Comparative genomics of carbapenem resistance and virulence associated with respiratory tract and bloodstream infections in carbapenem-resistant Klebsiella pneumoniae

ABSTRACT

The convergence of resistance and hypervirulence in Klebsiella pneumoniae poses a severe global health threat, yet the genomic mechanisms underlying adaptation to different clinical source remain incompletely understood. We conducted longitudinal genomic surveillance of 282 carbapenem-resistant K. pneumoniae (CRKP) isolates (225 sputum, 57 blood) between 2021and 2024 to characterize population structure, source-associated genomic adaptations, and plasmid evolution. ST11-CG258 dominated the cohort (77.3%), with 61.3% of isolates exhibiting the CR-hvKP phenotype. Longitudinal analysis revealed significant capsular plasticity within ST11, characterized by a temporal shift from KL64/47 to KL25 and the emergence of the highly invasive ST16-KL51 lineage, whereas non-ST11 clones displayed high capsular fidelity. Although bla_KPC-2 predominated, bla_OXA-181 was enriched among bloodstream isolates, whereas three bla_NDM variants were more frequently detected in respiratory isolates. Comparative genomics identified enrichment of genes assigned to COG categories associated with energy production and amino acid transport in bloodstream isolates, whereas sputum isolates showed enrichment of genes involved in DNA repair and insertion sequences. Plasmid clustering of carbapenem-resistant plasmid revealed six antimicrobial clusters (Aclusters 1–6), identifying hybrid plasmids within Acluster 1 co-carrying bla_KPC-2 and essential virulence determinants (iucABCD-iutA, rmpA, peg-344). Hypervirulent plasmids formed two major clusters, with Vcluster 2 diverging from Vcluster 1 via a 40kb deletion encompassing the ter (tellurium resistance) operon and metabolism-associated genes. These findings provide a comprehensive genomic characterization of CRKP from respiratory tract and bloodstream infections and support genome-based surveillance integrating clonal lineages, capsular types, and plasmid architectures to strengthen infection control and facilitate early identification of emerging high-risk CRKP lineages.

Introduction

Carbapenem-resistant Klebsiella pneumoniae (CRKP) continues to represent a critical threat to global health due to its rapid dissemination and the increasing prevalence of multidrug resistance [1,2]. Of particular concern is the emergence of carbapenem-resistant hypervirulent K. pneumoniae (CR-hvKP), which combines classical hypervirulence traits, such as rmpA/rmpA2 and the siderophore aerobactin, with carbapenem resistance [3–5]. Compared with classical CRKP, CR-hvKP is associated with higher risks of invasive infection, metastatic dissemination, and mortality [6,7]. Recent systematic reviews reported pooled mortality rates of approximately 57% for CR-hvKP infections, with bloodstream infection representing one of the highest-risk clinical presentations [8]. Unlike classical CRKP, which primarily manifests as nosocomial infections, CR-hvKP exhibits broad tissue tropism, with the ability to colonize the respiratory tract or invade the bloodstream [9,10].

The clinical manifestations of K. pneumoniae are highly heterogeneous, as the respiratory tract and circulatory system represent two primary, yet distinct, ecological niches during human infection [11,12]. The respiratory tract often acts as the initial site of colonization or localized infection, heavily subjected to the selective pressures of antibiotic interventions and the lung microbiome [13]. Conversely, successful translocation into the bloodstream represents a critical bottleneck, requiring the pathogen to evade host systemic immunity and adapt to a stringent hematogenous environment [14]. While epidemiological data frequently link pneumonia to subsequent bacteremia, the specific genomic determinants, adaptive mutations, and virulence repertoires that associated with from sputum to blood remain poorly understood.

The phenotypic flexibility and source-associated adaptability of K. pneumoniae are largely driven by variations in its capsular polysaccharide and the acquisition of mobile genetic elements (MGEs) [15]. The capsule, classified into diverse capsule locus (KL) types, is fundamental to immune evasion [16,17]. Recent surveillance indicates a dynamic shift in the epidemiology of these capsular types, underscoring the need to track temporal KL trends to identify emerging high-risk lineages [18]. Furthermore, the convergence of resistance and virulence is intrinsically linked to plasmid biology. While the presence of specific resistance (e.g. bla_KPC, bla_NDM) and virulence genes is well-documented, a purely gene-centric perspective obscures the structural complexity and evolutionary dynamics of the plasmids carrying them [19,20].

Although plasmid incompatibility (Inc) typing and replicon-based classification are widely used to characterize plasmids, these approaches classify plasmids primarily according to their replication systems and often fail to resolve structural variation, recombination, and mosaic plasmid architectures that are common among K. pneumoniae plasmids. Genome-wide plasmid clustering based on overall sequence similarity enables more comprehensive characterization of plasmid diversity and evolutionary relationships, particularly for hybrid resistance-virulence plasmids generated through extensive horizontal gene transfer and genetic rearrangements. Implementing plasmid-cluster analysis provides a high-resolution framework to categorize these plasmids based on their conserved backbones and modular architectures, offering deeper insights into how specific plasmid clusters mobilize and disseminate across different bacterial populations.

To address these knowledge gaps, this study presents a comprehensive genomic analysis of CRKP isolates recovered from sputum and blood samples between 2021 and 2024. By integrating comparative genomics with high-resolution plasmid profiling, we aimed to: (i) elucidate the shifting population structure and temporal trends of sequence types (STs) and KL types, (ii) define the core genomic differences and distinct genetic signatures differentiating respiratory and systemic isolates, and (iii) utilize plasmid cluster analysis to track the prevalence, distribution, and structural evolution of carbapenem-resistant and virulence-encoding plasmids. Ultimately, this study seeks to provide an updated evolutionary blueprint of K. pneumoniae adaptation and plasmid-mediated convergence in the contemporary clinical setting.

