The genomes of Kp52.145, SB2390 and SB3193 were sequenced by a combination of 454 and Illumina technologies using single and paired-end libraries. Finishing efforts resulted in the complete genome sequence of K. pneumoniae Kp52.145 (one chromosome + two plasmids), comprising 5.45 Mbp and 5,314 protein coding genes (Figure 1). SB2390 and SB3193 genomes were assembled in 11 and 17 scaffolds, respectively. The GC% of these three genomes ranged from 55.6% to 56.7%. The general features of the K. pneumoniae sequenced genomes are summarized in Table 1.

The three genomes sequenced in this study are compared to strains NTUH and MGH, previously published. PCG, protein coding genes.

Because Kp52.145 is a highly virulent strain, our analyses were focused on comparing its genome with the genomes of K. pneumoniae strains SB2390, SB3193, NTUH-K2044 and MGH78578. According to SEED subsystems annotations 25 , about 60% of protein-coding genes for each K. pneumoniae genome had predicted functions. More specifically, the largest percentage of annotated genes is involved in the metabolism of carbohydrates (approximately 20%), of amino acids and their derivatives (approximately 10%) and of cofactors, vitamins and prosthetic groups (approximately 8%) [see Additional file 1].

To define the common genome of the five strains, we used stringent BlastClust parameters of at least 90% identity and at least 80% coverage. This analysis identified 3,587 coding sequences common to the five genomes. The majority of proteins are involved in metabolic processes, such as energy metabolism and transporters, supporting the general concept that the core genome encompasses essential functions required for survival of the microorganism. The K. pneumoniae core genome comprised several sets of genes whose functions are related to bacterial survival in the environment or interaction with its host, and possibly virulence. This was the case, for example, of genes involved in quorum-sensing and biofilm formation, adhesins, and secretion systems, for which examples are detailed below.

Genes encoding for autoinducer-2 and type III fimbria, involved in biofilm formation in K. pneumoniae 26 27 28 , were present in all sequenced strains. In addition, genes coding for synthesis and transport of the poly-β-1,6-N-acetyl-D-glucosamine (PGA) adhesin (KpST66_4915 to KpST66_4918 in Kp52.145 genome) which is required for the structural stability of Escherichia coli biofilms 29 , and YidE (KpST66_0019), which mediates the hyperadherence phenotype of E. coli 30 , were also found in the core genome of K. pneumoniae strains.

Moreover, the five K. pneumoniae genomes contained the genes barA/uvrY (respectively KPST66_0986 and KpST66_3517) and ycjX/ycjF (KpST66_2441 and KpST66_2442) that may be involved in bacterial fitness and virulence. The two-component system BarA/UvrY (KpST66_3517 and KpST66_0986 in Kp52.145 genome) contributes to biofilm formation in Salmonella enterica and is a virulence determinant of urinary tract E. coli infections 31 32 . The E. coli YcjF protein is expressed in a septicemia murine model of infection in which the ycjF mutant is attenuated, thus suggesting its implication in the in vivo survival/multiplication of the bacteria 33 .

Several putative secretion systems were identified as part of the common genome of the K. pneumoniae strains, including one type I secretion system (T1SS) and one type II secretion system (T2SS). T2SS is composed of the pullulanase related genes pulA-O that are involved in the pathogenesis of several bacteria 34 35 . Streptococcus pyogenes PulA binds to host lung glycogen leading to a strong interaction with alveolar type II cells 36 . Similarly, the alpha-amylase AmyA degrades glycogen into cyclic maltodextrins, which increases the transepithelial translocation of Streptococcus 37 . Both amyA and pulA-O genes are encoded in K. pneumoniae genomes, but their functions in this bacterium remain to be characterized. Type VI secretion system (T6SS) putative genes were located in at least three different loci of the K. pneumoniae genomes, in accordance with a previous in-silico study 38 . T6SS clusters are usually found within pathogenicity islands or on chromosomal regions presenting virulence or host survival biases. Additionally, T6SS has been suggested to assist colonization and infection. Indeed, in a screen that identified K. pneumoniae mutants failing to colonize mice 39 , two of them were mutants in genes coding for proteins annotated as hypothetical, that have been subsequently re-annotated as T6SS proteins 38 . Type III secretion systems were not found in Klebsiella, but type IV secretion systems, possibly corresponding to conjugation apparatus, were present only in some strains (Kp52.145, SB2390 and NTUH-K2044) (Table 2).

