Preferred evolutionary routes of convergence in Klebsiella pneumoniae favor siderophore acquisition over hypervirulence

TL;DR

This study shows that Klebsiella pneumoniae strains evolve by gaining siderophore genes, not hypervirulence, to become more successful in spreading.

Contribution

The study identifies a universal evolutionary route involving siderophore acquisition in convergent Klebsiella pneumoniae strains.

Findings

Convergent Klebsiella strains commonly acquire IncFIB(Mar)/IncHI1B plasmids with incomplete virulence markers.

Aerobactin siderophore locus is universally acquired across 25 convergence events.

Convergent isolates show enhanced siderophore production but lack hypervirulence in vivo.

Abstract

The rise of Klebsiella pneumoniae combining antimicrobial resistance and virulence genes poses a major health threat, but the evolutionary routes and phenotypic consequences of this convergence are poorly understood. Here, phylogenetics of 1,468 isolates and population analysis of 7,520 plasmids, from >50 countries through the last two decades, reveal that convergence follows preferred, constrained evolutionary paths. The dominant route involves multidrug-resistant classical K. pneumoniae acquiring conjugative IncFIB(Mar)/IncHI1B plasmids carrying an incomplete set of virulence biomarkers. Across 25 independent convergence events, the acquisition of the aerobactin siderophore locus was the only universal feature. These convergent isolates exhibit enhanced siderophore production but consistently lack the hypervirulent phenotype in vivo. In contrast, genuine hypervirulent strains that gain resistance remain rare. We conclude that enhanced siderophore production, not hypervirulence, is the primary adaptive trait driving the success of globally emerging convergent lineages, representing a distinct evolutionary state optimized for transmission rather than systemic invasion.

Figures

A](#F1), blue triangles). Assemblies revealed 173 plasmid sequences, including 172 circular and 1 linear sequence, with between 1 and 7 plasmids identified per isolate (Table S3). Replicons were typed and, using a cluster-based approach ([31](#R31)), all study (MRSN) plasmid sequences were compared to each other and to a publicly available set of 7,347 complete *K. pneumoniae* plasmids (from genomes in NCBI, representing 306 STs from at least 47 countries throughout the last two decades) (Fig. S3, Table S4).

D](#F2)). Cluster pC4 was strongly enriched (50% vs 0.001% in other pCs, p < .01) for plasmids carrying all 5 virulence biomarkers ([Fig. 2D](#F2)), including the canonical, non-conjugative, pLVPK plasmid associated with phenotypic hypervirulence. Unlike plasmids found in pC4, pC2 plasmids rarely carried the complete set of virulence biomarkers (1%) and were more frequently associated with the carriage of only *iuc* and *rmpA2* (23%). In agreement with this population-level analysis, 68% of virulence gene-carrying plasmids characterized in this study belonged pC-4 and pC-2 ([Fig. 3](#F3), Tabl

AB](#F3)). The distribution of replicon types, STs, virulence genes, ESBL and/or carbapenemase genes were investigated and visualized for all plasmid sequences located in the 28 discrete clusters (n = 4,052) and the 44 singletons from this study ([Fig. 2](#F2), Table S5).

A](#F5)). The only exception was route Y, a ST-25 isolate from Georgia harboring *iuc3* on a pC23 plasmid. Across convergence routes, isolates that acquired an incomplete set of virulence genes on a pC2 plasmid (routes G-N) produced siderophore levels comparable to isolates harboring the canonical pC4 virulence plasmid (routes A-F; p > 0.05) ([Fig. 5AC](#F5)), a pattern associated with the *iuc1* aerobactin allele (Table S6). By contrast, isolates that acquired other virulence plasmid types (routes O-Y), most frequently encoding *iuc3*, displayed significantly lower, though still elevated, sid

Full text

Introduction

Klebsiella pneumoniae are ubiquitous bacteria and the causative agent of diverse infections including pneumonia, urinary tract, surgical site, and bloodstream infections (1). Classical K. pneumoniae (cKp) with acquired resistance determinants are one of the world’s leading causes of healthcare-associated infections (2). Notably, third-generation cephalosporin-resistant and carbapenem-resistant K. pneumoniae are listed by the World Health Organization as urgently needing novel therapeutics (3). Antimicrobial resistance (AMR) genes, including extended-spectrum β-lactamases (ESBLs) and carbapenemases, are readily acquired by conjugative plasmids and mobile genetic elements (4–6). Carbapenem-resistant K. pneumoniae were initially associated with globally disseminated clone ST-258, encoding KPC enzymes found on conjugative IncFII(K)/IncFIB(pQil) plasmids (4, 7, 8), however, other lineages (e.g. ST-307, ST-147, ST-395) have recently emerged internationally carrying blaNDM, blaKPC, and/or blaOXA−48 genes on diverse plasmids (9–12).