Methods and materials

Bacterial isolate collection and phenotypic characterization

A retrospective, longitudinal collection of Klebsiella pneumoniae isolates was conducted at Beijing Chaoyang Hospital, a 2500-bed tertiary medical center, between January 2021 and December 2024. Isolates were consecutively recovered from clinical sputum and blood specimens. These isolates were recovered using China Blue agar plates at 37°C for 24 h, and a single colony was identified by using MALDI-TOF MS as previously described [21,22]. China Blue agar was used as the routine selective differential medium in our clinical microbiology laboratory for the isolation of Gram-negative Enterobacterales. To minimize duplicate sampling, only the first CRKP isolate from each patient for each specimen type during the study period was included. Repeated isolates recovered from the same patient and the same specimen type at different time points were excluded. When CRKP was isolated from both sputum and blood specimens from the same patient, both isolates were retained because they represented distinct clinical specimens. In total, 6 patients contributed paired sputum and bloodstream isolates (12 isolates), which were included in specimen-specific comparative genomic analyses. Initial species identification and routine antimicrobial susceptibility testing (AST) were performed with the Vitek 2 GN09 automated system (bioMérieux, France). Minimum inhibitory concentrations (MICs) results for carbapenems (meropenem, imipenem) were interpreted according to the contemporary Clinical and Laboratory Standards Institute (CLSI) guidelines [23].

Short and long-read whole-genome sequencing

Genomic DNA was extracted using the DNeasy Blood & Tissue Kit (Qiagen, Germany). Short-read sequencing was performed on the Illumina NovaSeq 6000 platform to generate 150-bp paired-end reads. The quality of raw reads was assessed using FastQC v.0.11.5 (https://github.com/s-andrews/fastqc), and trimmed using Trimmomatic v0.39 with default settings [24]. The genome assembly was performed using SPAdes (version 3.10.0) [25] with clean reads. To achieve high-resolution plasmid structure analysis, a hybrid sequencing strategy was employed. Long-read sequencing was conducted using the PacBio platform. Raw sequencing data were processed using the Circular Consensus Sequencing (CCS) workflow to generate high-fidelity (HiFi) reads, with a minimum predicted accuracy of Q20 (≥99% accuracy). Adapter sequences and potential contaminants were identified and removed using HiFiAdapterFilt (version 3.0.0) [26]. The quality and length distribution of the resulting HiFi reads were evaluated using NanoPlot (version 1.46.2) [27] and MultiQC (version 1.34) (https://github.com/MultiQC/MultiQC). To achieve a highly accurate and contiguous genome assembly, hybrid assemblies were generated using long-read PacBio HiFi data and refined with Illumina short reads to improve base accuracy. The de novo genome assembly was primarily constructed using the long-read data with hifiasm (version 0.25.0) [28]. Subsequently, the assembly was refined by integrating short-read data to further correct potential base-call errors. Assembly completeness was assessed using BUSCO (version v6.0.0) [29].

Genomic typing, resistome, and virulence annotation

High-throughput genomic characterization was performed using Kleborate (version 2.2.0), an essential tool for profiling K. pneumoniae assemblies [30]. Kleborate was utilized to determine species, multi-locus sequence types (MLST), predict KL, clonal group (CG), acquired carbapenem and extended-spectrum beta-lactamase (ESBL) genes, and assign established virulence and resistance scores. K-locus predictions were independently verified using Kaptive (https://github.com/klebgenomics/Kaptive), yielding concordant assignments. Using a stringent genotypic threshold, CR-hvKP was defined as isolates possessing a virulence score ≥ 3 and a resistance score ≥ 1 [30]. The comprehensive resistome, virulome, plasmid replicons were further annotated using the Comprehensive Antibiotic Resistance Database (CARD), virulence factor database (VFDB), and PlasmidFinder database, respectively [31–33]. The peg-344 gene was screened using a BLASTn search of the draft genome assemblies with a minimum identity of 70%, with reference sequence from GenBank (accession no. NC_005249) [34].

Pangenome analysis

To elucidate the core genomic differences and evolutionary divergence between sputum-derived and blood-derived isolates, pangenome analysis was conducted using Panaroo (version 1.3.0) with strict parameters to account for assembly errors and fragmented genomes [35]. Genomic sequences were annotated by Bakta (version 1.9.3) (https://bakta.computational.bio/) and used to input for Panaroo. The genes that were significantly enriched in the sputum-derived and blood-derived isolates were then identified using Scoary (version 1.6.16) [36]. Because the pangenome-wide association analysis involved testing thousands of accessory genes simultaneously, p values were adjusted using the Benjamini–Hochberg false discovery rate (FDR) procedure implemented in Scoary. Only genes with an FDR-adjusted p value (Benjamini–Hochberg adjusted P, BH-P) <0.05 were considered significantly associated with specimen source and were included in subsequent functional enrichment analyses. The functional enrichment analysis of enriched genes was predicted using STRING (version 12.0) (https://string-db.org/).

SNP calling and phylogenetic reconstruction

Single nucleotide polymorphisms (SNPs) were identified using Snippy (version 4.6.0) with the first identified isolates as reference [37], and a core-genome alignment was extracted. Recombinant regions were detected and masked using Gubbins (version 3.0.0) to prevent phylogenetic distortion [38]. A maximum-likelihood (ML) phylogenetic tree was subsequently inferred from the recombination-free SNP alignment using IQ-TREE 2, applying the best-fit substitution model determined by ModelFinder, with branch support assessed by 1,000 ultrafast bootstrap replicates [39]. SNP sites were extracted from the multi-FASTA alignment file using SNP-sites (version 2.5.1) [40]. SNP distance matrix between isolates was generated using SNP-dists (version 0.8) (https://github.com/tseemann/snp-dists). Clonal clusters were identified as groups of isolates differing by fewer than 21 core-genome SNPs (cgSNPs), according to previous report [41], while singletons referred to isolates not grouping into any clusters. A minimum spanning tree based on the SNP distance matrix was constructed and visualized in GrapeTree (version 1.5.0) [42].

