The Capsule Regulatory Network of Klebsiella pneumoniae Defined by density-TraDISort

1Klebsiella pneumoniae is a ubiquitous Gram-negative bacterium, found both in the environment and as an asymptomatic coloniser of the mucosal surfaces of mammals. K. pneumoniae is also an opportunistic pathogen and can express a multitude of virulence factors which enable it to cause infections in humans. Historically associated with infections in the immunocompromised and in neonates, focus has been directed on K. pneumoniae following the emergence of antimicrobial-resistant and hypervirulent lineages. Hypervirulent lineages are a particular concern in a clinical setting, because they have the potential to cause infection in immunocompetent hosts, to metastasise, and to cause infections in unusual infection sites.

2Numerous factors contribute to K. pneumoniae virulence, such as the ability to produce siderophores, fimbriae, lipopolysaccharide (LPS), and extracellular polysaccharide capsule. Hypervirulence is associated with these factors, particularly with the overproduction of capsular polysaccharide, and in the absence of these virulence factors, K. pneumoniae virulence is reduced or abolished. The ∼200-kb K. pneumoniae virulence plasmid, which is also associated with the hypervirulent phenotype, encodes siderophores such as aerobactin and salmochelin and positive regulators of capsule biosynthesis. More than 100 capsule locus types have been identified in K. pneumoniae, though the majority of hypervirulent K. pneumoniae isolates represent strains of capsule types K1 and K2.

3Excessive capsule production is strongly associated with hypervirulence in K. pneumoniae, and several studies have sought to identify genetic determinants of hypervirulence. For example, the mucoviscosity-associated gene magA (now named wzy_K1) was originally identified by transposon mutagenesis screening. The rmpA and rmpA2 genes also affect hypermucoviscosity and encode transcription factors that positively regulate the K. pneumoniae capsule biosynthesis locus. These regulators can be either chromosomally encoded or plasmid-borne. Although rmpA is correlated with hypervirulence and strains lacking rmpA and aerobactin are avirulent in mice, it has been shown that this increased virulence is a consequence of the hypermucoviscous phenotype conferred by rmpA rather than a consequence of the presence of the gene itself.

4Capsule is also a potential therapeutic target in K. pneumoniae. Capsule-targeting monoclonal antibodies increased the killing of K. pneumoniae ST258 (an outbreak lineage) by human serum and neutrophils and limited the spread of a respiratory K. pneumoniae ST258 infection in mice. Specific capsule-targeting bacteriophage have been shown to clear or limit infections caused by K. pneumoniae strains of capsule types K1, K5, and K64, and in some cases protection could also be achieved by treatment with the capsule depolymerase enzymes produced by these phage, rather than the phage itself. Translating these early-stage findings to the clinic is a priority as extensively drug-resistant K. pneumoniae strains become more prevalent.

5Despite the absolute requirement for capsule in Klebsiella infections, its promise as a therapeutic target and the connection between capsule overproduction and hypervirulence and the biosynthetic and regulatory mechanisms governing K. pneumoniae capsule have not yet been fully explored. Genetic screens and reverse genetics, including transposon mutagenesis approaches, have been used to identify biosynthetic genes and activators of capsule production (including the rmpA and magA genes), and some of the cues that elevate capsule expression above basal levels have also been described—these include temperature, iron availability, and the presence of certain carbon sources. Defining the capsule regulatory network of Klebsiella pneumoniae in more detail not only would deepen our understanding of this pathogen’s interaction with the host environment but also could inform efforts to target this factor with new therapeutics.

6Here we employed transposon-directed insertion sequencing (TraDIS) to identify genes influencing K. pneumoniae capsule production. We performed density-based selection on mutant libraries of both K1 and K2 capsule type K. pneumoniae, using a discontinuous density gradient. This approach allowed the simultaneous selection of both capsulated and noncapsulated mutants from mutant libraries without requiring growth of bacteria under selective pressure. We have identified 78 genes which, when mutagenized by transposon insertion, reduce the ability of one or both of these strains to manufacture capsule. These included multiple genes not previously associated with capsule production in K. pneumoniae. We have also identified 26 candidate genes in NTUH-K2044 which cause an increase in capsulation when disrupted. Our results allow us to present an integrated model for capsule regulation in K. pneumoniae and establish a technology for study of capsule production that is applicable to other bacterial species.