A distinct pathotype from cKp, hypervirulent K. pneumoniae (hvKp) cause invasive community-acquired infections linked to high mortality, driven by metastatic spread, despite generally remaining amenable to antimicrobial treatment (13). hvKp infection outcomes are associated with five hvKp-specific virulence plasmid-encoded genes for siderophores aerobactin (iuc) and salmochelin (iro), mucoid and capsule polysaccharide regulators (rmpADC and rmpA2), and the metabolite transporter (peg-344) (14–18). Genetically distinct from cKp, hvKp clones have lower pan genome diversity, lower capsule type diversity (most commonly KL1 or KL2), and constitute fewer lineages (e.g., ST-23, ST-268, ST-65, ST-420), which are most prevalent in the Asia Pacific Rim (5, 6, 17). Conventionally, hvKp harbor non-conjugative virulence plasmids, mainly the described pLVPK or pK2044 with IncHI1B/IncFIB dual replicon (19, 20) or the rarer KP52.145pII plasmid with IncFIB replicon (21). However, acquired virulence genes are also found on chromosomal integrative conjugative elements (e.g. ICEKp10) and conjugative plasmids (6, 22).

While historically rare, the overlap of K. pneumoniae strains carrying both hvKp-specific virulence genes and carbapenemase and/or ESBL genes have been increasingly reported within the last decade (17, 23, 24). The genetic characteristics of reported convergent isolates include hvKp lineages (e.g. ST-23-KL1) acquiring conjugative resistance plasmids carrying AMR genes (24), or MDR-cKp lineages harboring pLVPK-like plasmids (25). Other evolutionary dynamics that have led to genomic convergence includes hybrid plasmids of the IncFIB(Mar)/IncHI1B dual replicon type carrying resistance (blaNDM−1, armA) and hvKp-specific virulence (iuc, rmpADC, rmpA2, and/or peg-344) biomarkers that are found in globally epidemic nosocomial MDR lineages such as ST-307 and ST-147 (26, 27).

Although some confusion remains, because of genetic-only inferences or inadequate virulence models (28), genomic convergence is not synonymous with phenotypic hypervirulence (12, 15, 27, 29). In this study we determine that the phenotypic outcomes of convergence in K. pneumoniae are lineage specific and the result of distinct evolutionary routes of plasmids acquisition. Using a unique global dataset of 1,468 multidrug-resistant isolates, and a population plasmidomic analysis of 7,347 complete plasmids (from strains spanning over 47 countries and the last two decades), we describe that convergence follows preferred paths, constrained by the plasmid biology, and that enhanced aerobactin-mediated iron acquisition, not hypervirulence, is the main adaptive trait driving the success of convergent lineages in hospital settings.

Results

Prevalence of MDR lineages in a global collection of Klebsiella pneumoniae

Global surveillance of multi-drug resistant K. pneumoniae (MDR-Kp) resulted in 1,468 clinical isolates recovered from 8 countries on 5 continents, between 2011 and 2021 (Fig. S1, Table S1). The frequency of isolate collection varied by year and country. In silico typing revealed a diverse population comprising 208 sequence types (ST) and 87 capsule types (KL) (Fig. S1, Table S1). Nevertheless, 8 lineages were represented by > 50 isolates and accounted for 57% (n = 834) of the population. ST-395 and ST-16 were the most prevalent (12% and 11%) and respectively grouped isolates that largely originated from the Republic of Georgia (99%) and Thailand (84%). In contrast, other major lineages (e.g. ST-15 and ST-147) were collected from multiple countries. Within those substantial genetic diversity was observed, with most isolates differing by > 100 allelic differences by core genome MLST (Fig. S2, Table S2). By contrast high genetic relatedness (≤ 10 allelic differences) was observed in geographically clustered subsets, including ST-395 from Georgia, ST-16 from Thailand, and subsets of ST-147 isolates from Thailand and Peru, consistent with local clonal expansion of emerging high-risk lineages.

Association of antibiotic resistance and virulence genes with global MDR-Kp lineages

For the resistance genes: in this collection, the rate of ESBL carriage remained remarkably stable, between 78 and 82% annually, and gene blaCTX−M−15 (74% of ESBL isolates) was the most represented (Fig. S1, Table S1). By contrast, the rate of carbapenemase-producing isolates increased markedly over time, from 6% of MDR-Kp isolates sampled between 2011–2016 to 44% (chi-square 175.5 and p < .01) between 2017–2021. This largely correlated with the sampling of clonal, geographically clustered isolates (Fig. S1, Table S1). Carabapenemase blaNDM genes were the most represented, alone (55%) or in combination with blaOXA−48 (22%).

