) were detected in two or fewer genomes. The most common K loci were KL1 (319 genomes), KL2 (203) and KL64 (140), which collectively account for 72.2% of genomes with a confident K locus, followed by KL57 (45) and KL20 (40). The dominant K loci are largely driven by their associations with over-represented STs (Fig. 5), many of which are known hvKp lineages such as KL1 in CG23 (which harbours KpVP-1) and KL2 in CG380 (KpVP2), CG65 and CG86 (KpVP-1). The next most common K locus, KL64, is associated with a sublineage of MDR clone ST11 that is widespread in China, known to have acquired variants of KpVP-1 [39].

Fig. 5

Distribution of rmp within K locus groups and KpSC clonal groups (CG) or species. Each circle represents an rmp+ genome with a confident K locus call (917 genomes) and is coloured by the functionality of the rmp locus as per the figure legend. Twenty-five K loci with 4 or fewer genomes were collapsed into ‘Other’: KL6, KL10, KL12, KL14, KL15, KL17, KL19, KL21, KL23, KL24, KL25, KL27, KL30, KL34, KL35, KL39, KL45, KL55, KL63, KL102, KL108, KL127, KL137, KL147 and KL149. Clonal group labels are coloured blue and red to indicate hypervirulent and MDR lineages, respectively. Kq: Klebsiella quasipneumoniae, Kv: Klebsiella variicola

Given previous reports and observations highlighting the common occurrence of truncations in rmpA due to insertions and deletions (indels) within a poly-G tract [14], we next examined the rmp locus for the occurrence of truncations. Truncations were detected in 30/85 rmpA, 9/77 rmpD and 16/42 rmpC alleles across 63, 16 and 55 genomes, respectively (Additional file 2: Fig. S1). The majority of truncations (45/55 truncated alleles) were caused by indels within homopolymer tracts resulting in premature stop codons arising from frameshift mutations (Additional file 3: Table S4). Further, some homopolymer regions had a higher frequency of mutations compared to others, including the previously reported poly-G tract in rmpA spanning nucleotide position 276 to 285 whereby indels were detected in 16 rmpA variants across 46 genomes. The frequency of truncated rmp loci compared to intact loci was lower within the hypervirulent clones and appeared to be more common outside of these clones (Figs. 4 and 5). Closer inspection of the phylogenies for hypervirulent clones CG23, CG65 and CG86 (Additional file 2: Figs. S2–S4) revealed a random distribution of truncated rmp alleles throughout these populations. Truncations in rmpA2 (n = 745/849) were also common, outnumbering intact variants (n = 95 rmpA2 intact; note n = 9 with truncation status not known due to fragmented rmpA2 sequences) and were detected at high prevalence (>55 to 100%) across both hypervirulent and MDR clones.

Discussion

We previously reported briefly on the sequence variation of the rmp locus, which clustered into four lineages and were each associated with distinct MGEs [22]. In this study, we provide detailed insights into the phylogenetic relationships, genetic contexts, distribution and phenotypic functionality of rmp variants. Building on the initial rmp scheme, we characterised an additional rmp4 lineage observed only in genomes belonging to a sublineage of K. pneumoniae (CG67), updating the number of rmp lineages to five.

The 1:1 correlation typically observed between rmp lineages and the MGEs that mobilise them (with few exceptions including rmp3 on ybt4 related plasmids, and the chromosomal versus plasmid variants of rmp2A) also extends to iro, shown here to be typically located adjacent to rmp (Fig. 2), and to iuc, which is often located elsewhere on the same plasmids [10]. This association highlights the co-evolution of these virulence loci, and presumably other common genes shared between the MGEs. While specific details or the order of events in the evolutionary history of these MGEs cannot be determined, similarities in the genetic contexts surrounding rmp do suggest a shared ancestry and the role of IS, particularly IS3, in the mobilisation of rmp/iro between different MGEs. Interestingly, the presence of IS3 upstream of rmpA2 has been flagged as being necessary for complete activation of the promoter for rmpA2 and may likewise serve a similar purpose for rmpA [40]. We also observed three instances of a novel genetic context for rmp, where a variant of rmp3 (ICEKp1) had been introduced into a yersiniabactin plasmid (i.e. another virulence MGE) [41] likely via IS3-mediated transposition (Fig. 2). Outside of K. pneumoniae, rmp (KpVP-1 and ICEKp1) was only detected in two other KpSC species at rare frequencies and did not appear to have been acquired naturally in any other species. K. pneumoniae is therefore likely to be the original host for rmp, although the same may not hold true for the other virulence loci given that iro and iuc variants have been detected in other non-Klebsiella species including Enterobacter spp. and Escherichia coli [10].

