Comparative analyses of Legionella species identifies genetic features of strains causing Legionnaires’ disease

Background The genus Legionella comprises over 60 species. However, L. pneumophila and L. longbeachae alone cause over 95% of Legionnaires’ disease. To identify the genetic bases underlying the different capacities to cause disease we sequenced and compared the genomes of L. micdadei, L. hackeliae and L. fallonii (LLAP10), which are all rarely isolated from humans. Results We show that these Legionella species possess different virulence capacities in amoeba and macrophages, correlating with their occurrence in humans. Our comparative analysis of 11 Legionella genomes belonging to five species reveals highly heterogeneous genome content with over 60% representing species-specific genes; these comprise a complete prophage in L. micdadei, the first ever identified in a Legionella genome. Mobile elements are abundant in Legionella genomes; many encode type IV secretion systems for conjugative transfer, pointing to their importance for adaptation of the genus. The Dot/Icm secretion system is conserved, although the core set of substrates is small, as only 24 out of over 300 described Dot/Icm effector genes are present in all Legionella species. We also identified new eukaryotic motifs including thaumatin, synaptobrevin or clathrin/coatomer adaptine like domains. Conclusions Legionella genomes are highly dynamic due to a large mobilome mainly comprising type IV secretion systems, while a minority of core substrates is shared among the diverse species. Eukaryotic like proteins and motifs remain a hallmark of the genus Legionella. Key factors such as proteins involved in oxygen binding, iron storage, host membrane transport and certain Dot/Icm substrates are specific features of disease-related strains. Electronic supplementary material The online version of this article (doi:10.1186/s13059-014-0505-0) contains supplementary material, which is available to authorized users.


Background
Among the many pathogens provoking severe pneumonia, the Gram-negative bacteria Legionella pneumophila and Legionella longbeachae are responsible for Legionnaires' disease, a severe pneumonia that can be deadly if not treated promptly [1]. Although several of the more than 60 species described in the genus Legionella may cause disease, L. pneumophila is the major agent, responsible for nearly 90% of all cases worldwide. L. longbeachae comes second, causing around 2 to 7% of cases with the exception of Australia and New Zealand, where it is associated with 30% of Legionnaires' disease cases [2]. Legionella micdadei, Legionella bozemanii, Legionella dumoffii, Legionella anisa, Legionella wadsworthii and Legionella feelei are rarely found in humans and the remaining Legionella species have never or only once been isolated from humans [2]. This highly significant difference in disease incidence among Legionella species may be due to different environmental distributions and/or to different virulence potential for humans. Few studies have analyzed the environmental distribution of Legionella, although one survey in France showed that L. pneumophila, which had a prevalence of 95.4% in clinical isolates, was found in only 28.2% of the environmental samples tested, whereas L. anisa was isolated in 13.8% of the environmental samples but found only once (0.8%) in a clinical isolate [3]. Similarly, a more recent report from Denmark showed that only 4.5% of clinical cases were due to non-L. pneumophila strains and reported a strong discrepancy in the occurrence of different Legionella species in clinical and environmental isolates [4]. For example, L. anisa was highly abundant in the environment, but never found in clinical isolates. In contrast, L. bozemanni, L. longbeachae and L. micdadei were identified in clinical samples but never or rarely in environmental samples [4]. Furthermore, different Legionella species also seem to have a different host range and different capacities to infect human cells [5,6]. Taken together, independently of the environmental distribution, different Legionella species seem also to possess different abilities to infect eukaryotic cells and to cause disease in humans.
After publication of the L. pneumophila genome sequence in 2004 [7,8] and that of L. longbeachae in 2010 [9,10] several additional L. pneumophila strains have been sequenced [11][12][13][14] as well as a few draft genome sequences of other species. However, apart from Legionella oakridgensis [15] none has been analyzed in detail. Thus the vast majority of comprehensively analyzed genome sequences are from the major human pathogens L. pneumophila (eight genomes) and L. longbeachae (two genomes). In order to deepen our insight into species never or rarely found in human disease, we completely sequenced and analyzed the genomes of three Legionella species, L. micdadei, Legionella hackeliae and Legionella fallonii (LLAP10), selected based on their different epidemiological characteristics compared with L. pneumophila and L. longbeachae. L. micdadei is found in less than 1% of community-acquired pneumonia [2], L. hackeliae has been isolated from humans only once [16], and L. fallonii has never been reported to cause disease. L. fallonii was originally designated LLAP10 for 'legionella-like amoebal pathogen 10' [17], a term coined by Rowbotham for bacteria that caused legionella-like infections in amoebae, but could not be grown on agar media.
Here we analyze and compare the L. micdadei, L. hackeliae and L. fallonii genomes and compare them with seven previously completely sequenced L. pneumophila (Paris, Philadelphia, Lens, Corby, Alcoy, Lorraine and HL06041035) [7,8,11,14] and one L. longbeachae NSW150 genome sequence [9]. We confirm that the presence of 'eukaryotic-like proteins' (ELPs) is indeed a specific feature of the genus Legionella and extend the knowledge of these proteins further by identifying additional eukaryotic motifs. Analyses of the virulence of the different Legionella species in protozoan and human cells correlated with the genetic content and allowed us to identify specific features of human pathogenic Legionella and to define a core set of 24 type IV secretion system (T4SS) effectors present in the Legionella species examined to date.