Plasmid reconstruction and high-resolution plasmid clustering analysis

To investigate the dynamics of carbapenem-resistant and hypervirulent plasmids, we performed an in-depth structural analysis of the plasmidome. Plasmids were reconstructed and initially typed utilizing MOB-suite (version 3.0.3), which assigned them to specific incompatibility (Inc) groups and predicted their mobility [43]. Because resistance and virulence plasmids frequently possess mosaic architectures and multiple replicons, plasmids were further grouped using MGE-cluster (version 2023) based on whole-plasmid sequence similarity (Mash distance ≤ 0.05), enabling comparison of plasmid structures beyond conventional replicon typing [44]. Hypervirulence plasmids were defined as plasmids with a virulence score of ≥3, determined as follows: a score of 3 indicated the presence of aerobactin without yersiniabactin or colibactin, a score of 4 indicated the presence of aerobactin with yersiniabactin (but not colibactin), and a score of 5 indicated the presence of yersiniabactin, colibactin, and aerobactin [30]. Carbapenem-resistant plasmids were defined as plasmids carrying at least one carbapenem-encoding gene, yielding a resistance score of ≥1 [30]. Plasmid genomes containing known replicons were clustered based on shared k-mer content and backbone synteny using the mge-cluster pipeline with a higher perplexity value (100). This allowed for the categorization of plasmids into distinct evolutionary clusters. The plasmids from this study were searched in plasmid database (PLSDB) to identify their closest relatives (https://ccb-microbe.cs.uni-saarland.de/plsdb2025/). Sequence comparisons and structural visualizations of major plasmid clusters were conducted using the BLAST Ring Image Generator (BRIG) (version 1.1.1) [45].

Mucoviscosity assay

The quantitative analysis of K. pneumoniae mucoviscosity usually involved measuring its resistance to low-speed centrifugation [46]. Bacterial strains were inoculated into 5 mL LB broth and cultured overnight at 37°C with shaking at 200 rpm. The overnight cultures in LB broth were centrifuged at 9400 g, resuspended in PBS to an OD₆₀₀ of 1, and then centrifuged for 5 minutes at 1000 g. The OD₆₀₀ was measured using a NanoDrop OneC spectrophotometer (Thermo Fisher Scientific, USA). The strain NTUH-K2044 was used as positive reference. All experiments were performed in three independent biological replicates.

Statistical analysis and visualization

All statistical analyses and data visualizations were executed in R (version 4.3.0). The phylogenetic trees, coupled with corresponding metadata (specimen source, KL type, AST profiles, and plasmid clusters), were visualized and annotated using the ggtree package [47]. Differences in the prevalence of specific genes or plasmid clusters between sputum and blood isolates were evaluated using the Chi-square test or Fisher’s exact test, as appropriate, with a two-tailed p < 0.05 considered statistically significant. Principal Coordinate Analysis (PCoA) based on accessory gene presence/absence matrices was plotted using ggplot2 to assess the genetic clustering of isolates by clinical source.

Results

Population structure and epidemic trend of CRKP isolates

In this retrospective study, a total of 282 non-duplicate CRKP isolates were systematically recovered from clinical specimens between 2021 and 2024. The collection comprised 225 (79.8%) respiratory isolates from sputum and 57 (20.2%) invasive isolates from blood, with a relatively balanced annual isolation rate across the four-year study period (Figure S1). In silico multi-locus sequence typing (MLST) and CG analysis revealed a highly structured, clonal population overwhelmingly dominated by ST11 (n = 218, 77.3%), the vast majority of which were assigned to the global high-risk CG258 (n = 219, 77.7% of total cohort) (Figure 1(A)). Other minority sequence types identified included ST15 (n = 33, 11.7%), ST16 (n = 14, 5.0%), and ST307 (n = 5, 1.8%), indicating a background of genetic diversity alongside the predominant clonal expansion of ST11. To evaluate the convergence of multi-drug resistance and hypervirulence, the genotypic resistance and virulence profiles of the cohort were assessed, with 61.3% (173/282) of the total cohort recording a virulence score of 3 or higher and 100% of isolates achieving a resistance score of ≥2 (Figure 1(A)). The CR-hvKP phenotype was almost exclusively restricted to the dominant ST11 clone (CG258), of which 168 (97.1%, 168/173) isolates belonged to ST11 isolates. The remaining five CR-hvKP isolates belonged to ST1764, ST2237, ST859, ST2237, and ST16. The PCoA analysis reveals a clear separation between K. pneumoniae strains isolated from blood and sputum samples, with the primary principal coordinate (PC1) explaining over 60% of the total variance (Figure 1(B)). This distinct clustering indicates that the ST distributions are significantly influenced by the specimen source, suggesting source-associated adaptation or different transmission dynamics between systemic and respiratory infections. The ST11-restricted PCoA continued to separate bloodstream-derived and sputum-derived isolates, indicating that the genomic differentiation remained evident within the dominant lineage (Figure S2).

Figure 1.