Furthermore, the core genome presented a large genomic region located between gly-tRNA and phe-tRNA loci in Kp52.145 genome, containing frdABCD genes coding for the fumarate reductase enzymatic complex responsible for fumarate respiration under anaerobic growth of bacteria. This complex is a virulence determinant for Helicobacter pylori, Mycobacterium tuberculosis, Actinobacillus pleuropneumoniae and S. enterica, as mutants on these genes are attenuated 40 41 42 43 . The ability to grow anaerobically allows bacterial pathogens to persist in host tissues, including in the lungs. Curiously, this genomic locus encoded in K. pneumoniae at least one more protein involved in anaerobic metabolism, the anaerobic C4-dicarboxylate transporter DcuA (KpST66_4904), supporting the idea that this GI provides K. pneumoniae advantages to grow under anaerobic conditions, possibly favoring infection. Additional proteins that might be involved in bacterial fitness to environmental stress conditions were encoded in this region. For instance, the putative small multidrug resistance protein SugE (KpST66_4882) has been shown to regulate biofilm formation and capsule expression 44 . A lipocalin-2 bacterial protein Bcl (KpST66_4881) is also encoded in this island of the genome. Eukaryotic lipocalins are small extracellular proteins that bind hydrophobic ligands and fulfill numerous biological functions including regulation of cellular homeostasis and immunity and are regulators of antibacterial defense 45 . Lipocalin2 is for instance a siderophore scavenger for several bacteria, including K. pneumoniae 46 , as well as a negative regulator of inflammatory response during Streptococcus pneumoniae pneumonia 47 . However, the role of bacterial lipocalins is not yet known.

Finally, a phosphatidylserine decarboxylase (KpST66_4873) might be important for the integrity of the bacterial membrane composition, as phospholipid biosynthetic pathways play crucial roles in the virulence of several pathogens 48 49 .

The accessory genome (genes absent in at least one of the five strains) included a large number of specific coding sequences (CDSs): 743 genes were found only in Kp52.145, 608 in NTUH-K2044, 806 in SB3193, 635 in SB2390 and 488 in MGH78578. About 50% of these genes code for hypothetical proteins or proteins of unknown function. The distribution of the putative virulence-related genes of K. pneumoniae among the sequenced strains is summarized in Table 2. The specific regions or genes of Kp52.145 are detailed below.

Strain Kp52.145 possessed two plasmids (Figure 1B). The first plasmid of 121 Kb carried the aerobactin cluster and the regulator of mucoid phenotype rmpA genes. The presence of this plasmid was previously correlated with the virulence of K. pneumoniae K1 and K2 isolates 21 50 . Additionally, a strain cured from this plasmid showed a 6 × 10⁴-fold reduction in virulence, establishing the link between this plasmid and bacterial virulence 21 . In addition to aerobactin and rmpA, this plasmid contained genes coding for F-pilin, purine metabolism, insertion sequences and proteins of unknown functions. Kp52.145 also carried a second, previously non-described, plasmid of about 90 Kb. Interestingly, it contained rmpA2, a homologue of the regulator of mucoid phenotype rmpA, which seems to be involved in capsule expression regulation 51 52 . F-pilin genes, a subtilisin-related serine protease, an AAA + ATPase, the UV protection system UmuD/UmuC and the genes encoding the toxin-antitoxin systems RelE/orf4 and VagD/VagC (Figure 1B) were also found on this plasmid. However, the potential role of these genes in virulence remains to be investigated.

In addition to the capsule synthesis operon and the iron-acquisition systems (yellow and orange boxes in Figure 1A, respectively), known to be involved in K. pneumoniae virulence, four additional regions of the Kp52.145 genome presented several characteristics of pathogenicity islands (red boxes, Figure 1A). These regions are defined by a different GC content in comparison to the average of the genome, represented large chromosomal regions (often > 30 Kb), were associated with tRNA genes or with the presence of insertion sequences, integrases and transposases, and were present in pathogenic strains while less frequent in less-virulent strains 53 54 . The four GIs present in the Kp52.145 genome were present or partially present in NTUH, but none of them was present in SB2390, SB3193 and MGH 78578, indicating that their occurrence is specific to pathogenic genomes. Figure 2 shows the four GIs identified, highlighting the putative virulence related genes.

