The diversity of Klebsiella pneumoniae surface polysaccharides

1Supplementary dataset S1 lists the ENA accession numbers for the 573 publicly available K. pneumoniae whole-genome Illumina read sets analysed in this study (http://www.ebi.ac.uk/ena).

2Representative rfb and cps locus sequences have been deposited in GenBank, see Table S2 for the list of GenBank accession numbers (http://www.ncbi.nlm.nih.gov/genbank/).

3Vaccines offer a potential remedy against the increasing threat of pan-drug-resistant Klebsiella pneumoniae strains. However the high diversity of surface antigens poses a challenge for vaccine design. This work is the first, to our knowledge, to catalogue naturally occurring polysaccharide antigen biosynthesis gene clusters in a globally representative collection of K. pneumoniae isolates, enabling us to identify novel serotypes and perform an epidemiological analysis. We show that only three O-antigen serotypes account for the majority of infections, offering a promising target for vaccine design.

4Klebsiella pneumoniae is a leading cause of hospital- and community-acquired infections (including urinary tract infections, pneumonia, bacteraemia and soft tissue infections), primarily afflicting the young and immunocompromised, despite being part of the normal human intestinal microbiota and able to colonize the skin and nasopharynx of healthy individuals. K. pneumoniae are naturally resistant to antibiotics such as amino-penicillins and carboxy-penicillins. Increasingly, treatment options are diminishing, leaving third-generation cephalosporins and carbapenems as the remaining alternative. However, the emergence of isolates carrying genes encoding extended spectrum beta lactamases (ESBL) and carbapenemases has raised alarm because it removes these last line treatment options and effectively negates the use of a whole class of antibiotics, with few alternatives. Therefore multiple agencies including the WHO, US Centers for Disease Control and Prevention and the UK Department of Health singled out K. pneumoniae as a global health concern.

5Historically Klebsiella isolates have been classified into serotypes and tracked using typing antisera. Serotyping is based on the recognition of distinct variations of surface-exposed polysaccharides, namely O-antigens and K-antigens, by specific antibodies, resulting in different O and K serotypes. O-antigens are the outermost part of the lipopolysaccharide (LPS), whereas K-antigens belong to the bacterial capsule polysaccharide (CPS). The number of serotypes has been estimated to be eight for O-antigens and 77 for K-antigens.

6Multivalent protein-conjugate polysaccharide vaccines have been demonstrated to be highly successful and effective against bacterial pathogens, such as Streptococcus pneumoniae. In K. pneumoniae, polyvalent vaccines based on the K-antigen have been developed and reached Phase I trials in humans. However, the high diversity of K-antigens and the confusing seroepidemiology render a vaccine with a broad coverage complex to develop and thus very costly. Compared with other Enterobacteriaceae, such as Escherichia coli [161 defined O serotypes] and Shigella flexneri [at least 47 O serotypes], Klebsiella has a surprisingly low number of reported O serotypes which promises a more viable alternative for vaccine development compared with K-antigen-based vaccines. Whole-genome sequence data allows us to verify and assess the diversity of these clusters and also estimate their frequency in contemporary isolates to assess the contribution of isolates bearing these markers.

7The O-antigen biosynthesis enzymes are encoded on the rfb locus. To date, seven O-antigen clusters have been defined for K. pneumoniae, associated to serotypes O1, O2, O3, O4, O5, O8 and O12 (Table S1) and the rfb O-antigen biosynthetic pathway is well described. In essence it is an ABC-transporter-dependent pathway functionally composed of three types of enzymes: those responsible for (i) biosynthesis of nucleotide-activated sugars, (ii) polysaccharide repeat-unit synthesis and (iii) assembly of the repeat units and transport across the membrane (flippases).