7K. pneumoniae capsule influences the centrifugation period required to pellet cells, and this property has been used to semiquantitatively compare capsule amounts between different strains. On the basis of this observation, we speculated that Klebsiella cells of different capsulation states could be separated by density-based centrifugation and that this method could be combined with TraDIS to screen for capsule-regulating genes. Discontinuous Percoll density gradients are routinely used to purify macrophages from complex samples and have also been used to examine capsulation in Bacteroides fragilis and Porphyromonas gingivalis. Tests with K. pneumoniae NTUH-K2044 (capsule type K1, hypermucoid) (Fig. 1A) and K. pneumoniae ATCC 43816 (K2, hypermucoid) and a noncapsulated Escherichia coli control showed that these strains differed in density and migrated to above 15%, above 35% and below 50% Percoll, respectively (see Fig. S1A in the supplemental material). Growth of the Klebsiella strains at 25°C (which reduces capsule production) decreased their density, with NTUH-K2044 localizing to above 35% Percoll and ATCC 43816 showing limited migration into the 35% layer. A third K. pneumoniae strain, RH201207 (nonhypermucoid, capsule type K106), was retained above 50% Percoll (Fig. S1A). We also examined the proportion of Klebsiella bacteria migrating to different locations within the density gradient (Fig. S1B). While the majority of K. pneumoniae NTUH-2044 bacteria were located above the 15% layer, a small fraction (4% viable count of the top) migrated to below the interface of this layer, suggesting some heterogeneity in capsule production. In both NTUH-K2044 and ATCC 43816, only a very small number of cells were present below the 50% layer (Fig. S1B); these numbers could be partially due to cross-contamination occurring during extraction of this part of the gradient, which was recovered by pipetting from the top.

8We then tested this method with a high-density transposon insertion library of K. pneumoniae NTUH-K2044 (generated as described in Materials and Methods). This mutant library gave rise to three fractions on a 15% to 35% to 50% Percoll gradient (Fig. 1B). The middle and bottom fractions separated from the NTUH-K2044 library contained less capsule than the input library culture, as shown by quantification of capsular uronic acids (Fig. 1C) and the hypermucoidy centrifugation test (Fig. S1C). To determine whether altered migration in a Percoll gradient was the result of stable mutant phenotypes, the fractions of the K. pneumoniae NTUH-K2044 gradient were cultured overnight and centrifuged on a fresh gradient. The “bottom” and “top” fractions migrated to the same position as before, while the “middle” fraction showed a partially heritable phenotype and was distributed across the top and middle positions (Fig. S1D).

9These data indicated that centrifugation on a discontinuous density gradient successfully separated K. pneumoniae populations on the basis of their capsule production and that variations in density among library mutants were, in part, due to stable phenotypes. Random-prime PCR analysis of several colonies from the bottom fraction of the K. pneumoniae NTUH-K2044 library identified insertions in known capsule biosynthetic or regulatory genes (see Text S1 in the supplemental material), and a wza insertion mutant was retained as a capsule-negative control for future experiments. We then wished to determine whether our method could be used for capsule-based separations of other species, for which we tested capsulated and noncapsulated Streptococcus pneumoniae 23F. These strains were separated reproducibly on a Percoll gradient (Fig. S1E), indicating that our method is applicable to other bacterial species.

10TraDISort is a term for transposon sequencing screens that employ physical, rather than survival-based, enrichment of insertion mutants from saturated libraries, which allows examination of phenotypes that are not linked to survival. We call our approach “density-TraDISort” to distinguish it from the original approach using fluorescence-based flow sorting as the physical selection method. The final density-TraDISort approach is shown (Fig. 1B). Briefly, transposon mutant libraries were grown overnight at 37°C in LB, applied to the top of a Percoll gradient, and centrifuged at a moderate speed for 30 min (see Materials and Methods). Following centrifugation, each bacterial fraction was extracted and subjected to TraDIS, as was a sample of the input culture. Data were analyzed using the Bio-TraDIS pipeline (see Materials and Methods). The K. pneumoniae NTUH-K2044 library contained approximately 120,000 unique transposon insertion sites (equivalent to an insertion every 45 bp), with a median of 14 insertion sites per gene. Statistics summarizing the sequencing results from each fraction are reported in Table S2 in the supplemental material.

11Unlike traditional growth-based transposon insertion screens, our density-TraDISort setup combines positive and negative selection within a single experiment; mutants with reduced capsule can be identified through their loss from the top fraction or by virtue of their enrichment in another fraction (see Fig. 2A for an example). We applied stringent cutoffs on the basis of both selections to identify putative capsule-related genes. Briefly, a gene was counted as a hit only if it was (i) lost from the top fraction (log2 fold change [log2FC] < −1; false-discovery-rate [q] value < 0.001) and (ii) enriched in another fraction (log2FC > 1; q value < 0.001) (see Table S3 and S4). The presence of a “middle” fraction also enabled us to identify genes which increased capsulation when disrupted (an example is shown in Fig. 2B). These were apparent in the raw TraDIS plot files as genes where all or nearly all the mutants localized to the top fraction following centrifugation (mutants with wild-type capsule were distributed between the top and middle fractions, in keeping with the migration pattern of wild-type NTUH-K2044) (Fig. S1A). We defined putative “capsule up” genes as those which showed a marked depletion in the middle fraction relative to the input (log2FC < −3; q value < 0.001), with genes with very low initial read counts (log2cpm < 4) excluded (Table 1; see also Table S4). Examples of TraDISort data for several genes known to affect capsule production are shown in Fig. S2C.