Population structure of 282 CRKP isolates between 2021 and 2024. (A) Maximum likelihood phylogeny of 225 CRKP isolates from sputum and 57 isolates from bloodstream infections. The phylogenic tree was built based on 92,188 cgSNP, which was called with the first identified isolate as reference. The heatmaps shows the sequence type (ST), clonal group (CG), sampling year, source, virulence score, and resistance score, respectively. (B) Principal coordinates analysis (PCoA) of principal coordinates analysis (PCoA) showing the ST distributions of CRKP isolates from blood and sputum samples. The ordination plot is based on a Bray-Curtis distance matrix derived from the relative abundance of STs isolated from blood (red) and sputum (blue) samples across four consecutive years (2021–2024). Each data point represents the total ST profile of a specific sample source for a given year, with shapes indicating the year of collection. (C) Dynamic shifts in capsular polysaccharide (K-locus) composition from 2021 to 2024. The stacked area chart illustrates the relative percentage of different K-locus types identified in all CRKP isolates. Each colored area represents a specific K-locus, with its vertical thickness corresponding to its proportion in the population for a given year. KL62: four ST15-KL62 and one ST11-KL62. KL10: right ST11-KL62 and one ST1128-KL62. KL19: 29 ST15-KL19, two ST2237-KL19, and one ST859-KL19.

Integrating ST and K-locus data revealed two distinct evolutionary strategies among the circulating clones (Figure 1(C)). The dominant ST11 lineage demonstrated remarkable capsular plasticity, diversifying into three major sublineages including ST11-KL64 (43.1%), ST11-KL47 (33.5%), and ST11-KL25 (18.8%). Conversely, non-ST11 lineages exhibited high degrees of capsular fidelity, 90.9% of ST15 isolates were strictly associated with KL19, and 92.9% of the emerging ST16 isolates were associated with KL51. The temporal dynamics of these ST-KL combinations (Figure 1(C), and Figure S3) highlight a period of rapid clonal replacement within the ST11 population. The traditional ST11-KL64 and ST11-KL47 combinations, which dominated the early years of the study, were increasingly superseded by the emergent ST11-KL25 sublineage by 2024. This transition, alongside the 2023 debut of the ST16-KL51 lineage, underscores a highly dynamic genomic landscape.

Potential epidemiological link

To delineate genomic cluster and clonal expansion patterns, we performed a 21-SNP threshold clustering analysis on 282 CRKP isolates, identifying 34 distinct genomic clusters (179 isolates) (Figure 2). These clusters were categorized into major clusters (≥5 isolates, n = 10) and minor clusters (<5 isolates, n = 24). The population exhibited a high level of clonal concentration, with 10 major clusters accounting for 62.0% of these clustered isolates. Longitudinal analysis revealed that 70.0% of these major clusters persisted across multiple years (Figure S4), confirming the long-term endemicity of high-risk lineages in this clinical setting. While sputum remained the predominant source (Figure 2(A)), major clusters demonstrated significantly higher bloodstream integration (20.0%-41.7%) compared to minor clusters (10.3%), with the newly emerged ST16 lineage (2023–2024) showing a relatively higher proportion of bloodstream isolates than other non-ST11 lineages (41.7% blood isolates). Genetically, ST11 remained the primary endemic driver, dominating 90.0% of major clusters (Figure 2(B)). Despite core genome stability (mean intra-cluster SNP distances of 8.75–18.39), these clusters exhibited substantial capsular plasticity, characterized by frequent switching between KL47 and KL64. Collectively, these results highlight a dynamic genomic landscape sustained by the long-term persistence of ST11 (CG258) and the rapid expansion of invasive, non-ST11 clusters, necessitating targeted genomic surveillance of these expanding high-risk lineages.

Figure 2.

Genomic clusters of 225 sputum-derived and 57 bloodstream-derived CRKP isolates. (A) Minimum spanning tree colored by STs illustrating the cgSNP difference among all CRKP isolates. (B) Minimum spanning tree colored by sample type illustrating the cgSNP difference among all CRKP isolates. All the minimum spanning trees were grouped based on 21 or fewer SNPs. Each circle or section of the pie chart represents a single isolate.

Comparative analysis of carbapenem resistant and virulent genes

A total of 104 AMR genes were identified, with no statistically significant difference in AMR gene counts between sputum and blood isolates (p > 0.05) (Figure S5A). The distribution of carbapenem genes in K. pneumoniae is characterized by a high genetic overlap (80%) and the universal dominance of bla_KPC-2 (>91%) across clinical source (Figure 3(A)). Despite this shared reservoir, source-specific preferences were observed. bla_OXA-181 demonstrated a distinct enrichment in bloodstream isolates (10.53% vs 3.56%), while bla_NDM-1 and the sputum-exclusive bla_NDM-13 were more prevalent in respiratory samples. Nevertheless, statistical validation (Wilcoxon signed-rank test, p = 0.625) confirms that the overall carbapenem resistance burden remains uniform across blood and sputum isolates. Twelve acquired ESBL genes were identified and compared between sputum- and blood-derived CRKP isolates. Most ESBL genes (66.7%) were detected only in sputum, while one (bla_SHV-129) was exclusive to blood isolates, only three ESBL genes were shared between blood and sputum isolates. bla_CTX-M-65 was dominant in both (about 55% positive rate). bla_SHV-12 showed a statistically significant sputum preference (48.4% vs 8.8% in blood). Seventy-five percent of genes had low prevalence (<5%), indicating a focused ESBL resistance profile dominated by bla_CTX-M-65 and bla_SHV-12, which guides targeted antimicrobial therapy (Figure S6).

Figure 3.