8Both the O1 and O2 antigen polysaccharide chains are based on a repeat-unit designated as d-galactan I. The difference being that the O1 antigen is capped by a distal d-galactan II unit, whereas O2 is not. d-galactan II is the only known O-antigen polysaccharide in Klebsiella for which biosynthesis is enabled by genes (wbbY and wbbZ) unlinked to the rfb cluster. The O1/O2 biosynthetic cluster occurs in two variants (Fig. 1a). Variant 1 consists of the transporter genes (wzm and wzt), three glycosyltransferases responsible for repeat-unit synthesis and an UDP-galactopyranose mutase, responsible for the synthesis of the polysaccharide subunit d-galactan I. Variant 2 is extended and carries three additional putative glycosyltransferases encoded by gmlABC. These additional genes are thought to explain the existence of a recently defined O2 subtype whereby the gml gene products modify d-galactan I, to give the chemically and antigenically distinct d-galactan III explaining the different subtype. The serology of O1 isolates is proposed to be unchanged by GmlABC, because in this instance d-galactan II (encoded by WbbY and WbbZ) is the dominant antigenic epitope.

9O-antigen polysaccharide structure and rfb sequence of K. pneumoniae serotypes O3 and O5 are identical to those of E. coli serotypes O9a and O8, respectively. Biosynthesis enzymes of O3 and O5 differ in the sequence of their mannosyltransferase (WbdA) and methyltransferase (WbdD). WbdD, in complex with WbdA, regulates the mannose chain length by capping the growing chain with a phosphate and methyl group in O3 and a methyl group only in O5.

10For completeness the O8 polymer is identical to the O1 antigen, with the exception of the partial O-acetylation of the d-galactan I repeat unit in O8 and serotype O9 may be considered as a subtype of serogroup O2. Serotype O9 Wzm and Wzt sequences share >95 % identity to the O8 Wzm and Wzt orthologues

11Whilst most members of the Enterobacteriaceae have one conserved LPS core, two types of outer cores have been reported for K. pneumoniae; they differ in a set of three genes within the waa operon, which is independently located in the genome.

12The seroepidemiology of K. pneumoniae has previously shown that for human-host-associated isolates the most prevalent O serotypes are O1, O2 and O3, with O1 being dominant in human disease. The observed prevalence of O4, O5 and O12 differs among the studies, whereas the significance of serotypes O8 and O11 remain equivocal. For O11 this is due to the lack of available rfb sequences and publicly available O11 isolates, so they could not be considered in this study. In addition, little is known about the association between O serotype and disease presentation, the only exception being the observation that serotype O3 isolates are more commonly isolated from blood samples compared with urine samples.

13The K-antigen biosynthesis enzymes are encoded on the cps (capsule polysaccharide synthesis) locus. To date the cps gene clusters of the 77 serologically defined K-types and nine additional distinct cps operons have been identified and published. The biosynthesis pathway is a Wzy-dependent polymerization pathway, identical to E. coli Group 1 capsule synthesis.

14There is some confusion in the literature regarding the prevalence of K serotypes and their association with disease outcome. This may be explained by the higher diversity of the K serotypes and possible cross-reactivity, making them much harder to define. In some studies K1 and K2 showed the highest prevalence and were associated with poorer disease outcome. In addition, these K serotypes have been associated with the emergence of hypervirulent K. pneumoniae causing community-acquired invasive infections such as pyogenic liver abscess, which has become of particular concern in parts of Asia. Hypervirulent K1 strains generally belong to the ST23 lineage, while K2 is found in more diverse backgrounds. However, in many studies outside of Asia, including an extensive seroepidemiological study of European and North American cases, K1 isolates are rare.

15Until now our knowledge of the nature and diversity of O- and K- operons and, by proxy, the antigens they encode is based on sequences of a limited number of reference isolates. Recently, the whole-genome sequences of more than 500 isolates collected from environmental samples, plants, mammals and non-human primates, as well as those from asymptomatic human carriage, from cases of invasive disease and from both the clinical and community setting were sequenced and published (Table 1). These studies highlighted that the ability to cause invasive infections is not determined by lineage but is associated with the presence of virulence factors such as siderophore systems and the rmpA mucosity factor, although these virulence determinants are overrepresented in lineages associated with the hypervirulence phenotype.

