Global genomic landscape of the Klebsiella oxytoca complex: diversity, resistome and plasmidome (1974-2025) Authors: Wentao Zhu, Xi Chen, Hong Shen, Chunxia Yang, Qian Liu, Li Gu Published: 2026 ABSTRACT AbstractBackgroundKlebsiella oxytoca species complex (KoSC) is an opportunistic pathogen causing various infections. However, our understanding of their population structure and the co-occurrence of specific plasmid replicons with ARGs remains limited.Methods3097 high-quality genomes of KoSC between 1974 and 2025 were included. The population structure was determined by phylogenetic analysis and ANI identity, while ARGs and VAGs were predicted based on genomic data.ResultsAll genomes belonged to six species, with K. michiganensis (44.2%) and K. oxytoca (41.1%) being the most common. The genomes were sourced from 63 countries, with the majority originating from the USA (25.3%). 460 STs were identified within the KoSC, suggesting a high level of genetic diversity within the KoSC. The number of ARGs (median of 5 per genome) varied significantly among species, with K. michiganensis harboring the highest number. Forty-four different types of carbapenemase-encoding genes were detected across 37 countries, with blaNDM-1 (19.4%) and blaKPC-2 (19.3%) being the most prevalent. A significantly higher number of VAGs were found within K. oxytoca and K. grimontii isolates compared to the other four species. Notably, 22 plasmid replicons and 76 ISs showed significant correlations with ARGs, suggesting their potential role in ARG dissemination.ConclusionsThis study provides a comprehensive genomic framework for understanding the global population structure and antimicrobial resistance landscape of the KoSC. These findings identify priority targets for genomic surveillance and improve our understanding of the mobile genetic elements contributing to antimicrobial resistance dissemination within a One Health context.Keywords: Klebsiella oxytoca species complex, Population structure, antimicrobial resistance, carbapenemase, plasmids Introduction The Klebsiella oxytoca species complex (KoSC) comprises a group of closely related, facultatively anaerobic, oxidase-negative, Gram-negative bacterial species within the genus Klebsiella (family Enterobacteriaceae) [1,2]. While it is part of the normal human gut microflora, it can also act as an opportunistic pathogen, causing a variety of infections [3,4]. These infections can range from mild diarrhea to severe conditions such as bacteremia and meningitis [[5], [6], [7], [8]]. Additionally, K. oxytoca has been implicated in outbreaks of healthcare-associated infections [9,10]. It is now recognized as an important human pathogen and the second most common member of the Klebsiella group causing clinical infections [11]. The KoSC is comprised of six known species: K. oxytoca, K. grimontii, K. huaxiensis, K. michiganensis, K. pasteurii, and K. spallanzanii, as well as three unnamed species (taxons 1, 2, and 3). These species are difficult to reliably differentiate based on phenotypic and chemotaxonomic characteristics [3]. As a result, many genomes deposited in public databases may be incorrectly labeled as K. oxytoca [[12], [13], [14]], leading to the underreporting of clinically relevant species such as K. michiganensis and K. grimontii in the literature [1,13]. However, whole-genome sequencing (WGS) enables accurate species-level discrimination within the KoSC, overcoming the limitations of conventional phenotypic identification and improving taxonomic classification and genomic surveillance [[15], [16], [17]]. Antimicrobial resistance (AMR) in the KoSC is an increasingly serious threat, largely driven by the inappropriate use of antimicrobials [18]. Aside from intrinsic resistance mediated by the chromosomally encoded β-lactamase gene blaOXY [19], acquired resistant isolates, especially those exhibiting multidrug resistance (MDR), have been increasingly observed in the KoSC [2,6]. The emergence of carbapenemase resistance remains an ongoing public-health concern causing severe outbreaks and dramatically limiting treatment options [20], especially when mediated by transferable carbapenemase-encoding genes [[21], [22], [23]]. Virulence genes also play a crucial role in the outcome of clinical infection. While the emergence of carbapenem-resistant hypervirulent K. pneumoniae has drawn increasing attention, little is known about carbapenemase-resistant hypervirulent K. oxytoca. Substantial knowledge gaps remain in our understanding of antibiotic resistance and virulence within the global KoSC. Although previous studies have explored the diversity of the KoSC [3,19], they have typically included only a few hundred genomes and have not assessed the co-occurrence of plasmid replicons with ARGs. In this study, we aimed to comprehensively characterize the population structure, resistome, and plasmidome of KoSC using a large genomic dataset, and to infer potential inter-host transmission events. Materials and methods Study design This study aimed to map the population structure of the global KoSC and identify the major determinants of AMR. To ensure the comprehensiveness of the analyzed genomes, all publicly available genomes belonging to KoSC from the National Center of Biotechnology Information (NCBI) were downloaded on May 1, 2025. Additionally, isolates