Genomic comparisons of CRKP isolates from sputum and blood infections. (A) Comparison of carbapenem gene prevalence between CRKP isolates from sputum and blood infections. (B) Comparison of key virulence gene prevalence between CRKP isolates from sputum and blood infections. (C) Clusters of Orthologous groups (COG) clustering of significantly different genes for CRKP isolates from sputum and blood infections. These genes were categorized into three groups. Information storage and processing (ISP) included genes related to translation, ribosomal structure and biogenesis (J), transcription (K), and replication, recombination and repair (L). Cellular processes and signaling included genes related to cell cycle control, cell division, chromosome partitioning (D), cell wall/membrane/envelope biogenesis (M), cell motility (N), posttranslational modification, protein turnover, chaperones (O), signal transduction mechanisms (T), intracellular trafficking, secretion, and vesicular transport (U), and nuclear structure (V). Metabolism included genes related to energy production and conversion (C), amino acid transport and metabolism (E), nucleotide transport and metabolism (F), carbohydrate transport and metabolism (G), coenzyme transport and metabolism (H), lipid transport and metabolism (I), inorganic ion transport and metabolism (P), and secondary metabolites biosynthesis, transport and catabolism (Q). (D) Network plot showing predicted protein-protein interactions from the database STRING. The middle genes (tufB and KPN_04724) are significantly different genes identified from this study.

There was no statistically significant difference in virulence gene counts between sputum and blood isolates (Figure S5B), and both sample types showed highly similar distribution trends, with isolates predominantly carrying 22–26 virulence genes (Figure 3(B)). Analysis of eight clinically significant virulence genes revealed a highly conserved virulence reservoir between respiratory and bloodstream isolates, with 100% detection overlap across both sources. A core virulence set of six genes (ybt, ICEkp3, peg-344, rmpA, rmpA2, and iucABC-iutA) maintained high prevalence (≥50%) in both groups, with ybt being nearly universal (>96%). While the overall prevalence categories were identical across sample types, specific hypervirulence markers exhibited distinct site preferences, with rmpA and rmpA2 enriched in blood isolates by 11.09% and 9.78%, respectively, suggesting their enhanced role in systemic dissemination and capsule regulation. Conversely, iron-acquisition systems (ybt, iucABC-iutA) showed negligible site-specific variation. These findings indicate that while CRKP maintains a baseline of high virulence regardless of the clinical source, the preferential presence of rmpA/A2 in the bloodstream serves as a key marker for invasive infection potential.

Comparative genomic analysis between sputum- and blood-derived isolates

Due to the high genomic similarity observed between sputum- and blood-derived isolates, pangenome and comparative genomic analyses were conducted to clarify both their shared and unique genetic characteristics (Figure S7). Across all isolates, 22,978 genes were identified, with 3,528 making up the core genome (present in ≥99% of isolates). We further investigated their functional annotation using Clusters of Orthologous Groups (COG), revealing distinct enrichment trends in key functional categories between the two groups (Figure 3(C)). After Benjamini-Hochberg correction for multiple comparisons, 227 signature genes were significantly enriched in sputum-derived isolates (OR >1, BH-adjusted p < 0.05), most of which were related insertion sequences (IS), followed by hypothetical proteins (Figure S8). Conversely, 79 signature genes were significantly enriched in blood-derived isolates (OR <1, BH-adjusted p < 0.05). Functional annotation and COG classification showed that sputum-derived signature genes were mainly involved in replication, recombination and repair (L), which are essential for antibiotic resistance mechanisms, and mobile genetic elements (information storage and processing) (Figure 3(C)). In contrast, blood-derived isolates exhibited significant enrichment in genes related to energy production and conversion (C), amino acid transport and metabolism (E), and cell wall/membrane/envelope biogenesis (M), which are functionally associated with bacterial metabolism, environmental adaptation (cellular processes and signaling). To assess whether these genes are functionally related, we analyzed them using the STRING database. The result revealed that the top five genes (present in ≥88% of isolates) enriched in sputum-derived isolates were present in two networks (Figure 3(D)), with one gene not similar to any known K. pneumoniae proteins, suggesting limited functional interaction among these genes.

Carbapenem resistant and hypervirulent plasmid clustering and transmission

To evaluate the horizontal gene transfer (HGT) potential and plasmid architecture within the population, we reconstructed 1,446 plasmids across the 282 CRKP isolates. The plasmid load per strain was remarkably stable, with a mean of 5.13 plasmids (range from 1 to 10), notably, 69.2% of strains maintained a conserved capacity of 4 to 6 plasmids (Figure S9A). Classification based on mobility (MOB) revealed a distribution comprising 607 non-mobilizable, 507 mobilizable, and 332 conjugative plasmids. Non-mobilizable elements were the most ubiquitous, appearing in 93.6% of isolates. Most significantly, 69.5% of strains harbored a sophisticated system containing all three mobility types simultaneously (Figure S9B). The high prevalence of mobilizable and conjugative plasmids detected in over 85% of strains underscored a widespread potential for the horizontal dissemination of resistance and virulence determinants throughout the cohort.

We next identified the plasmids carrying carbapenem-resistant and/or virulence gene. For carbapenem-resistant plasmids, IncR_1 and IncFII (pHN7A8)_1_pHN7A8 are the primary replication systems in KL types of ST11 isolates, except for ST11-KL10 isolates with IncFIB(K)_1_Kpn3 and IncHI1B_1_pNDM-MAR being most prevalent (Figure 4(A)). IncX3_1 and ColKP3_1 were prevalent in CRKP ST16 isolates, with FII (pBK30683)_1 in ST15 isolates. For virulence plasmids, IncFIB(K)_1_Kpn3 and IncHI1B_1_pNDM-MAR were highly prevalent in all isolates (Figure 4(B)).

Figure 4.