16Using this comprehensive whole genome dataset we set out to describe the genetic diversity of O-antigen and K-antigen and LPS core biosynthesis gene clusters and classify them by molecular serotyping. Using these data we then set out to determine if there was an association between O-, K- or LPS core types and disease outcome, phylogenetic lineage or the presence of other known virulence-related genes.

17Publicly available genome data derived from four different K. pneumoniae collections were analyzed in this study, totaling 573 sequenced isolates (Table 1). The global dataset consists of a globally representative collection from six different countries including isolates from different hosts and different infection types. The UK hospital dataset consists of a collection from the Cambridge University Hospitals NHS Foundation Trust and contains invasive isolates biased towards those with resistance to third-generation cephalosporins collected over a period of seven years. The Nepal hospital dataset contains human isolates from a single Nepalese hospital outbreak in 2012. The preantibiotic dataset contains strains isolated before the widespread use of antibiotics (pre 1949).

18Genome data, generated in the above studies by paired end Illumina sequencing, were sourced from the European Nucleotide Archive (accession numbers are listed in Supplementary dataset S1, available in the online supplementary material). Reads were de novo assembled using Velvet and Velvet Optimiser, and the resulting assemblies annotated using Prokka, as described previously.

19O serotyping was performed on the basis of the polysaccharide ABC transporters (flippases; Wzm and Wzt). The specificity of transporters to polysaccharide types has been noted previously. The protein sequences of Wzm were located and extracted from the assembled and annotated contigs using tblastn and grouped based on sequence clustering using CD-HIT with an identity threshold of 95 %. This threshold was empirically derived and shown to be able to differentiate between the previously described rfb clusters (Table S1. The resulting groups were assigned the O serotype of known Wzm sequences (Table S1), which were included in the clustering step. In cases where Wzm was missing, Wzt homology was used for assignment. Any rfb loci which differed from previously described loci in their sequence and gene content were assigned to putative novel O serotypes. Novel rfb loci sequences were given the designation OL[n], etc, (OL=LPS locus, to differentiate those defined on the basis of rfb locus sequences from those defined serologically; n is a numeric identifier beginning from n=100) (Fig. S1, Table S2). The differentiation of O1 from O2 serotypes was based on the presence of WbbY and WbbZ in O1 (NCBI accession number KJ451390, strain NTUH-K2044). LPS core types were assigned as described above, using WaaL clustering with an identity threshold of 80 %. The waa gene cluster was extracted by locating WaaC and WabG using tblastn. Several operons contained a contig break or stretches of low sequencing coverage; when no waaL could be identified or did not fit into any of the two main groups, these were marked as unassigned.

20Due to the much higher diversity of the K antigen, grouping was performed based on the full-length cps sequence, which was extracted by locating wzi and wzc using tblastn and extending the presumed locus until a gene on the opposite strand appeared. The extracted cps locus DNA sequences were grouped using UCLUST based on a 95 % identity. The groups were assigned to a serotype if they matched to any of the 86 previously described cps locus sequences.

21Additionally, a database of wzi and wzc sequences with known serotypes was used to verify the classifications. Any cps clusters which differed from previously described loci in their sequence and gene content were assigned to putative novel K serotypes. Putative full-length novel clusters were confirmed by generation of alternative assemblies using SPAdes 3.6.1 and inspection of the resulting assembly graph using Bandage. Sequences were confirmed as full length if they spanned from the 5′ galF to the 3′ ugd gene. Full-length clusters were given the designation KL[n], etc: KL=capsule locus, to differentiate those defined on the basis of cps locus sequences from those defined serologically; n is a numeric identifier beginning from n=101 (although the first two loci (KL101 and KL102) have already been named KN1 and KN2 in the literature, and thus we will continue to use these designations here). Representative nucleotide sequences of full length cps and rfb clusters were annotated using Prokka followed by manual curation, and deposited in GenBank (Fig. S1, Table S2). Novel wzi alleles were added to the K. pneumoniae whole-genome sequence typing database BIGSdb at the Institut Pasteur (http://bigsdb.web.pasteur.fr/).