For the virulence genes: 222 isolates (15%) carried at least one virulence biomarker (iuc, iro, peg-344, rmpADC, and/or rmpA2). By contrast, only 1% of MDR-Kp isolates collected from the U.S. military health system during this same time period carried an acquired virulence gene (30). Core genome phylogenetic analysis of the global virulence-carrying collection (n = 222) compared to virulence-carrying US isolates (n = 39, including 22 canonical hypervirulent ST-23 isolates) identified 32 distinct virulence gene-carrying lineages including ST-395 (n = 135), ST-23-KL1 (n = 29), ST-15 (n = 13), ST-23-KL57 (n = 13) and ST-2071 (n = 10) (Fig. 1). The prevalence of the 5 acquired virulence biomarkers was variable; all 261 isolates carried iuc siderophore loci, represented by four iuc alleles including the hvKp-associated iuc1 (n = 135), followed by iuc3 (n = 107), iuc5 (n = 17), and iuc2 (n = 2) (Fig. 1, Table S1). By contrast, the iro siderophore loci was only found in 23% (n = 60) of isolates, with the hvkp ST-23-KL1 isolates accounting for half. Most isolates carried only 1 biomarker (n = 145), mainly represented by lineage ST-395 from Georgia. Only 49 isolates carried all five biomarkers, predictive of the hvKp phenotype (15). Within those, ST-23-KL-1 was the most represented (n = 25) (Fig. 1, Table S1).

Increased prevalence of convergent isolates in recent years.

Convergence, defined by the co-carriage of an ESBL and/or carbapenemase gene in addition to ≥ 1 virulence biomarker (15, 17, 29), was detected for 198 (76%) of the virulence carrying isolates and was significantly enriched (chi-square 27.7 and p < .01) in the last 5 years of sampling (Table S1). Only 7 convergent isolates carried all 5 virulence loci. None of these 7 isolates carried a carbapenemase gene, instead all acquired an ESBL, including ST-420 (n = 2) and ST-268 (n = 1) from Thailand, ST-218 (n = 2) from Kenya, and ST-23-KL-1 (n = 1) from the United States (Fig. 1, Table S1). By contrast, 80% of the isolates with 1–4 virulence loci co-carried a carbapenemase gene. Isolates from Georgia were the most represented (89%) and two lineages were predominant: ST-395 (n = 111) and ST-23-KL57 (n = 13). Within each, differences in patterns of AMR and virulence genes were suggestive of distinct plasmid acquisition/loss events (Fig. 1).

Population analysis of virulence, resistance, or virulence/resistance carrying plasmids.

Complete genomes were obtained for 34 representative convergent isolates and 19 virulence-carrying isolates (Fig. 1A, blue triangles). Assemblies revealed 173 plasmid sequences, including 172 circular and 1 linear sequence, with between 1 and 7 plasmids identified per isolate (Table S3). Replicons were typed and, using a cluster-based approach (31), all study (MRSN) plasmid sequences were compared to each other and to a publicly available set of 7,347 complete K. pneumoniae plasmids (from genomes in NCBI, representing 306 STs from at least 47 countries throughout the last two decades) (Fig. S3, Table S4).

The K. pneumoniae plasmidome stratified into 43 discrete plasmid clusters (pC) (Fig. S.3). Within those, the study plasmids (n = 124) clustered in 28 pCs with 4,052 NCBI plasmids sequences (Fig. 2AB, Table S5). The remaining 49 MRSN plasmid sequences were singletons/unassigned.

Within the 28 pCs comprising at least one study plasmid, 22 grouped plasmids with an ESBL and/or carbapenemase gene (Fig. 2C). Cluster pC10, the most numerous, was very homogeneous and 82% of its plasmids carried an OXA-48 carbapenemase (Table S5). Plasmids within pC3, pC38 and pC39 were also frequent carriers of ESBL and/or carbapenemase genes (Fig. 2C), although a diversity of resistance alleles was noted (Table S5). In alignment with the population-level analysis, ESBL-carrying plasmids (n = 26) from this study were highly diverse and belonged to 11 pCs with pC-38 (n = 6, carrying CTX-M and IncR replicon) and pC-39 [n = 5, carrying CTX-M and IncFII(K)] being the most represented (Fig. 3, Table S3). Carbapenemase-carrying plasmids (n = 20) were also diverse, with 6 pCs represented, though the majority belonged to either pC-10 (n = 9 carrying OXA-48 and IncL/M replicon) or pC-2 (n = 4, carrying NDM and a IncFIB(Mar)/IncHI1B replicon). With a single exception (isolate 582610), conjugative plasmids pC2, −10, −38, and − 39 were exclusively found in MDR-cKp lineages (e.g. ST-395, ST-147 and ST-23-KL57) irrespective of geographical origins (Fig. 3, Table S3).