Distribution of plasmid replicons of CRKP isolates from sputum and blood infections. (A) Replicon distribution of carbapenem-resistant plasmids. (B) Replicon distribution of hypervirulent plasmids. The purple rectangular grids indicate the presence of the replicons, while the gray grids indicate the absence of the replicons. The ST, K-locus, and yersiniabactin (ybt) were grouped by colored blocks, shown at the top of the binary heatmaps. The types of plasmid replicons were labeled at the right of the binary heatmaps. The bar plot on the right summarizes the overall prevalence of each plasmid replicon across CRKP isolates.

We used MGE-cluster to cluster the carbapenem-resistant plasmids, respectively (Figure 5). 82% of the carbapenem-resistant plasmids were assigned to six clusters (Alusters 1–6) (Figure 5(A)). Acluster 6 was the most prevalent group, accounting for 41.9% of the total cohort. Although the clonal landscape was heavily dominated by Aclusters 3 and 6, Acluster 4 represented rarer, non-ST11 lineages (Figure 5(B)). And only one ST11 isolate was detected in Acluster 2. The presence of rmpA/rmpA2 served as the primary differentiator (Figure 5(C)), Aclusters 3 and 6 were significantly associated with the hypermucoid phenotype. Notably, Acluster 3 exhibited the highest invasiveness, accounting for the largest proportion of bloodstream isolates, and Acluster 4 was only detected in sputum-derived isolates (Figure S10). Acluster 5 represented a distinct ST11 sublineage that lacks rmpA/rmpA2, identifying it as a non-hypervirulent endemic reservoir despite its high-risk ST11 background. These findings indicate that plasmid-based clustering effectively maps the virulence evolution within the ST11 lineage, identifying the ST11-KL64/47-rmpA/rmpA2+ plasmid clusters (particularly Acluster 3) as the primary drivers of invasive infection in this cohort.

Figure 5.

Plasmid clustering of carbapenem-resistant and hypervirulent plasmids. (A) Clustering of plasmids carrying carbapenem genes. The carbapenem-resistant plasmids formed six antimicrobial clusters (aclusters) using the MGE-cluster analyses, which are labeled with different colors. tsne = t-distributed stochastic neighbor embedding. (B) Stacked column chart illustrating the proportion of ST in each Acluster. (C) The rmpA/rmpa2 distribution in each Acluster. (D) Clustering of hypervirulent plasmids. The hypervirulent plasmids formed two virulent clusters (Vclusters 1–2) using the MGE-cluster analyses, which are labeled with different colors. tsne = t-distributed stochastic neighbor embedding. (E) The comparison of ST distribution in each Vcluster.

In addition, the virulence plasmids were classified into two clusters (Vclusters 1 and 2), with Vcluster 2 being the most prevalent group (Figure 5(D)). Vcluster 1 exhibited a broad phylogenetic distribution with highly variable prevalence across diverse STs. In contrast, Vcluster 2 demonstrated a more concentrated pattern, often showing inverted prevalent levels relative to Vcluster1 within shared lineages. While a degree of co-occurrence was observed in a subset of STs, several lineages were characterized by the exclusive or dominant presence of a single Vcluster (Figure 5(E)). These diverging prevalence patterns suggest that Vcluster 1 and Vcluster 2 possess unique evolutionary or functional associations with specific CRKP lineages.

Structural variations of the carbapenem resistant and hypervirulent plasmids

To investigate the structural adaptability of carbapenem-resistant plasmids, their replicon types were first compared. Acluster 6 was characterized by a stable dual-gene signature (IncFII (pHN7A8)_1 and IncR_1) (Figure S11A). Conversely, Acluster 3 represented a highly homogeneous subpopulation with a simplified carriage pattern, exclusively harboring the IncR_1 replicon. High specificity was observed in four Aclusters (1, 2, 4, and 5) possessing unique markers. Notably, Acluster 1 exhibited absolute consistency (100% carriage) for IncFIB(K)_1_Kpn3 and IncHI1B_1_pNDM-MAR. Acluster 2 displayed the highest genetic diversity, characterized by unique signatures of IncX3_1 (94.12%) and ColKP3_1 (76.47%) (Figure S11B). The backbone structures of representatives of different Aclusters were aligned and compared, showing clear differences between clusters (Figure 6(A)). Notably, we found that the structures of carbapenem-resistant plasmids within Acluster 1 exhibited highly similar backbone structures based on comparative plasmid analyses, which contained bla_KPC-2 and iucABCD-iutA, rmpA, and peg-344, indicative of the emergence of carbapenem-resistant and hypervirulent hybrid plasmids (Figure 6(B)). Representative isolates from Acluster 1 were further subjected to long-read sequencing, confirming that bla_KPC-2 and the virulence loci (iucABCD-iutA, rmpA, and peg-344) co-localized on a single hybrid plasmid in these representative genomes (Figure 6(B)). Complete hybrid plasmid sequences obtained from representative isolates demonstrated that bla_KPC-2 and major virulence loci were co-localized on a single circular plasmid.

Figure 6.

Genomic comparative analyses of carbapenem-resistant and hypervirulent plasmids. (A) Circular comparison of the six antimicrobial clusters (aclusters). Each circle represents one carbapenem-resistant plasmids from corresponding Acluster. (B) Circular comparison of four carbapenem-resistant hypervirulent plasmids. The complete plasmid genome from isolate KP70468 is used as reference. All the convergent plasmids belong to Acluster1. (C) Circular comparison of the hypervirulent plasmids within different virulent clusters (vclusters). The plasmid (CP029383.2) is the closest relative of hypervirulent plasmid from isolate KP78521 (Vcluster 1), with the plasmid (CP155704.1) being the closest relative of isolate KP88688 (Vcluster 2). The closest relatives are identified based on mash screen with default parameters in the PLSDB database. (D) Organization of deletion regions in the virulence plasmids. The hypervirulent plasmid from isolate KP78521 (Vcluster 1) is compared to the plasmid from KP88688 (Vcluster 2). The plasmid from KP88688 (Vcluster 2) is compared to its closest relative (CP155704.1). Genes are depicted as arrows, and shaded areas represent regions of homology with greater than 95% nucleotide identity.