22To analyze the prevalence of different O and K serotypes in human disease and their putative association with disease outcome, the 216 human-associated isolates of the global collection were used to investigate their seroepidemiology.

23Three different infection properties were examined for association between O and K serotypes and LPS core variant: (i) infection site: we focused on isolates sampled from human blood (representing bacteraemia), sputum (pneumonia) and urine (urinal tract infections, UTI), isolates from other sites were excluded from this category, (ii) acquisition type: hospital-acquired isolates had been sampled in patients from 48 h after admission to hospital, community-acquired isolates had been sampled within 48 h of admission to hospital, isolates where no acquisition type was recorded were excluded from this category, (iii) infection status: carriage isolates are intestinal samples without any infection, non-invasive isolates are infections without bacteraemia (such as wound infection, pneumonia, UTI) and invasive isolates are from infections of normally sterile sites (such as blood).

24A database of putative virulence genes was compiled (Table S3) and the presence of these genes was characterized as described previously.

25MLST sequence types for all isolates were determined directly from sequence reads using SRST2 to type against the seven-locus MLST scheme. Sequence types (ST) are listed in Supplementary dataset S1.

26Comparative gene analysis and core genome definition was performed using Roary, using a blastp percentage identity of 95 % and a core definition of 99 %.

27Phylogenies were inferred from either a SNP alignment generated by mapping reads to the K. pneumoniae MGH78578 reference sequence (Fig. S2) or from a concatenated alignment of core genes extracted from Roary output (Fig. 2). Single-gene alignments (Fig. 1b and Fig. S3) were performed using Clustal Omega. Phylogenies were inferred from these alignments by running RAxML using a gamma distribution to model site-specific rate variation and 100 bootstrap replicates.

28The rfb gene cluster was located from the genome sequence data and classified on the basis of Wzm and Wzt protein sequence. The rfb cluster was universally located between the cps and his gene clusters in all of our samples (Fig. S1). It was evident from these data that there were rfb operons that did not match any of the known rfb gene clusters (i.e. they did not belong to O1, O2, O3, O4, O5, O8 or O12), either as a result of different gene content or differing nucleotide sequence. These clusters were defined as novel geno-/serotypes. Out of the 573 isolates analyzed in this study, 533 isolates could be assigned to six known O serotypes, 36 isolates to five novel serotypes, and four isolates remain unassigned due to contig breaks or missing sequencing coverage (Supplementary dataset S1).

29The clusters themselves range in size from seven to ten genes (8–10 kb). Both O1/O2 variants, O3, O4, O5 and O12, and four novel rfb clusters (OL101, OL102, OL103 and OL104) were observed in our dataset (Fig. 1a). The novel cluster, OL104, is identical in gene repertoire to O3 and O5, both synthesizing a mannose polymer repeated subunit. OL104 possesses the same WbdA as O3, that contains two mannosyltransferase domains and is distinct from the longer O5 WbdA, that contains three mannosyltransferase domains, which have been shown to have functional consequences in the mannosyl linkages they are able to synthesize. Within the isolates designated as serotype O3, two types of WbdD, the chain-length regulator, were observed. One (present in 34 of a total of 86 O3 isolates screened; 40 %) is identical to the E. coli O9a WbdD (Uniprot accession number Q47592), whereas the other type (present in the remaining 52 isolates; 60 %) has a deletion of 25 aa at the C-terminal region of its protein kinase domain. The WbdD of the isolates designated as serotype OL104 shows the same deletion (Fig. 1b). We speculate that OL104 and the two O3 WbdD variants might actually represent serologically distinct subtypes of the O3 serogroup.

30The K. pneumoniae O12 antigen is composed of an N-acetylglucosamine and rhamnose polymer. Based on their gene content and order, O12, OL101 and OL102 are also closely related rfb clusters differing by the presence and absence of single genes: Compared with the O12 cluster, OL101 includes an additional glycosylhydrolase located on the opposite strand, while OL102 lacks the terminal glycosyltransferase gene but possesses an additional rhamnan synthesis gene (Fig. 1a).