Contrasting with the breadth of diversity of resistance plasmids, only 7 of the 28 pC carried plasmids with virulence biomarkers and over 80% grouped in either pC4 or pC2 (Fig. 2D). Cluster pC4 was strongly enriched (50% vs 0.001% in other pCs, p < .01) for plasmids carrying all 5 virulence biomarkers (Fig. 2D), including the canonical, non-conjugative, pLVPK plasmid associated with phenotypic hypervirulence. Unlike plasmids found in pC4, pC2 plasmids rarely carried the complete set of virulence biomarkers (1%) and were more frequently associated with the carriage of only iuc and rmpA2 (23%). In agreement with this population-level analysis, 68% of virulence gene-carrying plasmids characterized in this study belonged pC-4 and pC-2 (Fig. 3, Table S3, Fig. S4). Plasmids pC-4, characterized by an IncFIB(K) or dual IncFIB(K)/IncHI1B replicon, were almost exclusively found in hvKp lineages (e.g. ST-23-KL1, ST-592, ST-420 and ST-86) irrespective of geographical origins (Fig. 3, Table S5). A single exception was the detection of pC-4 plasmids in isolates from MDR-cKp ST-23-KL57 from Georgia and Ukraine, which carried an incomplete set of virulence biomarkers (Fig. 3, Table S3).

Finally, only 6 (3.5%) plasmids from this study were hybrid and carried both virulence and AMR (ESBL or carbapenemase) determinants. Three belonged to pC-2 followed by pC-33 (n = 2) and pC-4 (n = 1) (Fig. 3, Table S3) and were mainly associated with MDR-cKp lineages (n = 2 ST-395 isolates from Georgia, and one each from ST-14, −37, and − 218 from Jordan, Thailand, and Kenya, respectively). The only exception was the pC-4 plasmid with a complete set of virulence biomarkers and an acquired CTX-M ESBL in the hvKp ST-23-KL1 lineage. Nevertheless, the pC2 cluster was the main source of overlap for virulence and ESBL/carbapenemase resistance genes (Fig. 2CD).

Convergence is largely driven by the acquisition of conjugative pC2 IncFIB(Mar)/IncHI1B plasmids globally.

Besides diversity, to quantify whether convergence resulted from preferred evolutionary routes, the complete set of 1,468 MDR-Kp isolates was analyzed to identify those that originated from shared or independent convergence events. Using maximum parsimony for the presence/absence of virulence and resistance alleles, and the tree topology, 25 independent convergence events were inferred (Fig. S5). These, hereby labeled convergence events A-Y, largely correlated with STs (i.e. one independent event per lineage) except within MDR-cKp lineages ST-395 and ST-23-KL57 for which convergent isolates were predicted to have emerged from distinct (n = 3 and n = 2, respectively) evolutionary occurrences (Fig. 3, Table S6).

At the gene level, carbapenemase and ESBL genes, were equally represented: 12/25 convergent events (48%) acquired carbapenemase genes (NDM-1, NDM-5, or OXA-48) and were non-susceptible to all carbapenems tested while isolates representing the remaining 13 events (52%) all carried an ESBL (CTX-M-15, CTX-M-55, CTX-M-63, SHV-2A) and were non-susceptible to all 3rd and 4th generation cephalosporins. By contrast, the acquisition of an incomplete set of virulence biomarkers was strongly favored (20/25 routes or 80%, p = 0.00012) and only 5/25 convergence events involved acquiring all 5 biomarkers, most predictive of the hvKp phenotype (15) (Table S6).

At the lineage level, am enrichment of convergence events was observed via the acquisition of virulence genes in an MDR-cKp background (events E-Y representing 21/25 routes or 84%). Combined with the plasmid data, the acquisition within a MDR-cKp lineage of a pC2 IncFIB(Mar)/IncHI1B or a pC4 IncFIB(K)/IncHI1B plasmid with an incomplete set of virulence biomarkers were the preferred routes of convergence (8/25 and 2/25, respectively, for a total of 40% of all independent convergent events) in the studied population (Fig. 3). By contrast, only 16% of the convergence events (labeled A to D: a ST-23-KL1 from the U.S., and three ST-268, ST-68, ST-420 isolates from Thailand) were due to the acquisition of an ESBL gene (either inserted within the canonical pC-4 virulence plasmid [Fig. S6] or harbored by a standalone plasmid) in isolates from recognized hvKp lineages (Fig. 3 and Table S6).

Recurring acquisition of genes that encode a functional aerobactin is the common denominator to global and regional emergence of convergent K. pneumoniae.