12In total, we identified 62 genes required for full capsule production in K. pneumoniae NTUH-K2044 (Table 1; see also Table S4). We also identified 26 putative capsule up mutants in this strain (Table 1; see also Table S4). The biological roles of the genes determined to influence capsule production are discussed in more detail below.

13Our “capsule down” mutants included almost every gene of the capsule biosynthesis locus of K. pneumoniae NTUH-K2044 (Fig. 2A), further confirming that our experimental strategy had successfully isolated capsule-deficient mutants. The manB and ugd genes were not called as hits, however, these contained very few insertions in the input sample (visible in the plot files above these genes in Fig. 2A). In addition to biosynthetic genes, the known capsule regulators rmpA, rcsB, and rfaH were identified (Fig. S2C).

14Other capsule down hits included multiple cell surface components, metabolic genes, and genes of global regulatory systems. Extended functional information corresponding to all of our hits, including reported links to capsule production in Klebsiella and other proteobacteria, is included in Table S4. Gene enrichment analysis using the TopGO package indicated that genes for polysaccharide metabolic processes (P = 1.7 × 10−5), molecular transducers (P = 0.0042), and protein kinases (P = 0.005) were overrepresented in our hits (all probabilities are from Fisher’s exact test, parent-child algorithm).

15Several global regulators were identified as affecting capsule production. These included the ArcB anaerobic/redox-sensitive sensor kinase, the OmpR-EnvZ osmotic stress response system, the BarA-UvrY system, the ArgR arginine repressor, and the CsrB carbon storage regulatory small RNA. Transcription regulators MprA and SlyA were also needed for full capsule production. Metabolic genes, including genes coding for the electron transport chain, glycolytic enzymes, and the GlnB nitrogen source regulatory protein, were also identified.

16The capsule up mutants of NTUH-K2044 included nucleoid proteins, membrane-bound transporters such as SapBCDF (Fig. 2B), and several metabolic gene products (Table S4). Interestingly, some of the capsule up hits, including CsrD (targets CsrB for degradation), H-NS (reported to be antagonized by SlyA), and GlnD (modifies and controls activity of GlnB), are known antagonists or regulators of genes in the capsule down hits.

17Finally, our results also included many genes for other cell surface components; genes for the synthesis and modification of LPS were identified, along with enterobacterial common antigen (ECA) genes and the Lpp lipoprotein gene.

18We wished to determine the extent of conservation of our capsule-regulatory hits across the Klebsiella species, and the overlap of the capsule-influencing genes in different Klebsiella strains. To explore this issue, we applied our density-TraDISort method to a second strain, K. pneumoniae ATCC 43816, a capsule type K2 strain which is commonly used in Klebsiella infection studies (Fig. S3A). We constructed a K. pneumoniae ATCC 43816 saturated transposon insertion library with ∼250,000 unique insertion sites (or an insertion every 22 bp) and a median of 36 insertion sites per gene (see Materials and Methods) (Fig. S3C and D). This library resolved into two fractions on a 35% to 50% Percoll gradient, with no obvious heritable “middle” fraction, and fractions were collected and their transposon insertion junctions sequenced as described for K. pneumoniae NTUH-K2044. Genes were counted as hits in this strain if they were lost from the top fraction relative to the input (log2FC < −1; q value < 0.001) and enriched in the bottom fraction (log2FC > 1; q value < 0.001). The lack of a “middle” fraction in this experiment meant that there was less sensitivity for identifying reduced-capsule mutants, particularly using the second, positive-selection-based filtering criterion. We used a pan-genome generated from 263 annotated K. pneumoniae genomes (Table S5) to define the common genes in NTUH-K2044 and ATCC 43816 and to determine their prevalence across the K. pneumoniae population (Table 1; see also Table S4).

19We identified 34 candidate capsule-regulatory genes in K. pneumoniae ATCC 43816. As observed for NTUH-K2044, the genes of the capsule biosynthetic locus were nearly all called as hits (Fig. S3E). Of the three genes that were not, two had very low initial insertion counts, and the third met our first selection criterion of being lost from the top fraction but was not enriched in the bottom fraction. Putative capsule-influencing genes of ATCC 43816 fell into diverse functional categories, with genes encoding cell surface components, metabolic genes, and genes associated with transporters and known regulators implicated.