31It is of note that the direction of transcription in all of the rfb operons run antiparallel to those genes belonging to the adjoining his operon, with exception noted above, the O4 cluster and the extended variant of the O1/O2 O-antigen cluster (O1/O2 Variant 2) (Fig. 1a). The O4 rfb cluster is the only rfb cluster in K. pneumoniae containing a transposase. The O4 antigen is based on a galactose and ribofuranose polymer.

32When assessing the genetic diversity of the waa LPS core biosynthesis operon we detected both of the known core types, but found no conclusive evidence for novel core types.

33The cps gene cluster showed a strikingly higher level of diversity both in sequence and gene content, compared to the rfb clusters (Fig. S1). Clusters ranged in size from 20.5 to 21.6 kb.

34Of the 573 isolates analyzed, 387 (68 %) were assigned to 52 known K serotypes by comparison to previously described cps clusters representing the serologically typed reference strains (Supplementary dataset S1). Three isolates (0.5 %) where only a partial sequence was available were assigned to a serotype with lower confidence due to only a partial match to the reference sequence. A total of 165 isolates (29 %) for which the cps sequence was >5 % different to those of any of the known cps clusters were putatively assigned to 68 novel serotypes. Eighteen isolates (3 %) remained unassigned because no wzi and wzc sequences could be identified in the genomes. Among the putative novel serotypes, 21 distinct full-length cps sequences were identified (Fig. S1). Five of these sequences matched or were transposase variants of those previously described in and, including one which was a transposase-negative variant of the cps cluster from K. pneumoniae HS11286 (genome accession NC_016845.1).

35Consistent with results from previous studies, the first gene in all cps clusters was galF followed by a putative glucose phosphatase (cpsACP), the JUMPstart sequence (‘just upstream of many polysaccharide starts’) followed by the translocation and surface assembly genes (wzi, wza, wzb and wzc). Although the genes responsible for transport into the periplasm and polymerization (wzx and wzy), UndPP-linkage (wbaP or wcaJ) and a 6-phosphogluconate dehydrogenase [gnd, a housekeeping gene not required for CPS synthesis] were universally present, they were not found at fixed locations in these clusters. The cps clusters are usually terminated by UDP-glucose 6-dehydrogenase gene (ugd). The gene encoding glucose dehydratase (rffG) separates the cps and rfb clusters.

36To understand the distribution of the rfb, cps and waa gene clusters across the K. pneumoniae species we reconstructed a whole-genome-based phylogeny for all isolates included in this study and show their distribution (Fig. 2). This highlights that, with some exceptions, rfb clusters are not restricted to any particular clade. In our dataset the O1 and O2 rfb loci are only found in the KpI sublineage (K. pneumoniae sensu stricto), for the other rfb clusters there is extensive evidence of horizontal gene transfer of the different rfb clusters between closely and more distantly related lineages. This is further confirmed by the non-concordance of the phylogeny of the rfb gene clusters and the whole-genome phylogeny [O1 and O2: Fig. S2(a); O3, O5 and Novel 4: Fig. S2(b)].

37The two waa LPS core types are equally not restricted to particular clades in the phylogeny, although we note a biased distribution with respect to the O serotypes. Out of the 573 isolates in the full dataset, 70 isolates (12 %) encoded core type 2; these core 2 strains are almost exclusively (65 isolates, 93 %) associated with serotype O1. The remaining isolates are core 1 (446, 78 %) or unassigned (57, 10 %). One previous study investigated the distribution of LPS core types in 100 K. pneumoniae strains based on PCR and dot blots and found a ratio of LPS core 1 to LPS core 2 isolates similar to our data (79: 19), and also noted the association of LPS core 2 with serotype O1 (10 out of 34 O1 strains); however we did not observe an association with K serotype K2 as has been previously reported. Our analysis shows an elevated number of strains carrying both the LPS core type 2 and K2 or K45, however this is likely to be strongly influenced by two clonal expansions within the sample collection we analysed (Fig. 2). Furthermore, we could only observe LPS core type 2 in subspecies KpI (Fig. 2).