Despite convergence largely (84%) resulting from the acquisition of an incomplete set of virulence biomarkers, it was noted that the aerobactin iuc siderophore loci was acquired in all 25 independent events detected globally (Fig. 3, Table S6). It was most often (11/25) acquired by the gain of a conjugative pC2 IncFIB(Mar)/IncHI1B plasmid but a large diversity of plasmids from other pCs or singletons carried iuc and accounted for the remaining 14 convergence events.

This recurring acquisition of iuc-carrying plasmids, albeit observed across MDR-cKp lineages, is best exemplified when reconstructing the evolution of ST-395. Within our sampling of this lineage, 3 independent convergent events (H, I and P) were inferred (Fig. S5) and Bayesian phylogenetics dated their origin within the last decade (Fig. 4). The proposed routes of convergence were reconstructed: routes P and H shared a most recent common ancestor predicted to carry a pC-38 (IncR) plasmid harboring a CTX-M-15 ESBL. From there, the routes diverged and route P involved a stepwise acquisition (between 2016 and 2021) of a 177 kb singleton plasmid carrying the iuc3 aerobactin locus (node 1), a pC2 (IncFIB(Mar)/IncHI1B) carrying the NDM-5 carbapenemase (node 2), and a transposon harboring the 16S methyltransferase rmtB which inserted into the pC-38 backbone (node 3) (Fig. 4). Unlike route P, the stepwise acquisition through route H could not be resolved but this convergent event involved the acquisition of a pC-10 (IncL/M) plasmid co-harboring an OXA-48 carbapenemase and the armA 16S methyltransferase, and a pC-2 (IncFIB(Mar)/IncHI1B) hybrid plasmid carrying 4 virulence biomarkers, including iuc1, and the NDM-1 carbapenemase (node 4). Finally, route I involved the independent acquisition of 3 plasmids (node 5): a pC-2 (IncFIB(Mar)/IncHI1B) only carrying two virulence biomarker (iuc1 and rmpA2), a pC-42 (IncFIB(K)/IncFII(K)) harboring a CTX-M-15 ESBL, and a pC-10 (IncL/M) plasmid with OXA-48 and armA like that observed in route H (Fig. 4).

Phenotypically, representative isolates from each of the ST-395 convergent routes (H, I and P) resulted in extensively drug-resistant isolates with non-susceptibility to all cephalosporins, all carbapenems, and all clinically relevant aminoglycosides (Table S6). Notably, all showed increased siderophore production compared to control isolate cKp1 lacking iuc (p < 0.001) (Fig. 5A), but all did not exhibit a hypervirulent phenotype and were categorized as cKp (as defined by LD_50_ >1×10^7^ in the outbred CD1 SQ challenge model [15]), unlike the control hypervirulent isolate kvKp2 (Fig. 5B).

Increased siderophore production, not hypervirulence, is the key adaptive trait driving global convergence in K. pneumoniae.

Phenotypic characterization of representative isolates from all 25 independent convergence events showed that, irrespective of lineage, country of origin, or plasmid background, all exhibited increased siderophore production relative to cKp1 control (Fig. 5A). The only exception was route Y, a ST-25 isolate from Georgia harboring iuc3 on a pC23 plasmid. Across convergence routes, isolates that acquired an incomplete set of virulence genes on a pC2 plasmid (routes G-N) produced siderophore levels comparable to isolates harboring the canonical pC4 virulence plasmid (routes A-F; p > 0.05) (Fig. 5AC), a pattern associated with the iuc1 aerobactin allele (Table S6). By contrast, isolates that acquired other virulence plasmid types (routes O-Y), most frequently encoding iuc3, displayed significantly lower, though still elevated, siderophore production (p < 0.001) (Fig. 5AC).

Despite this consistent enhancement of iron-scavenging capacity, all convergent MDR-cKp isolates carrying an incomplete set of virulence biomarkers, whether on pC2 or other plasmids (routes G to Y), retained high LD_50_ values and uniformly lacked the hypervirulent phenotype in the murine subcutaneous infection model (Fig. 5BD). In contrast, only hvKp convergent isolates that acquired ESBL or carbapenemase genes, and that maintained the full complement of five virulence biomarkers on a pC4 plasmid (routes A-D), exhibited low LD_50_ values (LD_50_ ≤1×10^7^) consistent with true hypervirulence. Although rare, only these events result in bona fide MDR-hvKp (Fig. 5BD).

Discussion

The increasing detection of K. pneumoniae isolates that harbor AMR and virulence determinants has raised urgent concerns for human health (23). Here, we combined evolutionary, functional genomic, and biologic analyses of a global collection of K. pneumoniae to show that plasmid-mediated convergence is frequent but follows preferred evolutionary paths dominated by MDR-cKp lineages acquiring IncFIB(Mar)/IncHI1B (pC2) conjugative plasmids carrying an incomplete set of virulence biomarkers, rather than by hvkp lineages acquiring resistance plasmids. Further, these convergent isolates showed increased siderophore production, due to increased aerobactin production (32), but lacked in vivo hypervirulence emphasizing a fundamental distinction between genomic and phenotypic convergence.