The structures of hypervirulent plasmids from Vcluster 1 and Vcluster 2 were also investigated and compared (Figure 6(C)), which all contained IncFIB(K)_1_Kpn3 and IncHI1B_1_pNDM-MAR replicons, and virulence genes, such as iucABCD-iutA, rmpA/rmpA2, and peg-344. Hypervirulent plasmids within Vcluster 1 had a high degree of conservation with their closest relative (CP029383.2) from PLSDB database (Figure 6(C)). However, the plasmids belonging to Vcluster 2 lacked a fragment of approximately 40kb compared with plasmids within Vcluster 1. The genes within the fragment encoded proteins related to tellurium resistance (terZEDCBAZXY), and metabolism-related enzymes, such monooxygenase, N-acetyltransferase, aliphatic nitrilase and 2-aminoadipate transaminase (Figure 6(D)). And, the hypervirulent plasmids within Vcluster 2 showed a slightly different structure compared with its closest relative (CP155704.1) from the PLSDB database, which lacked the iroBCD, fecI, fecA, and IS66 family transposase genes (Figure 6(D)). In addition, these clinical CRKP isolates harbouing plasmids from Vcluster 1 and Vcluster 2 both showed similar mucoviscosity compared with NTUH-K2044 (p > 0.05) (Figure S12). Notably, the clinical CRKP isolates carrying hybrid plasmids exhibited high mucoviscosity compared with isolates carrying other hypervirulent plasmids of the same Vcluster 2 (p < 0.05).

Discussion

In this retrospective genomic epidemiological study of 282 CRKP clinical isolates collected from 2021 to 2024, we observed a highly clonal population structure dominated by the high-risk ST11 lineage, consistent with recent systematic analyses highlighting the global dominance of ST11 among CRKP clones, particularly in Asia and South America [48]. The overwhelming prevalence of ST11-CG258 in our collection reinforced the notion that this clone remained a primary driver of CRKP dissemination in clinical settings and was associated with both antimicrobial resistance and convergent hypervirulent traits [48,49].

A significant finding of our longitudinal analysis was the dynamic shift in capsular types within the ST11 population. The transition from the traditionally dominant ST11-KL64 and ST11-KL47 to the emerging ST11-KL25 sublineage by 2024 may reflect capsule switching and/or expansion of preexisting ST11 sublineages carrying different capsule loci. These observations are consistent with previous reports showing that capsule switching can occur in ST11 CRKP; however, our cross-sectional genomic data cannot distinguish capsule switching from temporal replacement of preexisting ST11 sublineages carrying different K loci [48,50]. The rapid rise of KL25 highlighted a potential selective advantage in colonizing or infecting hospital patients [51], necessitating continuous high-resolution surveillance to monitor these high-risk capsular variants. Together, these findings address our first objective by defining the temporal dynamics of dominant ST/KL lineages and demonstrating that capsular replacement, together with the emergence of ST16-KL51, contributed to changes in the CRKP population during the study period.

Genomic cluster analysis revealed widespread persistence and endemicity of major clusters over multiple years. The long-term presence of these clonal groups aligned with documented endemic transmission patterns of CRKP in healthcare environments, where infection control challenges and selective pressure from antibiotic use favored clonal stability [52]. Additionally, the relatively high proportion of bloodstream isolates among ST16-KL51 suggests that this emerging lineage warrants continued surveillance. However, because only 14 ST16 isolates were included and most were collected during 2023–2024, larger multicenter studies are required to determine whether this observation reflects an intrinsic biological characteristic or a localized epidemiological phenomenon [53].

Our analysis of antimicrobial resistance determinants highlighted the predominance of bla_KPC-2 as the principal carbapenem, reinforcing global surveillance reports that identify KPC-type carbapenem as the most common resistance mechanism in ST11 and related clones [54]. The differential enrichment of bla_OXA-181 and bla_NDM variants between bloodstream and sputum isolates suggested source-associated selection pressures that warranted further functional investigation. Similarly, the focus of acquired ESBL genes, with bla_CTX-M-65 and bla_SHV-12 dominating, reflected the common ESBL spectrum observed in multidrug-resistant K. pneumoniae populations and underscored the challenge in tailoring empiric therapy in high-burden contexts. Because antimicrobial susceptibility testing was performed as part of routine clinical practice using the VITEK 2 system, quantitative MIC values for highly carbapenem-resistant isolates were frequently truncated at the upper limit of the assay. Consequently, we were unable to comprehensively evaluate whether different carbapenemase genotypes (e.g. bla_KPC-2, bla_OXA-181, or bla_NDM variants) were associated with differences in carbapenem MICs or therapeutic susceptibility profiles. Future prospective studies using reference broth microdilution methods will be required to address this question.

Our comparative genomic analysis identified distinct genetic signatures associated with respiratory and bloodstream isolates. While iron-acquisition systems like ybt were ubiquitous [55], blood-derived isolates were significantly enriched in rmpA and rmpA2. After restricting the analysis to the dominant ST11 lineage, the enrichment of rmpA and rmpA2 among bloodstream-derived isolates remained evident, although the differences reached only borderline statistical significance (Fisher’s exact test: p = 0.053 and p = 0.052, respectively). These findings suggest that lineage composition alone is unlikely to fully account for the observed trend, while additional studies with larger sample sizes will be required to confirm the independent contribution of these virulence regulators. Previous studies have shown that rmpA/rmpA2 regulate capsule production and may contribute to resistance against complement-mediated killing. Their enrichment among bloodstream isolates in our dataset is consistent with a potential role during bloodstream infection, although functional validation was beyond the scope of this study.