38Due to the high diversity, K serotype switching is more difficult to analyze. As an example, K1 isolates belong mainly to two lineages, one of which (ST23) is associated with the hypervirulent phenotype, whereas the K2 serotype is much more distributed across the phylogenetic tree, including lineages associated with the hypervirulent phenotype (e.g. ST25) and other non-hypervirulent lineages (e.g. ST14, Fig. 2). This is consistent with previous reports which showed that many K. pneumoniae sequence types include multiple CPS types.

39Looking across the full dataset of the 573 isolates the majority were genotyped as O1 (296 isolates, 52 %), followed by O2 (91, 16 %), O3 (86, 15 %), O5 (33, 6 %), OL101 (26, 5 %), O4 (18, 3 %) and O12 (9, 2 %). Each of the OL102 to OL104 serotypes occur in less than 1 % of isolates. The O1/O2 extended rfb cluster variant 2 was observed in 112 (38 %) of O1 genotyped isolates and 52 (57 %) O2 isolates (Supplementary dataset S1).

40The seven most common O serotypes described above were associated with 54, 39, 40, 22, 14, 7 and 9 distinct K serotypes respectively, suggesting there is a relationship between the number of representatives of each O serotype in our dataset and the number of distinct K serotypes with which each is associated. This relationship is best explored using only the global isolate set, which is not biased towards any particular K. pneumoniae lineage or isolate source. Within this dataset O1 is the most common O serotype (n=133) and was associated with 43 distinct K serotypes. O3 is the next most common O serotype (n=45), followed by O2 (n=39) and O5 (n=24), these were associated with 29, 30 and 20 distinct K serotypes, respectively. Among the same isolate set the most common K serotype was K2 (n=20), associated with two distinct O serotypes (O1 and O2). K1 (n=15) and K64 (n=14) are the next most common serotypes and are associated with two and four O serotypes respectively. All other K serotypes are represented by eight or fewer isolates in our collection and were each associated with just one or two distinct O serotypes. These data indicate that re-assortment of K and O types occurs frequently in K. pneumoniae.

41To investigate the seroepidemiology in human disease, we analyzed the association of O and K serotypes with the three different recorded infection properties (infection site, acquisition type and infection status), using the human-associated isolates of the global collection (Table 2, Fig. 3). Fisher’s exact test was performed to examine whether there was a significant association between serotypes or LPS core types and infection sites, acquisition type or infection status. Significant correlations were found between K2 and invasive infections (odds ratio 4.1, p=0.008), and between OL101 O type and asymptomatic carriage (OR 8.5, p=0.009) (Table 2). Of special interest is also K1, which was virtually restricted to community-acquired isolates (OR 8.4, p=0.016) and completely missing in carriage-associated isolates (p=0.024). No correlation was found for any association of either O or K serotype to infection sites and no correlation of LPS core type with any of the traits as above was observed.

42To assess any correlation of serotypes to geographical origins, the dataset was grouped based on acquisition type and infection site in order to remove any spurious relationship. The groups were analyzed as described above; no statistically significant correlation between serotypes or LPS core types and geographical origin could be found. No significant correlation of the presence of O1/O2 rfb cluster extended variant 2 to infection site, infection type or sample site was found.

43Multiple virulence factors have been defined for Klebsiella including three siderophores (yersiniabactin, salmochelin and aerobactin), the cytotoxin colibactin and rmpA/rmpA2, a regulator of CPS overexpression. Fimbrial adhesins are also known to play a major role in biofilm formation and are thus also classified as virulence factors. We determined if the distribution of these known virulence functions was correlated with O- or K-serotype using the existing global data set. The most noteable association was the significant enrichment of an array of siderophore and colibactin clusters and the mucosity regulator rmpA with K serotypes K1 and K2 (Fig. 4a). Since K1 and K2 isolates were almost exclusively also O1 serotypes, the same effect was observed in O1. These results are perhaps unsurprising given that these virulence genes are overrepresented in the hypervirulent K. pneumoniae lineages, which are also associated with serotypes K1 and K2.

