Klebsiella pneumoniae is regarded globally by the World Health Organization (WHO) as a priority antimicrobial-resistant (AMR) pathogen requiring new control strategies [1]. These include rapid identification and containment of high-risk AMR clones such as the carbapenemase-producing (CP) variants, augmented with vaccines, bacteriophages, or immunotherapies that target surface antigens. However, K. pneumoniae is highly diverse, hindering the development of such strategies and our ability to study its molecular epidemiology in a relevant time-frame.

This diverse bacterial species is generally associated with a range of differing community and healthcare-associated infections, but can be particularly problematic when the organisms gain access to sterile sites such as the cerebrospinal fluid and the bloodstream. Such infections are often characterised by rapid onset and multi-drug resistance (MDR), including resistance to third-generation cephalosporins and/or carbapenems. Antimicrobials are the primary treatment strategy but options can be severely limited by AMR, particularly for third-generation cephalosporin- and carbapenem-resistant strains causing invasive infections [2]. Concomitant with this are elevated mortality rates and treatment costs [3].

K. pneumoniae is among the most common cause of bloodstream infections (BSI) in South (S) and Southeast (SE) Asia [4–6] where it is associated with a high mortality rate [4, 7]. Available data suggest a heterogenous landscape in terms of drug resistance; for example, CP strains are rare in SE Asia (< 1–4% [5, 8, 9]) but common in S Asia (28–70% [10, 11]) and the prevalence of extended-spectrum beta-lactamase (ESBL, confers resistance to the third-generation cephalosporins) producing organisms varies from 12 to 79% in these regions [4–6, 8, 10]. Studies investigating ESBL and CP variants in S/SE Asia implicate CTX-M-15 as the most common ESBL type [12, 13], while NDM and OXA-48-like enzymes are the most commonly described carbapenemases [11, 14–16]. However, there is currently only limited information about the underlying population structure of these organisms in terms of multi-locus sequence types (STs), or genomically defined phylogenetic lineages.

Searching PubMed for reports of multi-locus sequence typing (MLST) or whole-genome sequence analysis of K. pneumoniae BSI (see “Methods” for search terms) yielded just nine studies reporting ST information for India, Nepal, Vietnam, Thailand, Laos, Cambodia and/or Hong Kong (the foci of this work). The majority of these (six of nine) were case reports or genome announcements from India, while an additional study from India described a screen of Enterobacteriaceae isolated from neonatal sepsis, including 12 K. pneumoniae [17–24]. Together, these works from India report ST information for a single ESBL-producing strain (ST2318) and 24 CP strains; five each of ST231 and ST347; two each of ST29, ST147, ST1224 and ST2558; and one each of ST11, ST43 and ST101. Carbapenemase genes were reported for 21 of the CP strains and included 11 bla_NDM-1, 8 bla_OXA-48-like, one each of bla_NDM-7 and bla_KPC. A single report from Nepal described two neonatal sepsis outbreaks [7], the first caused by ST15 carrying bla_NDM-1 and bla_CTX-M-15, and the second by ST1559 carrying bla_CTX-M-15 without any carbapenemases. The final report described K. pneumoniae BSI isolates from Hong Kong that produced DHA enzymes (resulting in third-generation cephalosporin resistance). Among the genotyped isolates were four ST11, and one each of ST17, ST23 and ST39 [25].

It is clear from these reports that several well-known globally distributed CP/ESBL-associated STs are present in S/SE Asia, e.g. STs 14, 15, 17, 29, 101, 147 and 231, a proposition that is further supported by studies exploring K. pneumoniae isolates from a broader range of clinical specimen types [12, 13, 15, 26, 27]. However, there remains a clear lack of systematic studies with which to fully understand the CP, ESBL and broader population genotypes of K. pneumoniae causing BSI in these regions.