from the “Klebsiella oxytoca database” of the pubMLST database (https://pubmlst.org/organisms/klebsiella-oxytoca) were also included. Metadata for publicly available genomes, including collection year, isolation host, and sampling country, were compiled from the corresponding NCBI Biosamples or the isolate database (https://pubmlst.org/bigsdb). Furthermore, thirty-one clinical isolates collected in this study between 2015 and 2024 were included in the downstream analysis. These isolates originated from eight clinical samples, including sputum (n = 9), drainage liquid (n = 5), urine (n = 4), and wound secretion (n = 3). Genomic sequencing and assembly The clinical isolates collected in this study were recovered on China Blue agar plate and incubated at 37°C for 24 h. Individual colonies was subsequently identified using MALDI-TOF MS. Genomic DNA extraction and purification were performed using the Wizard® Genomic DNA Purification Kit (Promega, USA), and extracted DNA was used to construct sequencing libraries with the NEBNext®Ultra™ DNA Library Prep Kit (NEB, USA). Whole-genome sequencing was conducted by Magigene Biotechnology Co., Ltd. (Shanghai, China) using the Illumina Novaseq 6000 platform (San Diego, USA) to generate 2 × 150 bp paired-end reads. Quality control of the raw reads was conducted using FastQC v.0.11.5 (https://github.com/s-andrews/FastQC), and adapters trimming was carried out using Trimmomatic v0.39 [24]. The resulting clean data were then used for de novo assembly with SPAdes v4.0.0, as previously described [25]. Genome processing and quality control Except for complete genomes, draft assembled genomes were excluded from future analysis if the assembly size was less than 5 Mb or greater than 7 Mb [26]. The completeness and heterogeneity of genomes were assessed using CheckM v1.2 (≥90%completeness and ≤5% contamination). Genomes containing more than 1000 contigs or with an N50s of less than 5 Kb were also discarded. The three new unnamed species (taxons 1, 2, and 3) were eliminated because only one genome was available for each species. To remove genomic redundancy, pairwise genetic distances of all genomes were calculated using Mashtree v1.4.6 [27], and genomes with a pairwise mash distance of greater than 0 were retained. Then, genomes were compared with the genome of type strains for each species (GCA_042137965.1 for K. grimontii, GCA_015139575.1 for K. michiganensis, GCA_900636985.1 for K. oxytoca, GCA_018139045.1 for K. pasteurii, GCA_003261575.2 for K. huaxiensis, and GCA_902158555.1 for K. spallanzanii) using FastANI v1.34 [28,29], and only genomes that showed ≥95% ANI identity to the references were kept in downstream analysis. Sequence types (STs) based on the standard seven housekeeping gene were predicted in silico for each genome using the MLST tool, which were BLASTn against the “Klebisella oxytoca species complex” database [30]. All obtained high-quality genomes were annotated using Prokka v1.14.5 using default parameters [31]. Phylogenetic analysis The annotated assemblies in GFF3 format were utilized to calculate the core and pan genomes using Roary v3.13.0 with parameters set of -i of 95 and -cd of 99 [32]. In total, 1044 core genes present in ≥95% of all isolates within KoSC were identified. Recombinant regions of the core-genomic alignment were excluded due to their high diversity [26]. SNP sites were then extracted from the multi-FASTA alignment and input into FastTree v2 (https://github.com/morgannprice/fasttree) to infer a maximum-likelihood phylogenetic tree. The best-fit evolution model was determined using ModelFinder. Predictions of potential One Health transmission events Core-genome multi-locus sequence types (cgMLSTs) for the KoSC were constructed using chewBBACA v3.3.10 (https://github.com/B-UMMI/chewBBACA). The pairwise allelic distances obtained were used to determine putative genomic clusters. To minimize potential large numbers of SNPs when inferring global transmissions, core genome SNPs (cgSNPs) were determined for each species using Snippy v4.6.0 (https://github.com/tseemann/snippy), with the same genomes mentioned above as reference, respectively. Recombined regions within the alignments were removed using Gubbins v3.4 [33]. SNP sites were extracted from the alignments using the SNP-sites (https://github.com/sanger-pathogens/snp-sites) and input to generate a SNP distance matrix using the SNP-dists v0.8.2 (https://github.com/tseemann/snp-dists). Putative One Health genomic transmission clusters were defined as groups of isolates differing by fewer than 100 cgSNPs [11,34]. We emphasize that this threshold was used to identify genetically linked isolates suggestive of recent epidemiological linkage rather than to infer direct transmission events or transmission direction. Singleton isolates were strains that did not belong to any genomic clusters. Minimum spanning trees were reconstructed and visualized using the GrapeTree v1.5.0 [35] based on the cgSNP distance matrix and corresponding metadata. Identification of antibiotic resistance and virulence-associated genes Based on the NCBI AMRFinderPlus database, antibiotic resistance genes (ARGs), such as carbapenemase and ESBL genes, were predicted for each isolate using ABRicate v1.0.0 with parameters set at ≥ 95% identity and ≥90% coverage [36]. The drug classes