44Two types of fimbriae are known to be present in K. pneumoniae: type 1 and type 3 fimbriae. Whilst type 3 fimbriae are known to bind to different human cell types, ranging from tracheal epithelia to renal tubular cells, as well as being important for binding to plants and abiotic surfaces, type 1 fimbriae appear more specific, so far they have been shown to be important for adhesion to uroepithelial cells and also for binding to plants. Both type 1 and type 3 fimbriae are also associated with biofilm formation in K. pneumoniae (type 3 fimbriae more strongly than type 1. Our data showed that serotype O1 isolates had a significantly higher likelihood of possessing type 1 fimbriae compared with any other O serotype (odds ratio 9.7, p=1.1E-16×E-16E-1610–16, Fisher’s exact test) (Fig. 4b). Differences in type 1 and type 3 fimbriae occurrence are very pronounced between K1 and K2: whereas K1 lacks type 1 fimbriae genes (none present, p=0.0019), K2 has a significantly higher occurrence compared with any other K serotype (OR 15.8, p=3.82×10–13); type 3 fimbriae genes are rarer in K1 (OR 0.03, p=6.43×10–17) compared with any other K serotype. The majority of K1 isolates possessing neither type 1 nor type 3 fimbriae (20 isolates without type 1 or type 3 fimbriae out of a total of 35 K1 isolates), stem from a single sequence type (ST82) present in the preantibiotic collection.

45To further investigate the fact that K1 occurs almost exclusively in community-acquired isolates and is absent in nosocomial infections, a comparative genomic analysis was performed to search for genes that are overrepresented in K1 compared with other K-serotypes. Strikingly, an alginate lyase isozyme was found to be present in all K1 isolates and virtually absent in all other K serotypes (p=4.82×10–57, Fisher’s exact test). This K1-associated alginate lyase is located just upstream of the thiamine pyrophosphate (TPP) biosynthesis cluster (Fig. 4c). The only other commonly occurring alginate lyase gene in the dataset was present within the K14 cps cluster and has a sequence identity to the K1 associated alginate lyase below 45 %. Alginate lyases enable the use of alginate as carbon and energy source.

46The rfb and cps gene clusters give rise to the dominant serotypical properties of K. pneumoniae and are therefore priority vaccine candidates for the treatment of this increasingly multidrug-resistant pathogen. We have catalogued the naturally occurring diversity of these gene clusters within a large collection of isolates taken from different geographic and clinical settings, hosts and disease manifestations. In doing so we identified six known and five novel O-antigen clusters and 45 known and 18 novel complete K-antigen biosynthesis gene clusters. Of particular interest is a previously undescribed rfb cluster (OL101), occurring in 5 % of isolates in our dataset, originating in Europe, Asia and North America and found in human, bovine and environmental samples. We also identified a further 46 putative cps clusters (for which only partial sequences could be obtained) which appear novel. Although this may be an overestimate as some of these may represent transposase variants or divergent forms of other cps types or result from assembly errors; additional work will be required to investigate and validate these further.

47This genetic catalogue of rfb and cps clusters enabled in silico serotyping of whole genome sequences. We focused on the seroepidemiology by analyzing the human-associated isolates from the global dataset, where clinical parameters such as infection site, type and acquisition mode have been recorded. We showed that O serotypes O1, O2 and O3 accounted for 80 % or more of all samples included in this study and that the relative prevalence of these O serotypes was approximately the same for all infection sites, infection types and acquisition modes. This contrasted with the K serotypes, for which an order of magnitude higher diversity was found and no single K serotype dominated this collection.

48Notable in the K serotype analysis was the distribution of serotype K1 sequences: most K1 isolates in our study belong to one of two lineages, ST23 and ST82 and showed a strong association with community-acquired infections, consistent with previous reports and the known association of ST23 with the hypervirulent phenotype. K1 isolates were found in six out of seven countries covered by our genome collection. However, despite the fact that K1 is regarded as one of the two most common serotypes it was only found once in 162 isolates collected over a period of seven years in a UK hospital (this study) and is absent in another study based on 703 isolates in 13 hospitals located in Western Europe and Northern America. This contrasts with reports from Taiwan, China and South Africa where K1 appears to have been dominant for a considerable period of time and has been associated with comparatively higher prevalence of hypervirulent infections in these countries, mostly linked to the ST23 lineage.