Capsule- or lipopolysaccharide- (LPS) targeted immunisation against K. pneumoniae has been proposed as an alternative strategy for prevention or therapy of MDR BSI. Hence, capsular serotype (K-type) and LPS (O-type) profiles among BSI isolates are also of interest. More than 130 capsular serotypes have been predicted on the basis of genome data [28]. Among the studies identified through the formal search described above, two did not report K-types and three included only a limited assessment, i.e. PCR screening for a small number of types that are associated with enhanced virulence (K1 and K2 [17, 24], or K1, K2, K5, K20, K54 and K57 [23]). Among the remaining three studies, two reported single K17 and K64 CP isolates in India [19, 20], while the Nepalese outbreak study reported the KL14 capsule locus, encoding capsule type K14 (ST1559) and a novel locus that was later defined as KL112 (ST15) [7]. None of these studies reported O-types. We searched PubMed for additional studies reporting K- and O-types of K. pneumoniae BSI in our regions of interest (search terms in “Methods”), revealing one additional report describing 47 isolates (including 28 BSI isolates) from a suspected neonatal intensive care unit outbreak in Madras, India [29]: Thirty-eight isolates were shown to be closely related by pulsed-field gel electrophoresis (MLST not reported), all of which expressed the same novel K and O serotype. Among the remaining nine unrelated isolates, three K2:O1 were identified in addition to one each of K23:O1, K54:O1, K62:O1, K62:O2, K-non-typeable:O1 and K2:O-novel.

The capsule, alongside the LPS, type I and type III fimbriae, and the enterobactin siderophore, are common to all K. pneumoniae strains and are required for pathogenesis. A number of accessory virulence determinants (including RmpA/RmpA2, which upregulate capsule expression; the colibactin genotoxin; and the yersiniabactin, aerobactin and salmochelin siderophores that promote systemic survival and dissemination [30–34]) are overrepresented among isolates causing invasive disease in human population studies [35–37]. Each has also been shown to enhance virulence in murine infection models through comparison of isogenic knockout mutant strains [31–34]. K. pneumoniae expressing several of these determinants are associated with severe community-acquired invasive disease, often manifesting as liver abscess. These are classified as “hypervirulent” infections; they occur globally but are most commonly reported in SE Asia [38]. They are predominantly associated with clones ST23, ST86 and ST65 [39, 40], which typically express the K1/K2 capsules. The dominance of a small number of clones combined with strong linkage between the rmpA/A2, aerobactin and salmochelin synthesis loci (which are co-located on virulence plasmids [41, 42]) has created difficulties for understanding the relative importance of these key virulence factors. However, recent studies implicate aerobactin as the most important siderophore and a key biomarker for hypervirulence alongside the peg-344 transporter gene (sometimes misannotated as pagO, in strong genetic linkage with aerobactin due to being located nearby on the same virulence plasmid) [37, 43, 44]. Notably, despite its apparent relevance as a hypervirulence biomarker, peg-344 isogenic knockout mutants did not show reduced virulence in either of two independent murine models of hypervirulent disease [44, 45].

To date the majority of hypervirulent K. pneumoniae have remained susceptible to antimicrobials [38, 46] (except ampicillin). However, the last few years have seen increasing reports of “convergent” K. pneumoniae that are both hypervirulent (defined as strains carrying the iuc aerobactin locus) and MDR ESBL/CP either due to acquisition of an MDR plasmid by a hypervirulent strain [47–51] or by acquisition of a virulence plasmid by an MDR strain [49, 52–56]. The majority of these reports represent sporadic isolations. However in 2017, it was reported that an outbreak of ventilator-associated pneumonia in a Chinese hospital, resulting in five deaths, was caused by a CP ST11 strain that had acquired a virulence plasmid harbouring the iuc locus, plus rmpA2 (without the salmochelin locus (iro), rmpA or peg-344) [52]. Notably, the outbreak strain showed enhanced virulence in the Galleria mellonella infection model compared to a wild-type ST11 lacking the virulence plasmid and a plasmid cured variant [52], further supporting the importance of the iuc locus. Subsequent retrospective investigations revealed that similar CP ST11 iuc + strains had already been silently spreading within China prior to the initial report [52, 57].

Given the high burden of both MDR and hypervirulent K. pneumoniae infections, S/SE Asia may represent a hub for MDR-virulence convergence, with the potential for outbreaks of severe disease with extremely limited treatment options such as that reported in China [52]. However, so far no studies have provided a complete molecular epidemiological picture (STs, serotypes, AMR and hypervirulence determinants) of K. pneumoniae BSI agents in S and SE Asia, leaving a gap in our knowledge about the local pathogen population and the prevalence and/or diversity of convergent strains. Here we present a genomic epidemiology study of BSI K. pneumoniae from seven major healthcare facilities across S/SE Asia, leveraging a recently established genomic framework that incorporates rapid genotyping of clinically important features (ST, AMR, hypervirulence, and capsule and lipopolysaccharide synthesis loci). The data are highly relevant to the design of K. pneumoniae control strategies, revealing a diverse population with high rates of AMR and virulence loci, and high prevalence of convergent strains.