20Although the majority of capsule hits are encoded in the core genome (48 of 62 in K. pneumoniae NTUH-K2044; 23 of 32 in K. pneumoniae ATCC 43816; Fig. 2C), only 16 genes were called as hits in both strains. These shared genes included five shared components of the two strains’ capsule biosynthesis loci, the transcription antiterminator rfaH, two enterobacterial common antigen genes, three genes of the arn operon responsible for modification of LPS lipid A with l-Ara4N, and the glutathione reductase gor gene. Strain-specific differences were also identified, with the caveat that NTUH-K2044 produces more capsule (Fig. S3B) and therefore afforded us greater sensitivity in our experiment. Genes that were identified as required for ATCC 43816 capsule production but not NTUH-K2044 capsule production included the yrbCDEF (mlaCDEF) ABC transporter genes and the mlaA gene, which are involved in maintaining outer membrane asymmetry through the cycling of phospholipids. Thirty-two genes common to both strains had a capsule down phenotype in NTUH-K2044 but were not called as hits in ATCC 43816 (note that capsule up hits are not included in our comparison as these were not resolved in ATCC 43816). These included electron transport pump rnf, global regulators such as arcB and ompR, transcription factor mprA, and several additional cell surface component biosynthetic genes. It appears that the influence of at least some conserved genes on capsule is strain specific.

21We generated a set of 10 single-gene-deletion mutants in K. pneumoniae NTUH-K2044, and a set of 3 in K. pneumoniae ATCC 43816, in order to validate the results of our density-TraDISort screen. Genes selected for mutagenesis were the known capsule and LPS regulator gene rfaH in both strains, the LPS O-antigen ligase gene waaL, the aerobic respiration control sensor gene arcB, and multiple transcriptional regulator genes (ompR, argR, slyA, mprA, and uvrY). We also deleted the arnF gene in ATCC 43816. The sapBCDF ABC transporter in NTUH-K2044 was examined in order to validate our assignment of capsule up hits.

22Single-gene-knockout mutants were grown under the same conditions as in the original screen and were subjected to density gradient centrifugation. Every mutant showed a banding pattern consistent with the results of density-TraDISort (Fig. 3A and B). The ΔarcB mutant appeared to have two populations, one with wild-type capsule and one with reduced capsule. Most of the other NTUH-K2044 mutants migrated to the position of the middle fraction, while the ΔwaaL, i-wza, and ΔrfaH mutants migrated to the bottom of the gradient. The ΔsapBCDF mutant showed higher density than the wild-type strain and remained above the 15% Percoll layer, with no movement into the gradient itself. Mutants of NTUH-K2044 were also tested for hypermucoidy and uronic acid production, and these experiments showed increased capsule in NTUH-K2044 ΔsapBCDF and reduced capsule in all other mutants (Fig. 3C; see also Fig. S4B). Of the K. pneumoniae ATCC 43816 mutants (Fig. 3B), the ΔrfaH mutant migrated to the bottom of the gradient and the ΔarnF mutant to the middle and bottom, while mutant ΔmprA showed the same pattern as the wild type. This is consistent with the TraDISort assignment of this gene as having a strain-specific effect on capsule, at least under these growth conditions (Table 1).

23Our hits included several genes with roles in LPS biosynthesis, which raised the possibility that LPS O-antigen may affect cell density independently of capsule. To test this possibility, we constructed a ΔwaaL deletion in our wza transposon insertion strain (see Materials and Methods). The resulting double mutant had no detectable reduction in density compared to the wza mutant on the standard 15% to 35% to 50% gradient or on 70% Percoll (representing the minimum concentration required to exclude the wza mutant; Fig. S4A), indicating that the LPS O-antigen alone does not affect the density of K. pneumoniae, at least within the resolution range of this experiment.

24We selected the transcription factors ArgR, SlyA, and MprA, along with the ABC transporter SapBCDF, for further characterization. ArgR represses arginine synthesis and transport as well as expression of other genes, SlyA is an antagonist of H-NS (known to suppress capsule in K. pneumoniae), and MprA is a transcriptional regulator with an effect on capsule in uropathogenic E. coli (UPEC). Both SlyA and MprA were also shown very recently to be virulence and capsule regulators in K. pneumoniae and were renamed KvrA and KvrB. SapBCDF has been reported to mediate resistance to antimicrobial peptides (AMPs) in H. influenzae by importing them for degradation and was presumed to have this activity in Enterobacteriaceae as well, though it has recently been reported that this pump functions as a putrescine exporter in E. coli and has no role in AMP resistance. ArgR and SapBCDF have not previously been linked to capsule regulation.