of ARGs were classified and summarized according to the Comprehensive Antibiotic Resistance Database. The virulence-associated genes (VAGs), such as iuc, iro, and rmpA2, were predicted against the virulence factor database (VFDB) (https://www.mgc.ac.cn/VFs/main.htm). Insertion sequences were identified based on the ISfinder database (https://isfinder.biotoul.fr/). The presence of plasmid replicons was determined for each genome using the PlasmidFinder database (https://github.com/genomicepidemiology/plasmidfinder). Plasmid sequences of draft genome assemblies were extracted using the MOB-recon v3.1.9 [37], and annotated by Bakta v1.9.3 [38]. Plasmid gene clusters were aligned and compared using clinker v0.0.31 (https://github.com/gamcil/clinker). Statistical analysis All statistical analyses were carried out using R v4.4.2. The correlations between plasmids, ARGs, and VAGs were assessed based on Spearman's correlation coefficient, with Holm's Method applied for multiple testing correction. The t-test was used to compare the gene count of individual isolates among different groups. Principal coordinate analysis (PCoA) plots were performed based on Bray-Curtis dissimilarity matrix. Metadata of publicly available genomes were extracted from public databases. For certain genomes, host and geographic information were unavailable; these missing data were excluded from analysesdescribing the distribution by host or country. When missing data are included elsewhere, they are labeled as “NA”. Because the study was based on publicly available genomes with heterogeneous sampling across regions, years, and hosts, all comparative analyses were considered exploratory and intended to describe patterns within the available dataset rather than estimate population-level prevalence. Results The global distribution of KoSC A total of 31 KoSC isolates were collected from a hospital in China between 2015 and 2024, belonging to K. oxytoca (n = 14), K. michiganensis (n = 14), and K. grimontii (n = 3). Among them, 11 isolates were resistant to carbapenems, with blaNDM-1 (n = 9) being the most common resistance gene, followed by blaKPC-2 (n = 2). To place the KoSC isolates in a global context, we combined our data with all publicly available genomes of KoSC for subsequent analyses. Core genome phylogenetic analysis based on 3097 non-duplicate genomes showed that these genomes belonged to six species (Fig. 1), including K. michiganensis (n = 1370), K. oxytoca (n = 1273), K. grimontii (n = 322), K. pasteurii (n = 112), K. huaxiensis (n = 12), and K. spallanzanii (n = 7). Genomes were collected from 1974 to 2025, with the number of isolates obtained in 2018 (16.7%), followed by 2017 (9.7%), 2019 (8.9%), and 2021 (8.7%) (Fig. 1). The genomes were obtained from sixty-three countries (Fig. S1), with the USA (25.3%, 782/3097), China (10.8%, 334/3097), and the UK (10.1%, 314/3097) being the most common sources. Sampling countries were categorized into seven continents, with 38.7% of them isolated from Europe, followed by North America (29.4%), Asia (19.2%), and Oceania (9.1%). The analysis of species composition for each continent revealed that K. oxytoca was the most common species in Europe, North America, and Oceania, with K. michiganensis being the most prevalent species in the remaining continents. Most genomes with available host information were isolated from humans (94.0%), with only small proportions originating from animal (4.3%), plant (1.0%), and the environment (0.7%) (Fig. S2A). Phylogenomic population structures of the global KoSC. (A) Core-genome phylogenetic analysis. The phylogenetic tree was constructed using 1044 core genes present on the chromosomes of >95% of strains. The colored rings, from the inside out, represent species, continents, countries, hosts, and years. “Other” indicates genomes collected from other countries not listed individually; details are provided in Table S1. “NA” indicates that the information was unavailable. (B) Distribution of genomes by year. (C) Pie chart depicting the proportion of samples from each continent. (D) Bar plot illustrating the species composition within each continent. A total of 3097 genomes were included after quality control. This figure is descriptive; no statistical comparisons were performed. A total of 460 STs were identified among isolates of the KoSC, with ST199 (n = 120), ST145 (n = 112), and ST176 (n = 105) being the most prevalent. There were 135 STs containing only one isolate and 95 STs with two isolates, indicating significant evolutionary divergence among them. Notably, more than half of all STs (n = 226) were identified during the years 2016-2020, followed by 98 STs identified during 2011-2015 (Fig. S2B). The dissemination of global KoSC The concurrence of closely related genomes could be used to investigate the direct or indirect transmissions. When the allelic difference was set to 50, the number of clusters obtained by cgMLST was the largest (Fig. S3). In total, 394 clusters that comprised of 1488 isolates were identified, with the remaining 1609 showing >50 allelic distances from other genomes in this dataset. Furthermore, we inferred potential One Health genomic clusters among the four main species of KoSC. Outbreak-associated clusters were defined as those containing two or more genomes with fewer than 100 cgSNPs (Fig. 2), with more than half of the isolates in each species being linked to outbreak. For isolates of K. oxytoca, all the 76 isolates in the biggest cluster belonged to ST145 and were observed in Poland from 2009 to 2019. The second largest cluster (comprising 42 isolates) belonged to ST2, which was prevalent in Argentina, the UK, Spain, and Germany during 2001-2022. Notably, a total of eight cross-source clusters were identified for isolates within K. oxytoca isolates, including six animal-human and two environment-human clusters (Fig. S4). For genomes within K. grimontii, only six cross-country clusters were identified. The biggest cluster contained only eight ST186 isolates that were collected in China in 2023. The genomes of the ST172 cluster were collected from Switzerland, Sweden, and Poland during 2012-2016. For K. pasteurii, a total of twenty-two clusters were obtained, with only one cross-country cluster (two ST570 isolates) observed between Pakistan and Germany. The number of isolates within clusters for both K. grimontii and K. pasteurii was less than 10, suggesting sporadic transmissions. For K. michiganensis, the genomes from the biggest cluster belonged to ST108, which were sampled from ten countries between 2012 and 2024. The second largest cluster (comprising 25 isolates) belonged to ST202, which was prevalent in Israel, the USA, the UK, China, Spain, Germany, and Australia during 2012-2022. Additionally, there were four cross-source clusters, including three animal-human clusters from ST29 (n = 2), ST205 (n = 5), and ST40 (n = 4), respectively, and one environment-human cluster (sixteen isolates from ST609) (Fig. S5). Genomic dissemination of the major KoSC species. Minimum spanning trees of (A) K. oxytoca, (B) K. grimontii, (C) K. pasteurii, and (D) K. michiganensis. Genomic clusters were defined using a threshold of ≤100 core-genome SNPs. Numbers on connecting branches indicate pairwise cgSNP distances. Node colors represent sampling year, and pie chart segments correspond to individual isolates. The major STs are labeled within the corresponding circles. Numbers in square brackets indicate the number of isolates sampled in each year, and each pie segment represents a single isolate. “NA” indicates unavailable metadata. These analyses illustrate genomic relatedness and putative genomic clusters and should not be interpreted as direct evidence of transmission events. The ARGs and VAGs of global KoSC Predictions based on the NCBI AMRFinderPlus database indicated that 310 different types of ARGs were present in the genomes of the global KoSC (supplementary excel table 1). These ARGs confer resistance to 18 classes of antibiotics, with particularly high prevalence observed for resistance to cephalosporins, fosfomycin, and antibiotics via efflux mechanisms. (Fig. 3). Comparisons of antibiotics classes among different continents revealed that aminoglycoside resistance was highly prevalent in genomes from South America and Antarctica (Fig. 3A). ARGs conferring resistance to carbapenemase and diaminopyrimidine were relatively prevalent in genomes from South America. Among different species, ARGs related to antibiotic efflux were absent in K. spallanzanii and K. huaxiensis (Fig. 3B). Aminoglycoside resistance was highly present in genomes from K. michiganensis. ARGs from these 18 drug classes were identified in both K. michiganensis and K. oxytoca. Additionally, ARGs of several drug classes, such as ESBLs, colistin, and rifamycin, emerged during 2006-2010 (Fig. 3C). An upward trend in the prevalence of carbapenemase resistance was also observed during the period 2021-2025. Distribution of antimicrobial resistance genes (ARGs), virulence-associated genes (VAGs), and plasmid replicons across the KoSC. (A–C) Heatmaps showing the prevalence of ARG classes, VAG categories, and plasmid replicon types according to continent, species, and sampling year. (D–E) Boxplots comparing the number of ARGs among species and continents. (F–G) Pairwise comparisons of ARG abundance among species and continents. (H–I) Boxplots comparing the number of plasmid replicons among species and continents. Boxes indicate the median and interquartile range (IQR); whiskers represent 1.5 × IQR. Statistical comparisons were performed using the Kruskal–Wallis test followed by Dunn's post hoc test with Holm correction for multiple comparisons. All tests were two-sided. * indicates a P value of less than 0.05. The median number of ARGs per genome was 5 (interquartile range: 4–13; range: 1–32). To investigate interspecies differences in ARG counts, we performed a comparative analysis, which revealed that K. michiganensis harbored the highest number of ARGs (Fig. 3D). In contrast, K. grimontii genomes contained the fewest ARGs, significantly fewer than those of K. michiganensis (P < 2.22 × 10−16), K. oxytoca (P < 2.22 × 10−16), and K. pasteurii (P = 3.9 × 10−7). Additionally, isolates originating from Asia and South America possessed a significantly higher number of ARGs compared to those from the other four continents, while genomes from Africa carried the fewest ARGs (Fig. 3E). However, because the publicly available genomes were unevenly distributed across regions and years, this finding should be interpreted as an observed pattern within the available dataset rather than a definitive global