The largest GI found in the Kp52.145 genome comprised 133,679 bp, presented a GC content of 52%, was inserted in an asn-tRNA locus and encoded 92 CDSs (Figure 2A). Most of the protein-coding genes found in this region were described as part of the IceKpI GI of NTUH-K2044 55 56 , although several differences were found. Kp52.145 GI-I begins at asn tRNA locus followed by several uncharacterized proteins and insertion sequence elements. GI-I coded for the synthesis of two polyketide/nonribosomal peptides (yersiniabactin and colibactin) and for the conjugative transfer machinery (T4SS) that allows horizontal transfer of the island 56 . In contrast to the previous description of IceKpI 56 , the Kp52.145_GI-I carried colibactin and did not contain the region coding for vagC-vagD, iroN-iroB-iroC-iroD and rmpA genes which are carried only by the 121 kB plasmid in Kp52.145.

We describe here a novel GI (GI-II). It is a 29,829 bp island, with a GC content of 49% and coding for 28 CDSs which is inserted in a leu-tRNA locus (Figure 2B). Potential pathogenesis-related genes coded for a putative cytotoxic outer membrane protein (cOMP, KpST66_4736) and a polyamine ABC transport system (KpST66_4729 to KpST66_4732). cOMP closest known homologue (34% identity) was a Plesiomonas shigelloides predominant virulence factor proposed to trigger cell death in host cells following infection 57 . Polyamine biosynthesis and transport mechanisms were intricately linked to fitness, survival, biofilm formation and pathogenesis, for instance in S. pneumoniae and Yersinia pestis 58 59 . Additionally, this GI encoded a 4-phytase gene (KpST66_4736), ugpQ3 (KpST66_4728) and xylA, xynT, xynB, xylR (KpST66_4724 to KpST66_4727) that are involved in xylose metabolism.

The third GI is characterized by a 49,657 bp region presenting a G + C content of 51% and contained 66 CDSs, most of them coding for phage structural proteins (Figure 2C). This GI included genes coding for proteins with homologues that were shown to play a role in bacterial adhesion and immune system escape 60 61 62 . These proteins encoded for an immunoglobulin domain-containing protein (KpST66_1506), a peptidase S24-like protein (KpST66_1511), two HNH family endonucleases (KpST66_1468 and 1486) and an exonuclease (locus 1464).

The forth island (GI-IV) was mainly comprised of phage-related genes. Among the 42 CDSs encoded within this genomic region, one gene coded for a SEFIR-domain containing protein (KpST66_1945; Figure 2D). A SEFIR domain is usually found in IL17 receptors and SEF proteins, acting in eukaryotes signaling pathways. Very little is known about prokaryotic SEFIR-containing proteins. Structural analyses suggested that these bacterial SEFIR domains share structural and electrostatic similarity with their mammalian homologues and, thereby, could potentially subvert host immunity by hijacking the IL17R signaling pathways 63 . Notably, local production of IL-17 is a significant factor in effective host defense against Gram-negative bacteria, including K. pneumoniae 64 . Therefore, further studies are required to elucidate whether KpST66_1945 is implicated in K. pneumoniae pathogenesis.

To investigate the distribution of the described GIs in K. pneumoniae, the presence of the putative virulence-related genes was searched using BlastN in 171 genomes, including 119 publicly available and 52 unpublished genomes (Bialek-Davenet, Brisse et al., unpublished work) representing 47 different sequence types (STs). Whereas the SEFIR-domain containing protein gene of GI-IV was only found in two (1.2%) isolates of sequence type ST15, the three other GIs described herein were more distributed among K. pneumoniae isolates (Table 3). GI-II genes were present in a total of 11 (6.4%) isolates, most of which belonged to ST375, ST65 and ST25, which were associated with severe infections caused by isolates of capsular serotype K2 9 65 . The genes of GI-II were always found in synteny. GI-III genes were observed in only seven (4.1%) isolates dispersed in several unrelated STs. The distribution of the ICEKpI element (similar to GI-I) has been previously analyzed 55 .

^aaccording to ref. 55 . Prevalence of virulence-related features encoded in GI-II, GI-III, GI-IV and T6SS insertion was based on >90% identity and >50% coverage in the length of the genes, using a database of 171 Klebsiella genomes representative of 47 different STs. STs, sequence types.

Recently, three different T6SS loci were defined in K. pneumoniae 38 . Within these loci, three putative valine-glycine repeat (Vgr) proteins and two hemolysin-coregulated proteins (Hcp) were described as potential effector proteins, through their sequence similarities to Vibrio cholerae and Pseudomonas aeruginosa effector proteins 65 66 67 68 . Accordingly, the Kp52.145 genome also presented three conserved T6SS loci syntenic to those previously described. The first two loci were identical in composition and orientation to the previously described ones. The third one, locus III, had conservation of adjacency limited to the imcF/impG/impH and impJ/ompA/vgrG gene clusters, as a region with nine genes was inserted between these two clusters. This insertion encoded for one hypothetical protein, five putative Sel-1 repeat containing lipoproteins and three putative phospholipase D family proteins (Figure 1A). Flanking sequences suggesting how this region was inserted were not found.