25To confirm that the alterations in capsule production observed in the NTUH-K2044 ΔargR, ΔslyA, ΔmprA, and ΔsapBCDF mutants were due to the deleted genes, each mutant was complemented by reintroducing the wild-type gene on the chromosome (see Materials and Methods). Although this complementation strategy ensures wild-type levels of expression, it cannot rule out polar effects. Complementation caused a complete restoration of wild-type capsule production, as measured by the hypermucoidy and uronic acid assays (Fig. 4A and B). To define changes in capsule architecture, each mutant strain was examined by transmission electron microscopy (TEM) (Fig. 4C). Wild-type K. pneumoniae NTUH-K2044 had a thick, filamentous capsule of roughly half the cell diameter. The ΔargR and ΔslyA mutants had capsules with slightly reduced thickness and finer filaments, while the ΔmprA mutant had extremely fine and diffuse filaments such that the boundary of the capsule was not clear. The ΔsapBCDF capsule had some thick filaments but at lower density than NTUH-K2044, with an additional gel-like layer visible outside these filaments. The virulence of the ΔargR, ΔslyA, ΔmprA, and ΔsapBCDF mutants, and their complements, was assessed by infection of research-grade Galleria mellonella larvae, an established invertebrate model for Klebsiella infections. Each of the reduced-capsule strains showed a virulence defect relative to the wild-type strain which was restored on complementation (Fig. 5A), while the ΔsapBCDF mutant did not have changed virulence compared to the wild type.

26We then examined the effect of ArgR, MprA, SlyA, and SapBCDF on resistance to human serum. After 2 h, the NTUH-K2044 wild type showed full survival with a slight increase in viable count, while the wza mutant was reduced in viable count by ∼25-fold. The ΔargR, ΔmprA, and ΔslyA mutants did not change significantly (Fig. 5B). The ΔsapBCDF mutant showed increased viable count compared to the wild-type strain, with a 7-fold increase over the course of the experiment. This increase was unexpected, as the wild-type strain is already fully serum resistant. A double ΔsapBCDF cps mutant was constructed (see Materials and Methods) and showed a drastic reduction in survival, suggesting that the increased serum survival of the NTUH-K2044 ΔsapBCDF mutant is capsule dependent (Fig. 5B).

27We also tested the ΔsapBCDF mutant for resistance to the peptide antibiotics colistin and polymyxin B. The drug MICs were approximately the same as that seen with the wild type, at 1 µg/ml for colistin and 0.75 µg/ml for polymyxin B, indicating that the Sap transporter in K. pneumoniae does not contribute to antimicrobial peptide resistance.

28We then wished to determine whether mutation of the Sap transporter increased capsule production by acting on transcription. RNA was extracted from late-exponential-phase wild-type and ΔsapBCDF cells, and the abundance of three capsule locus transcripts—manC, wcaG, and wza—was measured by reverse transcription real-time quantitative PCR (qRT-PCR). These genes are transcribed from separate promoters. The ΔsapBCDF mutant showed elevated expression of wcaG, at 2.5 times wild-type levels, while expression of wza and manC was not significantly changed (Fig. 5C).

29The Rcs phosphorelay system regulates capsule expression in E. coli and other enterobacteriaciae and is induced by cues such as membrane stress. RcsA is a component of the system which autoregulates and increases its own expression when activated. We measured rcsA transcript levels to determine whether loss of the ΔsapBCDF genes induces rcsA (Fig. 5C). Unexpectedly, levels of rcsA were much lower in the mutant than in the wild type, indicating that Sap-dependent induction of capsule expression does not occur through rcsA. Note, however, that RcsA is not required for all permutations of Rcs signaling as RcsB can interact with a number of partner proteins to regulate transcription.

30We identified numerous putative capsule regulators by density-TraDISort and validated the results of our screen with single-gene-deletion mutants. We propose an integrated model for how the genes we identified may collectively control K. pneumoniae NTUH-K2044 capsule. This model is based on our results and previous published work in K. pneumoniae and other enterobacteria (particularly Escherichia coli). Full details of the literature relevant to each hit are listed in Table S4.

31Major nodes for transcriptional control are the CsrB carbon source utilization system and the Rcs phosphorelay system. Each of these systems is itself regulated by multiple genes identified in our study—CsrB integrates signals from the UvrY-BarA two-component system (a capsule down hit) and various carbon metabolic genes, is activated by DksA, and is targeted for degradation by CsrD (a capsule up hit); the Rcs system is induced by MdoGH mutation, can cooperate with RmpA and RmpA2 to induce capsule, and also responds to carbon metabolism and some forms of enterobacterial common antigen. SlyA/KvrA and MprA/KvrB both promote capsule transcription. The SlyA/KvrA protein acts as a temperature-dependent switch which acts by relieving H-NS-mediated transcriptional silencing; H-NS suppressed expression of rcsA and the three capsule operons in a clinical K. pneumoniae strain of capsule type K39. Capsule is also affected by the composition of the cell envelope, and mutations in lpp or various LPS-related genes can reduce the retention of capsule at the cell surface. Note that several genes related to cell envelope composition and membrane stress have been shown to regulate the Rcs system; therefore, some of the cell envelope component genes identified in our study may act through RcsB. Our work has also uncovered novel regulators of capsule that, at this stage, cannot be tied to the wider regulatory network, such as argR, and the ABC transporter Sap. We intend to define the mechanisms by which these genes affect capsule in future studies.

