This study reveals significant genetic diversity underlying the biosynthesis of aerobactin and salmochelin in K. pneumoniae, but shows the distribution of iuc and iro locus variants is highly structured within the population. Our data indicate that most of the burden of these hypervirulence-associated siderophores in the K. pneumoniae population is associated with two dominant virulence plasmids, which we define here as KpVP-1 and KpVP-2, that differ in terms of gene content (Fig. 2) and are each associated with co-segregating sequences of the FIB_K replicon, iuc and iro loci (Fig. 1, Fig. 5). These dominant virulence plasmid types are each represented by one of the previously characterised K. pneumoniae virulence plasmids (18,20), pK2044 (KpVP-1, encoding iro1 and iuc1) and Kp152.145pII (KpVP-2, encoding iro2 and iuc2); both also carry hypermucoidy determinants, and together they account for 74% and 14% of the iuc+iro+ K. pneumoniae genomes analysed. Importantly, our data indicate that each of these common virulence plasmid variants is maintained at high prevalence in a small number of known hypervirulent clones: KpVP-1 in CG23 (96%, including pK2044 (18)), CG86 (80%, including pLVPK (19)) and CG65 (79%); KpVP-2 in CG66 (100%, including Kp152.145pII) and CG380 (100%) (Fig. 3). This suggests that both plasmid types can persist for long periods within a host bacterial lineage as it undergoes clonal expansion; indeed our recent study of the evolutionary history of CG23 indicates that KpVP-1 has been maintained in this clonally expanding lineage for at least a century (22). Notably we also detected KpVP-1 at low prevalence in numerous other K. pneumoniae lineages and KpVP-2 at low prevalence in one other lineage, suggesting the possibility of wider dissemination of both plasmid types by occasional transfer to new lineages (Fig. 3). Given the stability of the plasmids observed in several clonal groups, we speculate that some of these transfer events will result in the emergence of novel hypervirulent strains that can stably maintain the plasmid into the future. In contrast, the non-plasmid form of iro (iro3, occasionally integrated into the chromosomes of K. pneumoniae via ICEKp1 was found at low prevalence (<0.5%) and included just one of the 79 ST23 isolates analysed (NTUH-K2044, in which ICEKp1 was first described), 1/1 ST5, 1/21 ST111 (13%), 1/2 ST198, 2/15 CG25, 2/2 ST493 and 5/5 ST60. Hence while ICEKp1 is somewhat dispersed in the K. pneumoniae population, it shows little evidence of stability within lineages, consistent with our previous observations regarding ICEKp in general (8).

We also detected several novel iuc+ or iro+ plasmid types, the most common being the group of iuc2a plasmids (21% of all iuc+ strains) that were detected in respiratory isolates from CG3, CG82 and CG90 and mostly originated from historical collections (50). Interestingly these combine an iuc sequence closely related to that of KpVP-2 (Fig. 1) with a FIB_K replicon sequence very close to that of KpVP-1 (Fig. 5), and showed substantial mosaicism and gene content variation (Fig. 2, Fig. 4). The iuc3 lineage was also quite common (5.3% of all iuc+ strains) and associated with a variety of diverse plasmids, most of which carried the FII conjugative transfer region and thus are likely self-transmissible (Fig. 2, Fig. 4). It is notable that iuc2a and iuc3 plasmids were not only quite rare in the population, but also showed less evidence of stable maintenance within K. pneumoniae lineages (Fig. 3) and lower stability of gene content (Fig. 2) than the dominant KpVP-1 and KpVP-2 plasmids (Fig. 4). The position of iuc2a and iuc3 in the iuc trees (Fig. 1, Fig. S2) suggests that both derive from other K. pneumoniae loci, hence we speculate it is the properties of the plasmids mobilising these loci, and not the siderophore biosynthesis loci themselves, that makes these variants less widespread in the K. pneumoniae population. This variation in gene content may be a consequence of self-transmissibility, exposing the plasmids to a wider gene pool of host bacteria and providing opportunities for gene content diversification, which could potentially include AMR genes. Notably the iuc3 plasmids carry an arsenal of additional virulence loci involved in iron metabolism and resistance to heavy metals, reminiscent of KpVP-1 (Fig. 2).

The other novel plasmids appear to derive from outside K. pneumoniae (Fig. 1, Fig. S2). Most concerning are the four E. coli-derived plasmids we detected carrying iuc5 (and occasionally iro5) in USA and Brazil, three of which were found in the MDR hospital outbreak-associated clone CG258. Whether these aerobactin plasmids harbour AMR genes as they do in E. coli is not currently resolvable; however it seems that conjugative E. coli plasmids such as D3 plasmid A do have the potential to deliver hypervirulence and multidrug resistance to K. pneumoniae strains in a single step. A recent study of K. pneumoniae submitted to Public Health England used PCR to screen for isolates carrying both carbapenemase genes and rmpA, as a marker of the virulence plasmid, and identified a plasmid harbouring iuc, rmpA, rmpA2 and the AMR genes sul1, sul2, armA, dfrA5, mph(A) and aph(3’)-VIb (28). To our knowledge this is the first report of a complete sequence of a K. pneumoniae plasmid harbouring both AMR and virulence genes. The isolate (ST147) was not included in our original screen, however subsequent analysis using Kleborate plus manual inspection of the plasmid sequence reveals it carries iuc1 (AbST63, a novel single locus variant of AbST1 which is typical of hypervirulent clones CG23, CG65 and CG86) and appears to be a mosaic carrying sequences from KpVP-1 (40% coverage), an IncFII conjugative transfer region and transposons carrying AMR genes.

The presence of aerobactin synthesis loci in the iuc5+ K. pneumoniae isolates we identified here was not reported in the original studies (51,52), and thus it is not known whether they actually produce aerobactin or show enhanced virulence. This highlights the need to raise awareness of the iuc and iro loci as potentially clinically relevant hypervirulence factors, and to screen for them in isolates and genome data. The latter we aim to facilitate via the genotyping schemes established here, which can be used to easily screen new genome assemblies using Kleborate (https://github.com/katholt/Kleborate/) or BIGSdb-Kp (http://bigsdb.pasteur.fr/klebsiella/klebsiella.html), or new short read data sets using SRST2 (https://github.com/katholt/srst2). PCR primers suitable for screening for iro and iuc can be found in Lee et al (53). Notably many studies rely on the hypermucoidy phenotype to identify hypervirulent strains, however this is dependent on growth conditions (54), and recent studies indicate that aerobactin synthesis is a more important virulence determinant (16,13,14). Our data suggest that hypermucoidy screening would typically pick up most of the common aerobactin plasmids KpVP-1, KpVP-2 and iuc2a+ plasmids; but not those carrying iuc3, or the iuc5 plasmids from E. coli. Additionally, it is important not to conflate the presence of the core chromosomal receptor gene iutA with the ability to synthesise aerobactin, which is encoded in the iuc locus (6). False positive detection of the aerobactin locus version of iutA can be avoided by using an identity threshold of <20% divergence. Tellurite resistance has also been suggested as a phenotypic screen to identify hypervirulent isolates of CG23, CG65 and CG86(55); our data confirms this is a good marker for KpVP-1 (92.6% carry ter), but not for other aerobactin plasmid types (Table S10).