49When looking for genes closely associated with K1 we identified the presence of a gene predicted to encode an alginate lyase isozyme. Outside of K1 the only other alginate lyase detected is distantly related and part of the K14 cps cluster. Hence, this gene is almost exclusively found in isolates of the K1 serotype (including those of both major K1 lineages). Pseudomonas aeruginosa is known to produce alginate during biofilm biogenesis in chronic lung infections of cystic fibrosis patients. Alginate lyase enables cell detachment from the biofilm. However it is not known whether K. pneumoniae produces alginate biofilms. Moreover, since it is unusual to find alginate-lyase-producers, like the K1 K. pneumoniae, that do not use alginate as primary carbon source, it has been proposed that alginate lyase production could be related to coinfection of P. aeruginosa and K. pneumoniae in cystic fibrosis patients. Given the above it is suggested that the association with cystic fibrosis is important, but probably it is opportunistic, and its role in K. pneumoniae is much broader.

50We found no association of LPS core type with the isolates' infection sites, infection types and acquisition modes. This is in contrast to a study showing that LPS core type 2 contributes to the level of virulence in K. pneumoniae, although its mechanism is so far unknown.

51Virulence genes (siderophores, colibactin, rmpA and fimbriae) are expected to be overrepresented in isolates from infections compared with those from carriage or environmental sources and are particularly overrepresented in lineages associated with hypervirulent invasive disease, thus the significantly high presence in K. pneumoniae serotype O1 isolates was anticipated. However, their significantly lower abundance in isolates of serotype O2 or O3, which are both also prevalent in human infection, was more surprising.

52The striking difference in the distribution of the two major fimbriae types found in K. pneumoniae is also of note. Type 3 fimbriae are essential for biofilm formation, while type 1 fimbriae are important for adhesion to uroepithelial cells and are thus considered virulence factors for urinary tract infection. The differential distribution of fimbriae in K. pneumoniae probably allows isolates to adhere to different receptors and perhaps exploit or specialize in different niches, the presence of type 1 fimbriae in K2 isolates and the notable absence of type 1 fimbriae in K1 isolates is intriguing but will require more targeted sampling to unravel its true biological significance.

53It is suspected that certain CPS types are able to influence the accessibility of the LPS O-antigen, possibly masking it; the reports are however inconclusive: described the masking of the O1 antigen by K1 capsule but not by K2, whereas showed that the O1 antigen is accessible by antibodies irrespective of the capsule type. The potential masking of O-antigen by the capsule warrants further research and is the crucial next step towards a polysaccharide-based vaccine against K. pneumoniae infection.

54Our data indicate that O and K serotypes are frequently re-assorted within the K. pneumoniae population. There are far fewer distinct O serotypes than K serotypes but among the O serotypes, those that are most common are generally associated with a higher number of distinct K serotypes. While there are lineage-specific clustering effects (Fig. 2), these data support the hypothesis that the rfb and/or cps clusters are shuffled within the K. pneumoniae population via horizontal gene transfer. However given the wide diversity of K serotypes, much larger strain collections will need to be examined in order to detect a statistically significant divergence from random re-assortment of CPS and LPS loci.

55K. pneumoniae is considered to be a significant threat to human health with the rates of infection increasing globally and appearing to be driven by increasing levels of antimicrobial resistance to front line antimicrobials. The short-term solution has been to turn to old drugs such as colistin that are associated with significant nephrotoxicity. However, the development of resistance to colistin is rapid and explained by both intrinsic mechanisms, such as point mutations, as well as by the acquisition of genes by lateral gene transfer. With this in mind we found that O-serotype prevalence and distribution were stable with regards to different infection types and sites. In addition although O serotype switching has been a common occurrence in the evolution of K. pneumoniae K serotypes show a much bigger variance. Consequently O antigens, defined in detail in this study, offer a promising target for vaccine design that warrant further research.