Sel1 lipoproteins are poorly characterized and there was no evidence for their function in K. pneumoniae, but they are essential in Legionella pneumophila for invasion of host cells where they influence vacuolar trafficking of bacteria 69 . The three open reading frames encoding putative phospholipase D proteins in strain Kp52.145 encoded one full length protein (KpST66_3368, 623 aminoacids), one C-terminal region (KpST66_3371, 187 aa) and one N-terminal region (KpST66_3372, 317 aa). Phospholipase D family proteins have been described as important for host cell invasion, bacterial dissemination and disease progression 70 71 72 73 . The bacterial phospholipase D family comprises at least four classes of proteins with distinct functions: true phospholipase D, cardiolipin synthase, phosphatidylserine synthase and endonuclease 74 . Full length K. pneumoniae PLD1 and its closest homologs all presented the conserved motif HXK(X)₄D and a serine or threonine approximately eight residues after asparagine (Figure 3), but no other conserved domain was described in each family, thus making it difficult to infer protein function only by sequence analysis.

In order to obtain evidence that the product of these three genes coding either for full length or partial PLD-family proteins are important for bacterial survival in vivo, we checked by RT-PCR for their mRNA expression. We observed that these genes are expressed both in bacteria grown for four hours in Trypto Casein Soy broth (GTCS) medium, as well as in the lungs of mice infected for 24 hours (data not shown). These results prompted us to further check for a putative involvement of the full length PLD gene, called pld1, on K. pneumoniae virulence.

As PLD-family proteins have been shown to be involved in virulence 75 76 77 , we decided to characterize the role of the full length PLD1 protein. We first tested a pld mutant strain in a K. pneumoniae murine model. Mice were infected intranasally with 10⁸ of either the wild-type bacteria, a pld1 mutant, or the mutant strain complemented with a plasmid expressing the putative PLD1 protein (pPLD), and their survival was monitored for seven days. Interestingly, the mutant strain appeared avirulent in a mouse model of acute pneumonia while mice infected with the wild-type and the complemented strain succumbed in less than one week (Figure 4). However, the wild-type, the mutant and the complemented mutant strains grew equally well in Luria-Bertani broth (LB) broth (data not shown). These results indicated that the pld1 mutant was strongly attenuated in vivo, thus showing an important role for PLD1 in virulence.

To analyze the frequency and clonal distribution of the pld1 gene in K. pneumoniae, the 171 genomes were analyzed using BlastN. We observed that besides ST66, represented by strain Kp52.145, pld1 was present in 10 strains (6.4%) belonging to ST380, ST679, ST67 (K. pneumoniae subsp. rhinoscleromatis) and ST35, but in none of the other isolates. It is interesting to note that ST380 was associated with severe K. pneumoniae infections 9 65 and that K. pneumoniae subsp. rhinoscleromatis is the only Klebsiella strain to be able to survive intracellularly in macrophages 78 .

In order to demonstrate the phospholipase activity of PLD1 and characterize its involvement in lipid metabolism, the lipid composition of wild-type and mutant strains was analyzed by thin-layer chromatography (TLC). A remarkable lipid spot was absent from the pld1 mutant in comparison with the complemented strain, suggesting that the putative PLD1 is involved in lipid metabolism [see Additional file 1]. To reinforce this result, a plasmid carrying the pld1 gene was inserted into E. coli strain SD9 79 . This strain is deficient in phosphatidylserine and cardiolipin, thus presenting a simpler lipid composition than its parental strain and Kp52.145. Lipid profiles of SD9 and complemented strains had a different lipid composition. Notably, the PLD1-expressing strain contained an additional lipid spot in comparison to the SD9 strain, suggesting that PLD1 is responsible for this difference (Figure 5A). SD9 wild-type strain also presented an extra lipid spot in comparison to the PLD1-expressing strain, possibly representing the PLD1 substrate (Figure 5A). Densitometric analysis of iodine-stained lipids on TLC plates revealed that this lipid spot corresponded to 21% of the total amount of lipids in SD9 strain, but only 6% of SD9 complemented with a plasmid expressing pld1. Mass spectrometry (MS) analysis of total lipid extract was carried out to identify such lipid. Comparing lipid profiles by MS, we found a lipid of mass 788.4 present only in the PLD1-deficient strain (Figure 5B) and identified it as phosphatidyl glycerol (PG) using the LipidMaps database.