At the lineage level, 25 independent convergent events distributed across 22 sequence types were identified, highlighting the recurring and polyphyletic nature of convergence. The majority resulted from MDR-cKp lineages acquiring virulence genes rather than hvKp acquiring AMR genes. This asymmetry mirrors observations from other studies (29) and may be explained in that hvKp lineages rarely acquire resistance plasmids due to low conjugation efficiency because of their unique capsule and hypermucoid properties (33) and perhaps restriction systems (34). Without these physiological constraints, MDR-cKp are a more permissive recipient, as illustrated by the three genomic convergence events inferred within a single lineage, ST-395, which independently evolved through the last decade alone. These spawned the emergence of NDM and/or OXA-48 producing epidemic subclones which have seemingly outpaced the KPC-producing ST-258 and are now predominant in Georgia and other countries in Europe (35, 36).

At the plasmid level, our data indicates that convergence is facilitated by a subset of broad-host-range plasmids combining conjugation machinery with partial virulence cargo. Specifically, IncFIB(Mar)/IncHI1B plasmids (plasmid cluster pC2) represented the most common vector for virulence biomarker gene acquisition, albeit an incomplete set, accounting for nearly half of all independent convergence events. Consistent with their non-conjugative nature, the alternative IncFIB(K)/IncHI1B plasmids (pC4), which includes the canonical pLVPK and pK2044 with a complete set of 5 virulence biomarkers, remained primarily confined to hvKp backgrounds (5, 20). Further supporting this divide, plasmidome population analysis revealed that resistance and virulence plasmids occupy distinct evolutionary spaces, with limited recombinational events. As such, hybrid plasmids carrying both AMR and virulence loci were rare and largely confined to pC2, confirming previous observations that co-resident plasmids remain the predominant molecular mechanism for genomic convergence (29, 37).

At the gene level, the aerobactin synthesis locus emerged as the single consistent signature of genomic convergence. All 25 independent events involved the acquisition of iuc and phenotypic assays demonstrated all exhibited increased siderophore production, albeit more pronounced in iuc1-carrying isolates. The iuc1 variant, typically found on pLVPK-like IncFIB(K)/IncHI1B pC4 plasmids from hvKp strains, was most common, followed by iuc3 more frequently associated with IncFIB(Mar)/IncFII pC2 plasmids, which have been previously associated with strains from animal and environmental sources (38). The consistent selection of iuc underscores the ecological advantage conferred by enhanced siderophore production (32). Recent studies have shown that aerobactin provides an ecological advantage and enhances mucosal fitness and epithelial adherence independent of its ability to enhance systemic virulence (39). Further, increased iron acquisition capacity may improve host colonization as well as persistence in iron-limited environments such as infected tissues and hospital surfaces. These same properties could indirectly promote transmission between patients, suggesting that siderophore-mediated iron acquisition may represent a key adaptive trait underlying the epidemic success of genomic convergent lineages like ST-147 and ST-395 in hospitals worldwide.

At the phenotypic level, assays in the murine subcutaneous model reinforce this separation between genomic convergence and pathogenic potential. Adding to our previous studies (12, 27) and that of Kochan and colleagues (29), none of the convergent MDR lineages with an acquired pC2 plasmid harboring an incomplete set of virulence biomarkers displayed systemic lethality characteristic of hvKp, but instead possessed a cKp virulence phenotype. While some confusion originally persisted, it is clear now that genomic convergence resulting in the acquisition of some, but not all, of the virulence biomarkers does not equate to a hypervirulent phenotype but rather defines a new, stable ecological state within the nosocomial setting. Conversely, confirmed hvKp lineages that acquired ESBL genes (e.g., ST-23-KL1 with CTX-M-15) retained a pC4 plasmid and full hypervirulence, indicating that convergence in an hvKp background can yield clinically high-risk isolates, albeit less frequently. Although imperfect, for diagnostic purposes, such hypervirulent, drug-resistant isolates are currently best identified by their carriage of a complete set of 5 virulence biomarkers (15). Unlike the epidemic MDR-cKp convergent lineages, the MDR-hvKp have not spread globally and, so far, have only been sporadically detected in Asia and Europe (37, 40), a possible consequence of the metabolic burden of full hypervirulence (41).