The genomes of 393 presumptive K. pneumoniae BSI from seven countries across S/SE Asia were sequenced (Fig. 1a, Additional file 1: Table S1). Twenty-eight genomes were identified as Enterobacteriaceae species outside of the K. pneumoniae species complex and were excluded. Among the remaining 365 isolates, the majority (n = 331, 91%) of organisms were confirmed to be K. pneumoniae. Among the regional comparator samples, these accounted for a higher proportion of genomes from S than SE Asia (98% vs 87%, p = 0·011; Table 1). The remaining isolates (Additional file 2: Table S2) were K. quasipneumoniae subsp. similipneumoniae (n = 20, 5·5%), K. variicola (n = 9, 2·5%) and K. quasipneumoniae subsp. quasipneumoniae (n = 5, 1·4%), which are indistinguishable from K. pneumoniae by microbiological methods [35, 67]. Unfortunately, we did not have access to patient clinical data and hence were not able to assess the likelihood that individual BSI cases were hospital- or community-acquired, or to explore patient co-morbidities, which we expect to be quite diverse.

The 331 K. pneumoniae were highly diverse and comprised 120 individual STs (Simpson’s diversity index = 0·97), the majority (61%) of which were represented by a single isolate. Nevertheless, we observed four common STs that each accounted for > 5% of the sequenced organisms: ST15 (n = 37, 11%), ST23 (n = 28, 8·5%), ST14 (n = 22, 6·6%) and ST231 (n = 17, 5·1%; Fig. 1c). Among these, only ST15 was common across all sites, whereas ST23 was significantly associated with SE Asia (p = 0·015) and ST14 and ST231 were both significantly associated with S Asia (p < 0·01; Table 1).

This work represents the first broad genomic study of K. pneumoniae causing BSI in S/SE Asia and Hong Kong, regions that are facing a combination of community-acquired invasive hypervirulent K. pneumoniae, unregulated use of antimicrobials, and the emergence and spread of MDR pathogens.

The diversity of K. pneumoniae causing BSI in S/SE Asia represents a significant challenge to therapy using existing or novel agents. Our study revealed a highly diverse set of isolates, both in terms of chromosomal STs and surface polysaccharide loci. We were unable to stratify the isolates into hospital-associated and community-acquired cohorts, but anticipate that the former accounts for the majority of the diversity in our sample given that the majority of community-acquired K. pneumoniae BSIs are known to be caused by a small number of distinct clones with limited surface antigen diversity [38–40]. This is consistent with the hypothesis that the majority of hospital-associated BSIs originate from the patients’ personal gastrointestinal microbiota rather than from intra-hospital transmissions [67, 79]. The latter is highly relevant to the design of vaccines or other interventions targeting K. pneumoniae capsules or lipopolysaccharide, which are considered of high importance in response to increasing prevalence of ESBL and CP strains, both of which were common in our sampling locations (47% and 17% of all isolates, respectively). From our sampling strategy, we crudely estimate a non-cross-reactive capsule-targeted vaccine would need to include ≥ 16 serotypes in order to provide potential protection against > 50% of the BSI K. pneumoniae isolates across these study sites. The 16 most common K-loci would cover 61% of all ESBL and 68% of all CP K. pneumoniae across S/SE Asia. Alternatively, we estimate that a completely immunising vaccine targeting O1, O2 and O3b would hypothetically protect against 79% of the BSI K. pneumoniae isolates in this study, equating to 79% of all ESBL-containing isolates and 70% of CP.