trend. A total of 104 VAGs were identified across the KoSC genomes based on VFDB screening (supplementary excel table 2). Core genes associated with basic colonization and environmental adaptation, including ompA (99.6%), entB (99.4%), ecpA (yagZ) (99.4%), and entA (81.9%), were nearly ubiquitous across all genomes, indicating conserved functions in adhesion and iron acquisition. Notably, a large proportion of isolates harbored genes associated with the yersiniabactin iron uptake system, including ybtA (79.1%), ybtQ (79.1%), fyuA (79.0%), ybtX (79.0%), ybtS (79.0%), ybtP (78.8%), irp2 (78.0%), and irp1 (77.9%), suggesting that this siderophore system represents a widespread accessory virulence feature within the KoSC. In contrast, other virulence determinants such as mgtB (46.0%), fepC (10.2%), and toxin-related genes (astA, tcpC) were present at much lower frequencies. Importantly, classical hypervirulence-associated loci commonly observed in K. pneumoniae, such as aerobactin (iuc), salmochelin (iro), and regulator genes (rmpA/rmpA2), were rare (<0.5%) or absent in the majority of KoSC genomes, indicating a generally low prevalence of hypervirulence determinants in this species complex. When comparing the VAGs, it was observed that isolates originating from K. oxytoca and K. grimontii harbored a significantly higher number of VAGs compared to the other four species (P < 0.05) (Fig. 3F). Conversely, genomes from K. pasteurii and K. huaxiensis were found to carry the lowest number of VAGs, typically harboring 5-6 VAGs. Comparative analysis of VAGs among continents showed a significant difference for North America with Asia and Europe (Fig. 3G). In terms of plasmid replicons, genomes of K. spallanzanii contained the fewest number (Fig. 3H). In contrast, K. grimontii had the highest number of plasmid replicons, showing a significant difference (P < 2.22 × 10−16 for K. oxytoca; P = 0.0015 for K. michiganensis; P = 0.0058 for K. pasteurii; P = 0.043 for K. spallanzanii) compared to other species except for K. huaxiensis. Furthermore, a comparison of plasmid replicons among continents indicated that isolates from South America had a higher number, which was significantly higher than those from all other continents (P < 0.05) (Fig. 3I). The emergence of carbapenemase-resistant KoSC Among the genomes analyzed, carbapenemase-encoding genes were found in 38.0% of them, with 105 of these genomes carrying two or more genes. These isolates were collected from 37 countries across six continents between 1997 and 2025, with the majority identified after 2016 (Fig. 4A). The highest number of carbapenemase-encoding genomes was found in the USA (n = 329), followed by China (n = 148) and Poland (n = 107). The carbapenemase-encoding genes were primarily detected in two species across three continents: K. oxytoca in Europe and North America, and K. michiganensis in Asia, North America, and Europe. In total, 44 different types of carbapenemase-encoding genes were detected, with the most prevalent being blaNDM-1 (19.4%, 229/1178), followed by blaKPC-2 (19.3%, 227/1178), blaVIM-1 (15.6%, 184/1178), and blaKPC-3 (11.2%, 132/1178). Notably, these carbapenemase-encoding genes had also been found to emerge in animal hosts and the environment. Carbapenemase-producing genomes became more frequently represented among isolates collected in recent years within the available dataset, although this observation may have been influenced by increased sequencing efforts and unequal temporal sampling. Spatial-temporal distribution of global carbapenem-resistant isolates within the KoSC. (A) Sankey diagram illustrating the distribution of carbapenem-resistant isolates belonging to the KoSC. The diagram displays the sampled years, countries, continents, species, carbapenemase genes, and hosts of carbapenem-resistant isolates, each in a separate column. (B-C) Principal component analysis of carbapenemase-encoding genes from different continents and species, respectively. To improve interpretability of carbapenemase gene distribution analyses, we first summarized the number of isolates included from each continent: Asia (n = 283), Europe (n = 374), North America (n = 393), and Oceania (n = 54). The numbers of carbapenem-resistant isolates were 74 for K. grimontii, 556 for K. michiganensis, 495 for K. oxytoca, and 47 for K. pasteurii. For each genome, carbapenemase gene profiles were converted into presence–absence matrices. Based on these data, gene richness (number of distinct carbapenemase genes per genome) and gene composition patterns were analyzed across continents and species. Statistical comparisons of carbapenemase gene richness across groups were performed using the Kruskal–Wallis test followed by Dunn's post hoc test with Holm correction for multiple comparisons. Differences in gene composition between groups were further evaluated using principal component analysis (PCA) based on Jaccard distance. All statistical tests were two-sided, and adjusted P values < 0.05 were considered statistically significant. We continued our investigation into the distribution of carbapenemase-encoding genes among the main continents (Fig. 4B) and species (Fig. 4C). From a continent perspective, the community composition of carbapenemase-encoding genes in Europe, Asia, and North America was significantly more diverse