Collectively, our results establish a framework for understanding convergence in K. pneumoniae. Genomic convergence is frequent, structured, and dominated by MDR-cKp backgrounds acquiring iuc-carrying plasmids, but it rarely leads to the hypervirulent phenotype. These convergent strains represent an intermediate evolutionary state between cKp and hvKp, optimized for transmission rather than invasion. Understanding the molecular basis and ecological advantages of this intermediate state will be essential for anticipating the future evolution of hospital-adapted Klebsiella and refining surveillance strategies to distinguish true hypervirulence from adaptive convergence.

Materials and Methods

Bacterial Isolates

Klebsiella pneumoniae isolates were collected from Military Health System facilities across the United States and globally in collaboration with the US Department of Defense’s (DoD) Global Emerging Infections Surveillance branch. The 1,468 global isolates were recovered from Thailand (n = 362), Georgia (n = 359), Peru (n = 305), Jordan (n = 215), Kenya (n = 159), Uganda (n = 36), Philippines (n = 25), and Ukraine (n = 7) and were compared to isolates from the United States (n = 39) (Table S1).

Whole Genome Sequencing

Genomic DNA extraction and Illumina short read WGS were performed as previously described (42). Libraries were constructed using Kapa HyperPlus Library Preparation kit (Roche Diagnostics) and quantified using the KAPA Library Quantification Kit - Illumina/Bio-Rad iCycler^™^ (Roche Diagnostics) on a CFX96 real-time cycler (Bio-Rad). Libraries were normalized to 2 nM, pooled, denatured, and diluted to 1 nM. Whole genome sequencing was performed using a MiSeq or NextSeq-500 (Illumina) with MiSeq Reagent Kit v3 (600 cycles; 2 X 300 bp) or NextSeq-500 Reagent kit 500/550 vs (300 cycles, 2 × 150 bp) (Illumina). De novo draft genome assemblies were produced using Newbler v2.7 (Roche Diagnostics). Minimum thresholds for contig size and coverage were set at 200 bp and 49.5+, respectively.

Long read sequencing was performed on 53 isolates. 8 isolates were sequenced using single-molecule real-time (SMRT) by Pacific Biosciences RS II Instrument (Pacific Biosciences). Assembly was completed using the Hierarchical Genome Assembly Process (HGAP v3.0). The resulting contigs were imported into Geneious and circularized. Assemblies were polished using Illumina paired-end short reads. 45 isolates were sequenced on a Minion platform using MinION Mk1B (Oxford Nanopore Technologies). Library preparation on genomic DNA was principally performed using EXP-NBD114 and sequenced on a R9.4.1 flowcell. Basecalling was performed using Guppy v6.1.7 using the super accurate basecalling model (r9.4.1_450bps_sup). Prior to assembly, reads were filtered by Filtlong (https://github.com/rrwick/Filtlong) to capture the top 95% of reads by quality score. De novo assembly was performed using Trycycler [v0.5.3] (43). Trycycler outputs a consensus assembly from multiple long-read dedicated assemblers including Flye (v2.9.1), Miniasm (v0.3), and Raven (v1.8.1). Next, the trycylcer consensus assembly was polished with Illumina short reads using Medaka and Illumina short reads using Polypolish (44).

Bioinformatic analysis.

Species identification, MLST typing, virulence locus, capsule (K), and lipopolysaccharide (O) loci were identified using Kleborate v3.0.5 (Table S1) (17). peg-344 was identified using BLASTn search of draft genome assemblies (query sequence pLVPK, accession number AP006726.1). AMRFinderPlus v3.9.8 (45) and ARIBA v2.14.4 (46) were used to identify resistance alleles from draft assemblies and processed reads, respectively, followed by deduplication of redundant alleles calls. clast allele assignment and minimum spanning tree generation were performed with SeqSphere (47). Plasmid replicons were identified using PlasmidFinder v2.1.6 (48).

We created a core gene phylogeny for the 261 isolates harboring ≥ 1 hvKp gene (Table S1). Assemblies were annotated using Prokka v1.14.6 (49) and were used as input for Roary v3.13.0 (50) and a SNP-based alignment was generated. Recombination was filtered from the alignments using Gubbins v2.4.1 (51) and a maximum-likelihood tree was generated with RAxML-NG [v1.1] (52) using the GTR + G (50 parsimony, 50 random) model 100 random bootstrap replicates. The tree was imported in iTOL [v6.8.1] (53) for visualization with metadata.