32We have developed a simple, robust technology for genome-wide studies of bacterial capsule, density-TraDISort, and applied it to identify capsule regulators in two strains of K. pneumoniae. In doing so, we have identified multiple positive and negative regulators of capsule production, including several genes not previously linked to capsule in this species.

33To our knowledge, this was one of the first studies employing physical selection independently of bacterial survival and growth to separate TraDIS libraries and represents the first time that density-based physical selection has been applied to studying capsule regulation in K. pneumoniae. TraDISort/FAST-INSeq technology with fluorescence-based sorting has to date been used to identify genes affecting efflux of ethidium bromide and mutations influencing expression of a Salmonella enterica serovar Typhi toxin reporter. We have expanded the utility of this method by adding a selection step based on cell density, allowing us to resolve different capsulation states. We envisage that, in addition to facilitating genome-wide screens for altered capsulation in other bacterial species, density-TraDISort could be used to identify genes affecting cell size and shape or cell aggregation.

34Our study was the second application of TraDIS to screen for genes affecting bacterial capsule production, following a recent study focused on UPEC. The UPEC study utilized a capsule-specific phage to positively select transposon insertion mutants lacking capsule; two novel capsule regulators were identified in this way. Compared with phage-based selection, our method offers increased sensitivity—mutants with a range of capsule phenotypes can be identified, in addition to capsule-null mutations. In addition, there is an option for very stringent selection of hits, as cutoffs can be applied on the basis of both negative selection (loss from the top fraction) and positive selection (enrichment in another fraction). However, density-based selection is less specific to capsule than phage infection, and there is the possibility that mutations could affect cell density in a capsule-independent manner. Interestingly, one of the novel Klebsiella capsule regulators identified in this study, MprA, was also shown to regulate capsule in UPEC. In Klebsiella pneumoniae, this gene increases capsule production above a baseline in hypermucoid strains (Fig. 3A and 4; see also Fig. S4B in the supplemental material), while a UPEC ΔmprA mutant did not produce capsule at all.

35We have shown that capsule production in K. pneumoniae NTUH-K2044 is controlled by many different global regulatory systems, allowing us to provide a detailed snapshot of the control of capsule in this strain (Fig. 6). Note that our assay was performed on bacterial cells at the late-stationary-growth phase, in LB medium, under microaerophilic conditions. This condition was used in this study because the associated level of capsule production is high, offering good resolution for capsule-based selection of mutants. Additional regulators, linked to different cues and stresses, are likely to be involved in different environments. Many of the regulators identified in this study were called as hits only in the hypermucoid strain, K. pneumoniae NTUH-K2044. It remains to be seen whether these same regulators control capsule (though to a degree outside the resolution of our gradient) in other K. pneumoniae strains; note, though, that several genes of K. pneumoniae ATCC 43816 (including uvrY, barA, csrB, rcsA, and rcsB) met our first screening criterion of being lost from the top fraction but not the second of being enriched in the bottom fraction. We speculate that capsule production is subject to complex environmental control across the Klebsiella species but that the hypermucoid phenotype is more costly to maintain and more sensitive to disruptions in its regulatory network.

36Many of our hits are involved in the synthesis of other cell surface polysaccharides; these included genes for enterobacterial common antigen (ECA), as well as genes for the synthesis or modification of LPS. ECA is a nonimmunogenic surface glycolipid found in various forms in Enterobacteriaceae, and structural modifications in this moiety can induce the Rcs system. LPS is a major contributor to K. pneumoniae pathogenesis in sepsis, though to a lesser extent in pneumonia, and various LPS modifications have roles in immune modulation during infection. We are confident that the LPS mutations identified in our study affect capsule retention or biosynthesis, rather than density per se, because (i) deletion of the O-antigen ligase waaL gene did not reduce cell density in an acapsular K. pneumoniae NTUH-K2044 strain (Fig. S4A); (ii) some LPS, but not all, biosynthesis genes were hits in our screen; and (iii) the glucuronic acid moieties on the core LPS polysaccharide are required for capsule retention in K. pneumoniae. In both of the strains that we studied, disrupting genes of the arn operon reduced capsulation (Table 1; see also Table S4 in the supplemental material). The arn operon has been shown to be responsible for modifying lipid A of LPS with 4-amino-4-deoxy-l-arabinose to mediate resistance to peptide antibiotics but has not previously been linked to capsule. The arnEF genes encode a flippase thought to translocate the modified arabinose across the cell membrane, while arnD is involved in its biosynthesis. We hypothesize that the reduced capsule of mutants of arnD and arnEF is independent of Lipid A modification, because other genes in this operon did not affect capsule and because a previous study showed that lipid A modification with l-Ara4N does not occur in cells grown in LB. Overall, our results hint at a high degree of interdependence among the three major surface polysaccharides of K. pneumoniae.