Alongside the high prevalence of ESBL and carbapenemase genes, our data revealed high prevalence of known hypervirulence determinants with the iuc locus detected in 28% of all isolates included in this study, more than double the prevalence seen in previous studies focused outside of this region [78]. We suggest this likely reflects the combination of a comparatively greater prevalence of strains from community-acquired infections (particularly in SE Asia) plus comparatively elevated prevalence among MDR hospital-associated strains. Accordingly, 33 of 210 (16%) genomes predicted to be MDR also carried iuc (note includes 5 genomes that were MDR variants of known hypervirulent clones). While the iuc locus itself was common in both S and SE Asia, the distribution of iuc lineages differed (Table 1). Specifically, iuc1 (associated with the characteristic KpVp-1 virulence plasmid [78]) and iuc3 were more common in SE Asia where they were each detected among numerous distinct STs. This is consistent with their local dissemination or recent acquisition from a common source, a finding supported by the discovery of four novel but similar iuc3 encoding plasmids in these isolates. Of greatest concern was the detection of at least nine distinct convergence events between AMR and hypervirulence (encompassing 7·3% of isolates) plus seven novel dual AMR + iuc containing plasmids, increasing the total number of convergent plasmids reported so far by almost 50% [49, 80–83]. Given that our isolate collection represented a snapshot from seven diagnostic laboratories, we predict that many additional convergent plasmid variants await future identification.

It remains unclear if convergent MDR-virulent K. pneumoniae plasmids and/or strains are fit for widespread dissemination or simply represent transient events that are rapidly purged from the population. The latter could explain the historical absence of reports of MDR hypervirulent infections and is consistent with the apparent lack of dissemination of any of the numerous reported convergent MDR ST23 variants, despite ample evidence that the susceptible hypervirulent ST23 clone is capable of global distribution [39, 40, 49, 84–87]. However, it is important to note that other factors have also been proposed; for example, the possibility that hypervirulent and MDR clones preferentially inhabit distinct ecological niches outside of the human host, thereby limiting the opportunity for transfer of genetic material between them; or that there may be physical and/or mechanistic barriers to genetic exchange [88]. Additionally, there is now clear evidence that at least a subset of the globally recognised MDR clones with acquired virulence plasmids are able to cause persistent local problems and spread more broadly: These include the CP ST231 iuc + strain reported here from the Indian hospital, a previously reported CP ST15 iuc + in a Pakistani hospital [53] and CP ST11 iuc + in China, which was able to disseminate for several years without detection [52, 57]. Fortunately, most MDR-virulent infections thus far appear to be hospital-associated rather than community-acquired, suggesting that unlike the hypervirulent clones, these convergent strains may be less able to spread outside of the hospital environment and/or unable to cause invasive disease in otherwise healthy hosts. Nevertheless, the combination of enhanced virulence plus limited treatment options is an important concern that warrants careful attention. Combined with our data indicating high prevalence and diversity of convergent strains in the S/SE Asian region, it is clear that there is a need for enhanced K. pneumoniae surveillance to rapidly identify and monitor convergent strains and/or plasmids.

A limitation of this work is that the data represents a retrospective convenience sample of K. pneumoniae BSI isolates by the participating diagnostic laboratories during routine activities during overlapping time periods. This prevented statistical comparisons between individual sites and should be considered when extrapolating gene prevalence information to the broader K. pneumoniae population. Additionally, we did not have access to patient clinical data, which has limited our interpretation of the data in the context of patient co-morbidities, and stratification of isolates into community-acquired and hospital-associated cohorts. We focussed on BSI, as this allowed for convenient and consistent retrospective identification of isolates associated with invasive disease for inclusion in the study. While this presentation arguably reflects the greatest clinical need, the siderophore virulence loci are known to be more prevalent in BSI compared to other infections [35], and hence the prevalence of iuc estimated from our sample may be higher than that of the broader population of K. pneumoniae causing infections in these regions. In addition, we note that our inferences about AMR are based on genotypic information, which is highly predictive of AMR in K. pneumoniae but not perfectly correlated [67, 89]. Nevertheless, our analyses reveal valuable insights and provide essential data to motivate enhanced public health surveillance.

K. pneumoniae BSI isolates from South and Southeast Asia represent diverse STs, capsule and LPS serotypes, with high prevalence of MDR, hypervirulence-associated loci and convergent MDR-virulent strains. Our study represents a blueprint for genomic surveillance of emerging AMR pathogens in this region and we urge the coordination of similar activities internationally. By rapidly detecting resistance and virulence genes in the context of clonal and surface antigen diversity, our approach provides critical information that can be used to simultaneously track the emergence and dissemination of clinically important variants, guide antimicrobial therapy and assess potential mechanisms and targets for containment and intervention. In combination with rapid reporting and data sharing, this approach will permit researchers and public health professionals to recognise and control the growing public health threat of AMR K. pneumoniae.