Bayesian Evolutionary Phylogenetic Analysis

To evaluate the strength of the temporal signal, TempEst v1.5.3 was utilized to visualize the relationship between root-to-tip genetic distances for samples with known collection dates (54). The bayesian phylogenetic inference was performed using BEAST2 v2.6.5 on a recombination free alignment, removing samples with uncertain collection dates, and accounting for constant sites with beast2_constsites (https://github.com/andersgs/beast2_constsites). The HKY substitution model was selected based on evaluation of all possible substitution models in bModelTest v1.2.1 (55). The random clock model was selected based on support by the marginal likelihood value using the Nested Sampling package v1.1.0 (56). BEAST2 was run under a coalescent constant population model, with a Markov chain Monte Carlo length of 1 × 10^8^ sampling every 5 × 10^3^ steps. Analyses were repeated 5 times to confirm consistency between the obtained posterior distributions. Parameter estimates were computed using Tracer v1.7.1. Posterior trees were combined with LogCombiner and summarized in TreeAnnotator after a 50% burn-in. The final MCC target tree was visualized in FigTree v1.4.4 (https://github.com/rambaut/figtree) and annotated using iTOL (53).

Plasmidome Analysis

mge-cluster plasmid typing [v1.1.0] (31) was used to investigate plasmid families present in our dataset compared to a global public collection. Mge-cluster is a unitig based classification and pairwise Jaccard distances from unitig presence/absence matrices. All default parameters were used except perplexity, which was set to 100 and unitig filtering set to 30% to deduplicate redundant sequences.

First, 7,347 complete plasmid sequences from 2,153 isolates were retrieved on NCBI. Clustering analysis resulted in 43 distinct plasmid clusters that grouped 5,810 plasmid sequences (grey dots, Figure S3). 1,500 NCBI plasmid sequences were unassigned a cluster group and 399 plasmid sequences were excluded from clustering analysis by mge-cluster (Fig. S3). Next, we incorporated our 173 plasmid sequences (recovered from the 53 isolates) into the plasmid cluster network from which 124 plasmids grouped with 28 of the NCBI clusters (44 plasmids were unassigned a cluster group (singletons) and 5 plasmids were excluded from the clustering analysis) (Fig. 3AB). The distribution of replicon types, STs, virulence genes, ESBL and/or carbapenemase genes were investigated and visualized for all plasmid sequences located in the 28 discrete clusters (n = 4,052) and the 44 singletons from this study (Fig. 2, Table S5).

Quantitative Siderophore Assay

Strains were grown overnight at 37°C in iron-chelated M9 minimal medium containing casamino acids (c-M9-CA) (18) and culture supernatants were assessed using the chromeazurol S dye assay as described (32). Standards with concentrations of 0, 1.5, 3.1, 6.25, 12.5, 25, 50, and 100 μg/ml enabled quantitation. A minimum of 3 biological assays with 3 technical repeats were performed and the results were reported as the mean ± the SD.

Mouse Subcutaneous (SQ) Infection Model

Animal studies were reviewed and approved by the Veterans Administration Institutional Animal Care Committee and the University at Buffalo-SUNY and were carried out in strict accordance with the recommendations in the guidelines delineated in the “NIH Guide for the Care and Use of Laboratory Animals”(revised 1985) and the “Ethics of Animal Experimentation Statement” (Canadian Council on Animal Care, July, 1980) as monitored by the Institutional Animal Care and Use Committee. All efforts were made to minimize suffering. Veterinary care for the animals was supplied by the staff of Veterans Administration Animal Facility under the direction of a fully licensed veterinarian. CD1 male mice, 4–6 weeks old, were obtained from Charles River Laboratories, quarantined for 7 days before use, and then challenged via a SQ injection with the isolates of interest (100 μL of bacterial suspension serially diluted to the required titers in 1 × PBS diluted and injected using a 0.5 mL insulin syringe), as previously described (57). The animals were closely monitored for 14 days after challenge for the development of the study endpoints, survival, or severe illness (in extremis state)/death, which was recorded as a dichotomous variable. Signs that were monitored and which resulted in immediate euthanasia using methods consistent with the recommendations of the American Veterinary Medical Association Guidelines included hunched posture, ruffled fur, labored breathing, reluctance to move, photophobia, and dehydration.

Statistical Analyses

Desired comparisons between strains for experiments assessing siderophore production were made via ordinary one-way ANOVA, using Šidák’s multiple comparisons test (Prism 10.4.2 for MacIntosh, GraphPad Software Inc.).

LD_50_ values were estimated using a logistic regression model as described (15). Pair-wise comparisons of the dose-response curves were used to generate LD_50_ values. Desired comparisons between LD_50_ values were made by employing a blend of the empirical logit function along with least-squares regression incorporating strain and inoculum factors (CFU/mL) to derive p-values for comparing dose-response curves based on LS-means.

Supplementary Files

This is a list of supplementary files associated with this preprint. Click to download.

• TableS3plasmidcontigs.xlsx
• TableS528clustersn4052.xlsx
• TableS4AllNCBIContigsn7347.xlsx
• TableS6ConvergenteventsandPhenotypes.xlsx
• TableS2cgMLSTdistance.xlsx
• TableS1allisolates.xlsx
• SupplementalFiguresandTables.docx