compared to that in Oceania (P = 0.001). The richness of carbapenemase-encoding genes exhibited significantly distinct patterns among different species (P = 0.001). Overall, these analyses describe differences in carbapenemase gene richness and composition among continents and species within the available genomic dataset, rather than estimating global prevalence. Plasmids and ISs: driving force for the dissemination of ARGs To investigate the roles of plasmids and ISs in the dissemination of ARGs, the co-occurrence of ARGs with plasmids and ISs was explored (Fig. 5). In this study, a correlation was considered significant when the P value was <0.05 and the absolute value of R was >0.5. We found that 22 plasmid replicons were found to co-occur with 20 ARGs, resulting in 27 positively correlated pairs (Fig. 5A). The mcr-9.1 was strongly correlated with three replicons (IncHI2A_1, IncHI2_1, and RepA_1_pKPC-CAV1321). Notably, the carbapenemase-encoding genes, blaOXA-181 and blaOXA-2, were related to ColKP3_1 and IncFII(SARC14)_1_SARC14, respectively. From the perspective of ISs, a total of 333 different ISs were detected in the global KoSC, with 76 IS types correlating with 82 different ARGs. There was a total of 130 pairwise correlations, including three negative correlations (Fig. 5B). The majority of pairwise correlations involved beta-lactamase genes (41.5%, 54/130), followed by aminoglycoside resistance genes (20.0%, 26/130). Notably, blaKPC-2 was positively correlated with ISKpn27, and blaKPC-3 was positively correlated with ISKpn6 and ISKpn7. In addition, nine ISs (IS1203, ISEca11, ISKpn11, ISEc9, ISEc59, IS1006, IS150, ISEc12, and ISEc19) were found to co-occur positively with blaCTX-M-like genes. Associations between antimicrobial resistance genes (ARGs), plasmid replicons, and insertion sequences (ISs). (A-B) Co-occurrence networks depicting the relationships of ARGs with plasmid replicons and ISs. Nodes in the network represent ARGs, ISs, and plasmid replicons with their respective names clearly labeled. All connections in the network reflect significant correlations, defined by a significance threshold of PSpearson < 0.05 and a correlation coefficient |RSpearson| > 0.5. The thickness of each line is proportional to the strength of the correlation. (C) Alignment of plasmids encoding carbapenemase and hypervirulence genes. Carbapenemase and hypervirulence genes are annotated within the corresponding genomic regions, and accession numbers are provided at the start of each alignment. Lines connecting gene clusters between genomes represent their homology. These analyses integrate ST, ARG, plasmid, and IS information to distinguish potential clonal expansion from horizontal gene transfer. The strength of evidence for antimicrobial resistance gene mobilization varied depending on genetic context. While co-occurrence of ARGs, plasmid replicons, and IS elements across different sequence types suggests potential horizontal gene transfer, identical resistance gene-plasmid-IS combinations observed in multiple unrelated STs provide stronger support for plasmid-mediated dissemination events. In contrast, enrichment of specific ARGs within individual ST likely reflects clonal expansion. These complementary patterns indicate that both mechanisms jointly contribute to the dissemination of antimicrobial resistance within the KoSC. Compared with carbapenemase-resistant hypervirulent K. pneumonia, the emergence of carbapenemase-resistant hypervirulent K. oxytoca (CRKO) has rarely been reported. Here, four K. oxytoca isolates carrying the carbapenemase (blaDNM-5) and hypervirulence (iucABCD-iutA) genes were identified (Fig. 5C). These four CRKO isolates were found in North America and Europe, respectively, including one isolate from Slovenia in 2021 and three isolates from the USA in 2024. Discussion Klebsiella is the most prevalent gram-negative pathogen in clinical settings, and it is becoming increasingly difficult to treat [39]. To our knowledge, this study represents one of the largest high-resolution genomic analyses of the Klebsiella oxytoca species complex (KoSC), comprising 3097 genomes collected globally across multiple hosts, countries, and years. This scale enables a more comprehensive characterization of KoSC population structure, resistomes, and plasmidomes than previously reported studies, which were largely limited to regional or smaller datasets. A total of 460 STs were detected, with a median of three genomes per ST. No dominating STs were detected in KoSC, indicating a high degree of heterogeneity within this pathogen complex. Although some studies have enhanced our comprehension of bacterial transmission networks within a One Health context, there is little agreement on the appropriate levels of relatedness [40]. In this study, we conduct a One Health genomic surveillance of KoSC using the <100 cgSNPs threshold, which was determined based on the intersection of cgMLST- and SNP-based phylogenetic distance [11,34]. The cross-source genomic clusters were identified among isolates of K. oxytoca and K. michiganensis, respectively, which were absent among other species. Despite the lack of epidemiologically informative metadata, transmission inferred from closely related genomes from different sources could provide sufficient evidence. In addition, genomic clusters