Conversely, sputum isolates were enriched in genes related to DNA recombination and repair (COG category L), which may reflect differences in the genomic composition of respiratory-associated isolates. In contrast, bloodstream isolates were enriched in genes involved in energy production and conversion, amino acid transport and metabolism, and cell wall/membrane biogenesis. These functional categories have previously been implicated in bacterial survival during systemic infection [56]. However, our comparative genomic analysis cannot determine whether these genes directly facilitate bloodstream adaptation or instead represent lineage-associated genomic features enriched among bloodstream isolates [56]. Accordingly, these findings should be as source-associated genomic signatures rather than direct evidence of source-specific adaptation, and their biological significance requires future functional validation [57]. Notably, the additional ST11-restricted analysis yielded highly consistent functional enrichment patterns, with IS-associated genes remaining the predominant category among sputum-associated isolates, suggesting that these genomic signatures are not solely explained by sequence type composition. Nevertheless, because a phylogeny-aware bacterial GWAS was not performed, residual effects of population structure cannot be completely excluded.

Clustering of carbapenem-resistant and hypervirulent plasmids further illustrates the evolutionary trajectories of high-risk CRKP lineages. Comparative plasmid analyses identified an Acluster 1 group with highly conserved backbone structures. Long-read sequencing of representative isolates confirmed the existence of hybrid plasmids co-carrying bla_KPC-2 and hypervirulence loci. Although these findings strongly support the presence of convergent resistance-virulence plasmids within Acluster 1, systematic physical validation across all cluster members was beyond the scope of this study. The presence of these hybrid elements represents a notable example of resistance-virulence convergence identified in this study [58]. Complete genome sequencing confirmed that these plasmids utilized IncFII and IncR replicons to maintain structural stability while minimizing the metabolic fitness cost to the host cell. Although complete plasmid assemblies confirmed the co-localization of resistance and virulence determinants on the same plasmid in representative isolates, the transferability and stability of these hybrid plasmids were not evaluated in the present study.

Intriguingly, we observed a significant structural divergence in hypervirulent plasmids (Vcluster 1 vs Vcluster 2). Comparative analysis revealed that Vcluster 2 lacked an approximately 40-kb region containing the ter operon and several metabolism-associated genes relative to Vcluster 1. The biological consequences of this deletion remain unclear. Previous studies have suggested that the ter operon contributes to tellurite resistance and may participate in bacterial responses to oxidative stress and host-associated environments [59]. However, whether loss of this region influences plasmid stability, bacterial fitness, virulence, or adaptation in CRKP remains unknown. Further functional studies will be required to determine the phenotypic significance of this structural variation. Finally, plasmid clustering addressed our third objective by resolving the structural diversity of resistance and virulence plasmids beyond conventional replicon typing and identifying convergent plasmid architectures within the dominant ST11 lineage.

This study has several limitations. First, although 25 patients contributed paired sputum and bloodstream isolates, these accounted for only 4.3% (12/282) of the study collection. Consequently, comparisons between respiratory- and bloodstream-derived CRKP isolates were primarily based on independent isolates recovered from different patients rather than longitudinal within-patient sampling. Therefore, the observed genomic differences should be interpreted as source-associated genomic signatures rather than direct evidence of within-host adaptation or respiratory-to-bloodstream transition. Future studies incorporating longitudinal paired sampling and statistical approaches accounting for repeated measures are warranted to validate these observations. Second, although several cgSNP-defined genomic clusters were identified, they indicate close genetic relatedness rather than confirmed transmission events because detailed epidemiological linkage data were unavailable. Third, while complete plasmid assemblies confirmed the co-localization of carbapenem resistance and virulence determinants on hybrid plasmids, their biological properties, including conjugative transferability, plasmid stability, and fitness effects, were not experimentally evaluated. Future studies combining functional assays with plasmid transmission and stability analyses will be essential to determine their dissemination potential and evolutionary dynamics.

Conclusion

From a clinical perspective, genomic surveillance should prioritize dominant ST11-KL64 clones, the emerging ST16-KL51 lineage, and hybrid resistance-virulence plasmids carrying bla_KPC-2 together with major virulence loci. Integration of genomic surveillance into routine microbiological diagnostics may facilitate early detection of high-risk CRKP clones and support targeted infection prevention and antimicrobial stewardship strategies. These findings provide an evidence base for hospital surveillance programs and early-warning systems aimed at limiting the dissemination of high-risk CRKP lineages and resistance-virulence convergence.

Funding Statement

This work was supported by the National Natural Science Foundation of China [82502766 to WZ]. The funding source had no role in study design, data collection, data analysis, data interpretation, or writing of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The genome from this study have been submitted to the NCBI database under accession numbers PRJNA1466030 and PRJNA1254431. The supplementary data supporting the findings of this study, including Figure S1, Figure S2, Figure S3, Figure S4, Figure S5, Figure S6, Figure S7, Figure S8, Figure S9, Figure S10, Figure S11, and Figure S12, are directly available on Figshare at 10.6084/m9.figshare.32515284 [60].

Ethics statement

This study was conducted in accordance with the ethical principles of the Declaration of Helsinki. The clinical bacterial strains used in this study were isolated from routine, leftover clinical diagnostic samples. No additional specimens were actively collected from patients for this research. The study was approved by the Ethics Committee of Beijing Chaoyang Hospital, Capital Medical University (Number: 2025–10-30–3). Because the study used fully anonymized, retrospective biological material and posed no risk to patients, the requirement for written or verbal informed consent was formally waived by the ethics committee.

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/21505594.2026.2716486