As mentioned above, the bacterial PLD-family proteins can be classified in four subfamilies. One of them, the cardiolipin synthase is able to convert two PG molecules into glycerol and cardiolipin, or to catalyze the opposite reaction, leading to PG formation 80 . Our results suggest that PLD1 belongs to the cardiolipin synthase subfamily and that it plays a role in balancing the PG and cardiolipin content.

It has been shown that humans and mice with bacterial pneumonia have markedly elevated amounts of cardiolipin in lung fluid and that it impairs surfactant function, lung mechanics, modulation of cell survival and cytokine networks and lung consolidation 81 .

There is evidence that bacteria are able to adjust their relative concentrations of phosphatidylethanolamine and PGs when subjected to environmental stresses. Such an alteration in headgroup composition seems to be a means for changing membrane permeability and, hence, preserving stability 82 . Therefore, we hypothesize that PLD1 alters the membrane composition and charge, affecting bacterial interaction with the host environment.

Recently, Russel el. al. demonstrated that diverse phospholipase proteins encoded within the T6SS loci of several prokaryotic genomes are antibacterial effectors, conferring competitive advantages on the donor strain during interbacterial interactions 83 . These proteins are generally designated as ‘T6SS Lipase Effectors’ (Tle) and classified in five sub-families, according to the sequence conservation and number of catalytic motifs present. Kp52.145 PLD-family protein could be considered a Tle5 member, as it presents two conserved HxKxxxxD motifs. However, in the Kp52.145 genome we did not identify any gene similar to the cognate immunity genes - a hallmark of the genomic islands described by Russel et al. Moreover, we did not observe such an antibacterial effect of Kp52.145 or its PLD1 mutant strain upon competition with E. coli [see Additional file 1]. These results showed that pld1 is implicated in virulence without being an anti-bacterial factor and is, so far, unique.

Strain Kp25.145 (a derivative of B5055, the reference strain of serotype K2) is a laboratory strain used to study K. pneumoniae pathogenesis 11 21 and was chosen as the focus of this work. SB2390 (cur15505, isolated in Curaçao, 2002, urinary tract infection; belongs to ST14) and SB3193 (IPEUC-744, isolated from a metritis case in a mare, 1981 in France; belonging to ST82) 23 are non-virulent strains that were sequenced to allow the comparison between virulent and non-virulent strains.

The pld1 transposon mutant was isolated by Anna Tomas and Jose Bengoecheoa during a screen of a K. pneumoniae mutant library made by Tn5 transposon insertion (manuscript in preparation). In this mutant, the transposon was inserted at position 1,625 of pld1 gene.

The complementation of the mutant strain was achieved through bacterial transformation using electrocompetent cells and a plasmid carrying pld1 gene. pld1 gene was amplified by PCR and cloned at the multiple cloning site of puc18 plasmid. Cloning was confirmed by DNA sequencing.

K. pneumoniae strains were sequenced using a combination of 454 and Illumina reads. Single and paired-end 454 reads with an average of 400 nucleotides were assembled into contigs and scaffolds by Newbler. Illumina reads of about 76 or 36 nucleotides were aligned to scaffolds in order to confirm and correct possible homopolymer errors in the 454 reads. Coverage was as follows: Kp52.145 genome: 170 X using GAIIX (76 nt) + 13.8 X using MP titanium + 18 X using SR titanium; SB2390 genome: 81 X using GAIIX (36 nt) + 6.4 X using MP titanium + 22 X using SR titanium; SB3193 genome: 209 X using GAIIX (76 nt) + 5.4 X using MP titanium + 20 X using SR titanium.

Following the primary assembly of the genomes, an in silico finishing approach, based on the methods described by Pop et al. 84 , was performed in order to identify small and identical repeats on the genomes. In such cases and without high quality bases discrepancies, the contigs containing the small repeats were manually duplicated and added to the assembly.

Scaffolds were aligned to experimentally determined BglII optical maps generated by OpGen company from purified chromosomal DNA, using MapSolver version 3.1 software. Such alignments were used to check for the quality of the assemblies. Additionally, specific pairs of primers were designed in order to close all remaining gaps. PCR products were purified in NucleoFast 96 plates (Macherey Nagel, Düren, Germany) and aliquots were used for sequencing reactions with the BigDye Terminator Cycle Sequencing Ready reaction Kit (Applied Biosystems, Foster City, CA, USA) on a ABI Prism 3730XL DNA Analyzer (Applied Biosystems). The resulting sequences were added to the previous assembly using Phred/Phrap/Consed.