spanning multiple years and countries were detected among the four main species. These results provided a comprehensive understanding of potential transmission events within the global KoSC. From a clinical and public health perspective, the widespread distribution of antimicrobial resistance determinants within the KoSC highlights the potential for these organisms to serve as important reservoirs of resistance genes in both healthcare and community settings. The frequent detection of carbapenemase genes, particularly blaNDM and blaKPC variants, underscores the risk of limited therapeutic options in invasive infections caused by these organisms. In addition, the co-occurrence of resistance determinants with plasmid replicons suggests ongoing mobilization of antimicrobial resistance within clinically relevant genetic backgrounds, which may complicate infection control strategies in healthcare environments. It should be emphasized that genomic clustering based on a predefined cgSNP threshold identifies genetically closely related isolates rather than direct transmission events. Although isolates from different hosts or environmental sources that differ by fewer than 100 cgSNPs are compatible with recent shared ancestry and potential epidemiological linkage, genomic relatedness alone cannot determine transmission direction, timing, or route. Therefore, the cross-source clusters identified in this study should be interpreted as putative transmission clusters that warrant further epidemiological investigation, rather than definitive evidence of animal-to-human or environment-to-human transmission. Integrating whole-genome sequencing with detailed epidemiological metadata, temporal information, and environmental sampling will be essential to confirm transmission pathways within a One Health framework. The atlas of ARGs was predicted based on WGS, which could be useful for routine surveillance of AMR. In this study, we sought to understand the geographic patterns and temporal dynamics of ARGs in the global KoSC. The most common resistance observed was intrinsic resistance (cephalosporin and efflux pump), which resulted from chromosomally encoded blaOXY-like genes and oqxA-oqxB [41]. However, it is noteworthy that the emergence of acquired resistance in KoSC, such as carbapenemase, aminoglycoside, ESBLs, and colistin. The AMR pattern varied depending on the species, with K. michiganensis being a hotspot for aminoglycoside resistance (aph(3′)-Ia and aph(3″)-Ib), and efflux pump encoding genes being absent in K. huaxiensis and K. spallanzanii. The number of VAGs showed significant differences among species, though variation among continents was less pronounced. The emergence of multidrug-resistant isolates contributed to the enrichment of VAGs (blaTEM-1, dfrA12, fosA_gen, sul1, and tet(A)) in K. oxytoca and K. grimontii. Although virulence-associated genes were less diverse compared with classical hypervirulent K. pneumoniae, our analysis revealed that genes involved in adhesion (e.g., fimbrial systems) and iron acquisition (particularly the yersiniabactin system) were widely distributed across KoSC members. These factors may enhance environmental persistence and colonization capacity, potentially facilitating opportunistic infections in vulnerable hosts. Notably, the presence of siderophore systems such as yersiniabactin may provide a selective advantage in iron-limited host environments, suggesting that even low-virulence backgrounds may contribute to opportunistic pathogenicity under appropriate conditions. The regional, species-specific, and temporal differences in antimicrobial resistance observed in this study should be interpreted with caution. Because the analyses were based on publicly available genomes, the dataset is inherently affected by unequal geographic representation, differences in sequencing capacity among countries, temporal variation in surveillance activities, and potential overrepresentation of outbreak-associated isolates. Consequently, the observed patterns may not fully represent the true global epidemiology of the KoSC but rather reflect the characteristics of the currently available genomic collection. Nevertheless, the large sample size and broad geographic coverage provide valuable insights into the contemporary genomic diversity and antimicrobial resistance landscape of the KoSC. Plasmids and ISs play crucial roles in the dissemination of ARGs between bacteria, leading to the spread of antibiotic resistance in various environments, including clinical settings [42]. This study conducted the comprehensive screening for plasmids and ISs driving the dissemination of ARGs in the KoSC. The findings revealed that 22 plasmid replicons significantly contributed to the dissemination of ARGs among global KoSC isolates. It is worth noting that the study identified the concurrence of blaOXA-181 and ColKP3_1, highlighting the role of plasmids in the spread of carbapenemase genes within the KoSC. Moreover, the study found that ISs were widely present in global KoSC isolates, with ISKpn27, ISKpn6, and ISKpn7 playing important roles in the dissemination of carbapenemase genes. The mobilization of different ARGs was not solely dependent on one type of IS; rather, it was observed that one IS could be associated with several ARGs or multiple ISs could be linked to one