37The sap ABC transporter, when mutated, was found to promote capsule production by increasing the expression of capsule middle genes (Fig. 3A, 4A and B, and 5C) (Table 1; see also Table S3 and S4). To our knowledge, our study is the first to implicate sapABCDF in capsule regulation, though its full functions (or, indeed, the substrate of this transporter in Klebsiella) are not known. The H. influenzae Sap homologue mediates resistance to antimicrobial peptides by importing them for degradation and is also required for haem uptake, while the Sap pump in E. coli has been reported to export putrescine and facilitate potassium import through TrkGH. We found that the Sap transporter did not affect antimicrobial peptide resistance, which was also observed in E. coli. It is unclear how Sap mutation induces wcaG while suppressing an important component of the Rcs system—more work will be needed to define the role and mechanism of this transporter in K. pneumoniae. For the four mutants characterized in detail in this study, it would be interesting to examine their phenotypes in mammalian models in addition to the invertebrate model used here, to see how these genes influence specific host-pathogen interactions.

38We have developed a simple, broadly applicable method for studies of capsulation and used it to define the regulatory network that controls capsule in K. pneumoniae NTUH-K2044. We have also identified genes required for full production of capsule in a K2 strain. Although the majority of regulators are located in the core genome of K. pneumoniae, there are differences in the specific regulators deployed in the two strains that we investigated, and it would be interesting to determine whether this pattern of strain-specific regulatory networks comprising primarily core genes holds across the Klebsiella phylogeny. This intraspecies comparison, together with our data showing that density-based capsule selection can be used in other capsulated bacteria, also opens the possibility for robust interspecies comparisons of capsule regulation.

39K. pneumoniae strains were cultured routinely in LB media supplemented with 1.5% (wt/vol) agar as appropriate. Cultures were supplemented with 12.5 μg/ml chloramphenicol and 12.5 μg/ml tetracycline when required. S. pneumoniae strains were grown on blood agar plates (Oxoid; CM02718) in microaerobic candle jars containing CampyGen sachets at 37°C or in static brain heart infusion (BHI) liquid media (Oxoid; SR0050C). The list of strains, plasmids, and oligonucleotides used in this study is reported in Table S1 in the supplemental material.

40TraDIS libraries were generated using the mini-Tn5 transposon delivery plasmid pDS1028, introduced into the recipient strain by conjugation. Full details are provided in Text S1 in the supplemental material.

41Bacterial mutant libraries were separated on the basis of their capsule expression by centrifugation on a discontinuous Percoll (GE Healthcare) density gradient for 30 min at 3,000 × g (Fig. 1B). Full details are provided in Text S1.

42Genomic DNA (gDNA) was prepared from overnight cultures of single reduced-capsule mutants using a DNeasy blood and tissue kit (Qiagen). Random-prime PCR to identify the transposon insertion site in each gDNA template was performed as previously described using primers FS57-59 and FS109 and Herculase II polymerase (Agilent). Amplicons were sequenced using primer FS107.

43Genomic DNA (gDNA) was prepared from each Percoll-resolved fraction by phenol-chloroform extraction. Two micrograms of DNA from each gDNA preparation was used to prepare TraDIS transposon-specific sequencing libraries as described previously, using primer FS108 for specific amplification of transposon junctions. Sequencing was carried out on an Illumina MiSeq platform using primer FS107.

44The analysis of TraDIS sequencing results was carried out using the Bio-TraDIS pipeline as described previously, with minor modifications (see Text S1). All scripts used in this study are available at https://github.com/sanger-pathogens/Bio-Tradis and https://github.com/francesca-short/tradis_scripts. Comparisons between fractions were based on normalized read counts per gene. Genes with (i) reduced mutant abundance in the top fraction and (ii) increased mutant abundance in the middle or bottom fraction were called as decreased capsule hits, with thresholds of an absolute change in log2FC of >1 and a q value of <0.001. Increased capsule hits in NTUH-K2044 were defined as those with severely reduced mutant abundance in the middle fraction (log2FC < −3; q value < 0.001) without enrichment in the bottom fraction (log2FC < 1), with genes containing very few reads in any fraction excluded (i.e., the value corresponding to log2 counts per million in the top fraction was greater than 4).