The sequences of K. pneumoniae genomes have been deposited to the European Nucleotide Archive and are accessible under the accession numbers: FO834904, FO834905 and FO834906 (strain Kp52.145), CCBO00000000 (strain SB2390) and CCCQ000000000 (strain SB3193).

In order to gain functional insights about the genome sequences, protein-coding genes were predicted and annotated using the CAAT-box genome browser 85 , using a combination of GeneMark predictions and Blastx results against the Uniprot database. All the putative open reading frames (ORF) longer than 120 nucleotides presenting similarity to sequences of the Uniprot database or positive GeneMark result were considered for further analysis. The final set of CDSs underwent a manual annotation process based on description of similarity. Pfam and COG database searches, as well as SignalP, TMHMM and PredTMBB predictions were performed to improve the degree of annotation confidence, if necessary. CDSs were described as ‘highly similar to’ , ‘similar to’ or ‘weakly similar to’ if they presented more than 70%, between 50% and 69% or less than 50% similarity to the protein hit sequence. Additionally, information on partial homology was included. The start codon for each CDS was automatically chosen and manually validated, based on a combination of GeneMark results and Blast alignments. RAST was used to classify proteins in functional categories. Structural RNAs were searched using tRNAscan.

The common genome of the five strains was determined using the BlastClust algorithm 86 using minimum parameters of 90% identity and 80% length coverage for proteins to be included in the same cluster.

Lungs of control or Kp52.145-infected mice were homogenized in cold TRI reagent (Sigma, Gillingham, Dorset, UK) using a Precellys lysing kit (Precellys, Saint Quentin en Yvelines, France). Total lung and bacterial mRNAs were extracted according to the manufacturer’s instructions. RNA (2 μg) was reversed transcribed in cDNA using Superscript II enzyme (Invitrogen, Foster City, CA, USA). Aliquots were used in PCR reactions using specific primers [see Additional file 1].

BALB/cJ mice were purchased from Janvier (Le Genest St. Isle, France). Mice were housed under standard conditions of feeding, light and temperature with access to food and water ad libidium. Experiments were performed according to the national and Institut Pasteur guidelines for laboratory animal experiments. Protocols were approved by the Institut Pasteur animal care and use committee (protocol 05–59) and the Direction des Services Vétérinaire de Paris (permit 75–713 to RT). Six- to eight-week-old mice were anesthetized with acepromazine (Calmivet, 1.5 mg/kg, Vetoquinol, Lure, France). and ketamine (Imalgene, 31.25 mg/kg, Merial, Lyon, France). and then infected intranasally with 20 μl bacterial suspension. The virulence of K. pneumoniae strains was tested on six-week-old BALB/c mice, as previously described 23 . Seven mice per test condition were infected with 10⁸ bacteria. Mice were followed every day for one week. Experiments were performed at least twice.

Bacterial lipids were extracted by the method of Bligh and Dyer 87 . Briefly, bacterial stationary-phase cultures were concentrated 10 times and mixed with chloroform:methanol. After centrifugation at 1,000 rpm for five minutes, the organic phase was recovered. Lipid profiles were analyzed by two-dimensional TLC using TLC Silica gel 60 F₂₅₄ plates (Merck, Whitehouse Station, NJ, USA). as the stationary phase and a chloroform 9:1 methanol mixture as the mobile phase in both dimensions. Staining was performed by iodine vapor. TLC calibrated images were aquired in ImageScanner using LabScan v5.0 software. The relative intensity of each spot was calculated in ImageMaster two-dimensional Platinum v7.0 software. Alternatively, lipid extracts were analyzed by MS and MS/MS in an ESI-Q-Tof Micro (Waters), in positive ion mode. Resolution was typically lower than 10 ppm.

Competition assays were performed as previously described 88 . Briefly, K. pneumoniae Kp52.145 or its pld1 mutant cells grown overnight on an agar plate were resuspended in LB, normalized to OD₆₀₀ of 0.5 and mixed at a ratio of 5:1 with a spontaneous nalidixic acid (nal) resistant mutant strain of E. coli MG1655. The mixture was incubated for four hours on a prewarmed agar plate. Recovered cells were plated out on antibiotic selective media and viable cells were reported as the total number recovered per co-culture spot. Serratia marcescens DB10 was used as a positive control.