Even with the larger genome dataset screened in this present study, the prevalence of rmp overall (7.5%) and each of the MGEs was very similar to that reported in our earlier study characterising iuc (8.7%) and iro (7.2%) (i.e. 14,000 versus 2700 genomes). The rmp1 lineage (KpVP-1) was by far the most dominant accounting for 80% of the rmp burden, in part driven by their association and maintenance via clonal expansion within the key hypervirulent clones such as CG23 [11], but also due to the more recent transmission and spread within MDR clones such as ST11 and ST15 (Figs. 1 and 4). Based on growing reports of ‘hypervirulent ST11’ from China, it is possible that KpVP-1 and variants thereof are also being stably maintained within this clone following its acquisition. Variants of KpVP-1 which include deletion or recombinant variants, and chromosomal integrations, are commonly reported in ST11 and have been proposed to confer a reduced fitness cost [39, 42, 43]. A more recent study has highlighted at least two plasmid variants of KpVP-1 (referred to as ‘PIUC1-IncFIB(K)₃₇’ and ‘pIUC1-IncFIB(Mar)’) that account for the majority of hypervirulent carbapenem-resistant K. pneumoniae in China [44]. In another study investigating the virulence plasmid in a collection of ST11 genomes, the authors noted as low as 25% coverage of their chosen KpVP-1 reference (pK2044, accession AP006726) across recombinant KpVP-1 plasmids, and 19–41% coverage for chromosomal integrations of KpVP-1 [41]. A similar range in coverage was also observed in this dataset not only in ST11 but also other clones, and manual inspection of these assemblies suggested the presence of several KpVP-1 plasmid variants as well as chromosomal integrations. With the exception of KpVP-1-rmp1 and ICEKp1-rmp3, which were detected in multiple STs/CGs, the remaining MGEs appeared to be stably conserved within a small number of clones in which they were detected (93.9–100% prevalence; Fig. 4).

Each of the rmp lineages was shown to be functional, resulting in HMV and elevated capsule production when a representative of each was introduced into a single strain background, KPPR1S $\Delta$ rmp. Notably, this ST493 strain has a K2 capsule, which is one of two dominant capsule types amongst known hvKp clones such as CG86 (Fig. 5). We also observed variability in the extent of HMV and capsule production for different rmp, which further reiterates that HMV and capsule production are two separable traits [15, 16]; for example, the expression of rmp4 does not restore HMV to the same level as KPPR1S rmp3 but does restore capsule production (Fig. 3). It is unclear if expressing these rmp (or other representatives of the same lineages) in other KpSC strains with different serotypes will have the same impact, especially given the extensive diversity of Wzc in different K loci and its interaction with RmpD. In another study, expression of KpVP-1 rmp in strains with different capsules (KL1, KL2 and KL64) yielded differences in HMV and capsule production [45]. These insights will be particularly useful for predicting the impacts of rmp acquisition in other clones, including those considered to be MDR, and is the subject of ongoing work by our team.

Loss of function mutations of rmpA arising from indels within the poly(G) tract has been well documented in many studies [14], and while this site (i.e. bases 267 to 285 of the rmpA_2 reference) does account for the majority of rmpA truncations in this dataset, indels within additional homopolymer tracts in rmpA, rmpD and rmpC were also observed (Additional file 3: Table S4). Loss of function mutations in the rmp genes were observed in at least 145 genomes (14.6% of rmp+ genomes, including three with multiple rmp), most often in the non-hypervirulent clones. Further, we observed parallel evolution of the same loss of function mutations on multiple occasions within different hypervirulent clones (e.g. allele rmpA_4 in CG23, CG65 and CG86; Additional file 2: Figs. S2–S4). Taken together, these findings highlight the potential reversibility of homopolymer tract mutations, which may take place to help alleviate the negative selection pressure following acquisition of KpVP-1 or other MGEs. Importantly, these loss of function mutations also need to be carefully considered when interpreting data based solely on PCR detection of rmpA/A2, and may partly explain the discrepancies in the literature reporting on the association between rmp presence and HMV, although other reasons also include inconsistencies and the unreliability of string testing, which is heavily influenced by temperature dependencies and strain genetic background [46].

Conclusions

Our findings reveal that, similar to the other key virulence loci co-localised on the same MGEs, genetic variation within the rmp locus is highly structured and this information can be harnessed to track novel rmp acquisitions. To this end, detection and genotyping of the rmp genes (i.e. RmST typing), alongside the reporting of locus disruptions, has already been implemented in our genotyping tool for KpSC genomes, Kleborate (github.com/klebgenomics/Kleborate). The tool also outputs a virulence score that is currently calculated from the presence of various siderophores (ybt and iuc) and colibactin (clb), and does not take into account rmp (or iro). Given the apparent role of rmp in hypervirulence, the virulence score in addition to rmp presence should both be considered when assessing the virulence of a given strain or genome. Ongoing research investigating the expression of rmp variants in different strain and capsule backgrounds will yield additional important insights into the function of rmp, particularly for novel strains such as those from the MDR clones that acquire them, and can be used to further improve the genotyping output for rmp.