45Generation of the pan-genome and the method of enrichment analysis are described in Text S1.

46Single-gene-knockout mutants were constructed in K. pneumoniae by allelic exchange. Upstream and downstream sequences (> 500 bp) for each target gene were amplified and joined by overlap PCR, cloned into pKNG101-Tc, and introduced into the recipient strain by conjugation with the E. coli β2163 donor strain. All primers used, and the resulting constructs, are listed in Table S1. Conjugation patches were incubated for 1 h at 37°C and then for 16 h at 20°C. Single-crossover mutants were selected on LB agar plus 15 μg/ml tetracycline. Double-crossover mutants were selected on low-salt LB agar plus 5% sucrose at room temperature and were subsequently patched onto LB plus sucrose and LB plus tetracycline plates to confirm loss of the vector. Mutants were confirmed by PCR across the deleted region. Mutants were complemented by reintroduction of the relevant gene into its original location on the chromosome by allelic exchange as described above, using a vector carrying the gene and its flanking region. The ΔsapBCDF cps (K. pneumoniae 1_3713 [KP1_3713]) double mutant was generated by random transposon mutagenesis of the ΔsapBCDF strain with the pDS1028 vector, followed by selection of acapsular mutants from the pool by density-gradient centrifugation and random-prime PCR to identify the insertion site.

47Capsule extraction and quantification of uronic acids were performed as described previously, with modifications (see Text S1).

48Cultures of K. pneumoniae were grown overnight in 5 ml LB medium at 37°C. These cultures were sedimented at 1,000 × g or 2,500 × g for 5 min (room temperature). The optical density at 600 nm (OD600) of the top 500 μl of supernatant was determined by spectrophotometry. Results were expressed as a ratio of the supernatant OD600 to that in the input culture.

49Colonies were taken directly from an agar plate, frozen at high pressure in a Balzers HP010, and freeze-substituted for 8 h in acetone containing 0.1% tannic acid and 0.5% glutaraldehyde at −90°C followed by 1% osmium tetroxide–acetone for 24 h at −50°C. They were then embedded in Lowicryl HM20 monostep resin. Ultrathin sections were cut on a Leica UC6 ultramicrotome and contrasted by the use of uranyl acetate and lead citrate. Images of bacteria were taken on an FEI Spirit Biotwin 120 kV TEM with a Tietz F4.15 charge-coupled-device (CCD) camera.

50Bacteria were grown in LB to an OD600 of 1, pelleted, and resuspended in sterile phosphate-buffered saline (PBS). Human sera (Sigma-Aldrich S7023) (400 µl) was prewarmed to 37°C and added to 200 µl bacterial suspension, and the mixture was incubated at 37°C for 2 h. Viable bacterial counts were determined before and after incubation.

51Galleria mellonella infection. Larvae of G. mellonella were purchased from BioSystems Technology Ltd. (United Kingdom) (research-grade larvae) and used within 1 week. Bacteria were grown overnight, subcultured and grown to an OD600 of 1, and then resuspended in sterile PBS. Larvae were infected by injecting the bacterial suspension (105 cells) into the right hind proleg of the larvae using a Hamilton syringe. Infected larvae were incubated at 37°C and monitored every 24 h and were scored as dead when they were unresponsive to touch. Thirty larvae were used per strain, and these were infected in three batches of 10 using replicate cultures.

52Strains were grown overnight, subcultured, and grown to an OD600 of 1.0. This culture (100 µl) was spread on the surface of an LB agar plate and dried, and an Etest (bioMérieux) strip was placed on the surface of the plate. Plates were incubated face up at 37°C, and the result was read after 6 h to avoid overgrowth of the capsule interfering with the reading.

53Bacteria were grown in LB medium at 37°C to an OD600 of 1.0. Cultures were mixed with 2× volumes of RNAProtect reagent, centrifuged, and RNA extracted using a MasterPure Complete DNA and RNA purification kit (Epicentre) according to the manufacturer’s instructions. Samples were then subjected to in-solution DNase I digestion (Qiagen) and cleaned up using a Qiagen RNeasy minikit. Reverse transcription of 200 ng RNA was performed using ProtoScriptII enzyme (NEB) per the supplied instructions.

54Transcripts were quantified using a StepOne real-time PCR instrument with a Kapa SYBR FAST qPCR kit. Relative abundances were determined using the relative standard curve method with K. pneumoniae NTUH-K2044 gDNA as a standard and recA as the reference gene.

55Sequences generated during this study have been deposited into the European Nucleotide Archive (ENA; http://www.ebi.ac.uk/ena) under study accession number ERP105653.