ARG. Additionally, the study identified the presence of the K. pneumoniae virulence factor (aerobactin) and carbapenemase gene (blaNDM-5) within the same genome [11,43], suggesting the emergence of CRKO strains. From a One Health perspective, the detection of genetically related isolates across human, animal, and environmental sources supports the hypothesis that KoSC lineages circulate at the human-animal-environment interface. However, due to the retrospective nature of publicly available genomic data, the directionality and mechanisms of transmission cannot be inferred. Future studies should therefore prioritize prospective, longitudinal sampling across clinical, veterinary, and environmental reservoirs, combined with high-resolution genomic surveillance and detailed epidemiological metadata. Such integrated approaches will be essential to accurately resolve transmission pathways and to assess the role of environmental reservoirs in the maintenance and dissemination of antimicrobial resistance within the KoSC. Our study has several limitations. Cross-source transmissions were identified using clinical samples derived exclusively from humans, while animal and environmental samples were based solely on database entries. Furthermore, the analysis does not adequately address potential biases inherent in the retrospective data range, particularly the impact of evolving sequencing technologies on genome quality. The study may also underrepresent less-developed regions, such as Africa, potentially biasing transmission inferences and masking rare but epidemiologically significant events due to the high cost of sequencing. Moreover, inter-host transmissions in this study only depended on phylogenomic clustering and host information, which requires additional supporting evidence and further research. In addition, we acknowledge that this threshold is not specific to the KoSC, but it may still be reasonable. Consequently, the identified cross-source genomic clusters should be interpreted as evidence of potential epidemiological linkage rather than confirmed transmission events. In addition, despite removing redundant genomes and applying stringent quality-control criteria, residual sampling bias could not be completely eliminated. Publicly available genome collections are influenced by differences in regional sequencing capacity, surveillance priorities, and the inclusion of outbreak-associated isolates. Furthermore, incomplete epidemiological metadata prevented formal adjustment for region, year, and source. Therefore, the regional, temporal, and host-associated trends reported in this study should be interpreted as observations within the available genomic dataset rather than definitive estimates of global epidemiological patterns. Conclusion This study provides a high-resolution global genomic atlas of the KoSC, revealing substantial species diversity, increasing antimicrobial resistance, and widespread dissemination of mobile resistance determinants. Our findings support the implementation of genomic surveillance programs that prioritize high-risk species, particularly K. michiganensis and K. oxytoca, as well as internationally disseminated lineages such as ST145, ST108, and ST202. Surveillance should also routinely monitor major carbapenemase genes, including blaNDM-1, blaKPC-2, blaVIM-1, and blaKPC-3, together with key mobile genetic elements such as ColKP3_1, IncHI2-type plasmids, ISKpn27, ISKpn6, and ISKpn7 that facilitate their dissemination. Because closely related isolates were identified across human, animal, and environmental sources, integrated One Health genomic surveillance is warranted to detect cross-sector transmission at an early stage. In addition, the emergence of carbapenem-resistant hypervirulent K. oxytoca carrying both blaNDM-5 and aerobactin genes highlights the need for routine genomic screening for convergent resistance and virulence. Both clonal expansion and horizontal gene transfer contribute to resistance dissemination. Collectively, these findings provide actionable priorities for strengthening global surveillance and guiding targeted infection prevention and antimicrobial resistance control strategies. Data availability These genomes sequenced in this study have been deposited in the National Center for Biotechnology Information database under BioProject accession PRJNA1271987. All assembled genome sequences supporting the conclusions of this article have been listed in Table S1. Ethical statement Ethical approval was waived for the genomes from NCBI, as the isolates were public available. Isolates collected in this study was approved by the Ethics Committee of Beijing Chaoyang Hospital (2024-381). Funding This work was supported by the National Natural Science Foundation of China (82502766), and the Beijing Chao-Yang Hospital Golden Seeds Fundation (CYJZ202220). CRediT authorship contribution statement Wentao Zhu: Conceptualization, Funding acquisition, Supervision, Writing – original draft, Writing – review & editing. Xi Chen: Data curation, Formal analysis. Hong Shen: Formal analysis, Resources. Chunxia Yang: Resources. Qian Liu: Conceptualization, Validation, Visualization, Writing – original draft, Writing – review & editing. Li Gu: Project administration, Writing – review & editing. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.