Results and discussion L. micdadei, L. hackeliae and L. fallonii show different virulence in amoeba or macrophages Little to nothing is known about the environmental distribution and the virulence of different Legionella species for human cells. Similarly, it is not known why L. pneumophila and L. longbeachae are so predominant in human disease compared with other Legionella species. As a first step to understand these differences we analyzed the capacity of L. micdadei, L. hackeliae and L. fallonii to infect the protozoan species Acanthamoeba castellanii and the human monocytic cell line THP-1. As shown in Figure 1A, L. micdadei replicated in THP-1 cells, similar to L. pneumophila, while L. fallonii and L. hackeliae were unable to replicate in these cells, although they are phagocytosed efficiently as seen from the higher numbers entering the cells after one hour of infection ( Figure 1A). In contrast, L. fallonii was able to replicate in A. castellanii ( Figure 1B). However, neither L. hackeliae nor L. micdadei replicated in this amoeba. Thus, additional experiments are necessary to analyze whether A. castellani is their environmental host or not ( Figure 1B). Similar results have been obtained using Dictyostelium discoideum as a host where L. micdadei can replicate in this model amoeba but L. hackeliae cannot [6]. In contrast, it was reported that L. micdadei is able to replicate in A. castellani [6,18]. Puzzled by these contradicting results we further analyzed the infection capacity of L. micdadei. Our infection assays had been carried out at 20°C whereas Hägele and colleagues [6] performed their infections at 30°C. We thought that the different results might be due to the different temperatures used. We thus carried out infection assays at 30°C and also used amoeba plate testing [19] at 37°C and 30°C ( Figure 1C). Indeed, L. micdadei was able to replicate in A. castellani at 37°C and also at 30°C, although to a lesser extent compared with L. pneumophila (Additional file 1). This suggested that the replication capacity of L. micdadei in A. castellanii is temperature dependent.
Taken together the replication capacity of the different Legionella species in amoeba and human cells differed in a way similar to the epidemiological data for these species. This suggests that common as well as species-specific mechanisms might be involved in Legionella infection and replication in human cells.
The Legionella genomes have similar genomic features but very different genome content At approximately 3.5 Mb, the genome sizes of L. hackeliae and L. micdadei are similar to that of L. pneumophila whereas that of L. fallonii is similar to that of L. longbeachae at approximately 4 Mb ( Table 1). The GC content is highly homogenous (approximately 39%) and the gene order is relatively well conserved. Apart from L. micdadei, each strain contained one or two plasmids between 14 and 238 kb in size (Table 1). When five different L. pneumophila genomes were compared the pan-genome comprised 2,957 genes, the core-genome of the species L. pneumophila contained 1,979 genes and the calculation of the rarefraction curves indicated that L. pneumophila has an open pan-genome [11]. This held true when we analyzed 11 Legionella genomes here (seven L. pneumophila strains and one strain each of L. longbeachae, L. micdadei, L. hackeliae and L. fallonii); the Legionella pan-genome increased considerably to 9,194 genes and the core genome was 1,388 genes ( Figure 2A) or 1,415 genes when comparing one strain of each sequenced species (L. pneumophila Paris as representative) ( Figure 2B). Thus, the core genome of Legionella represents only about 15% of the pan-genome, indicating that the Legionella accessory genome is large. The complete annotation of these three newly sequenced genomes is available in the LegionellaScope database [20] and at the Institut Pasteur, LegioList [21].
To establish a whole genome-based phylogeny of these Legionella species we used either 29 housekeeping genes or 816 orthologous genes shared among the 11 Legionella strains analyzed. Coxiella burnetii was used as outgroup. Phylogenetic reconstructions using either the nucleotide or the amino acid sequences gave the same tree topology for the different species. In contrast, the tree topology of the L. pneumophila strains was different depending on the data set or the phylogenetic method used, probably due to the high recombination rate of this species [12,22]. Our phylogenetic analyses showed that L. pneumophila, L. fallonii and L. longbeachae group together, with L. fallonii being phylogenetically the closest to L. pneumophila. L. micdadei and L. hackeliae formed a second cluster ( Figure 3). Except for the place of L. fallonii, this is in agreement with previous phylogenies of the genus Legionella [23,24]. In previous work L. pneumophila was described as phylogenetically closer to L. longbeachae than to L. fallonii [25] or L. fallonii closer to L. longbeachae than to L. pneumophila [26]. However, these studies are based on 16S RNA sequences and bootstrap values associated with the corresponding nodes to evaluate its statistical support are not provided.
In conclusion, the general features of the Legionella genomes are very similar but each Legionella species has a distinctive genomic content with about 60% of genes being species-specific. Interestingly, human pathogenic and nonpathogenic species were mixed in the phylogeny, which indicates that virulent traits favoring human infection have been acquired independently during the evolution of the genus.
Type II and IVB secretion systems are part of the core genome of Legionella Like in other bacterial genera, the core genome of Legionella contains the genes encoding fundamental castellanii plate test at 37°C and 30°C. L. pneumophila strain Paris wild type (wt) and ΔdotA were used as positive and negative controls, respectively. Intracellular replication for each strain was determined by recording the number of colony-forming units (CFU) through plating on BCYE agar. Blue, L. pneumophila strain Paris; red, ΔdotA; orange, L. micdadei; violet, L. hackeliae; green, L. fallonii (LLAP10). Results are expressed as Log10 ratio CFU Tn/T0 and each point represents the mean ± standard deviation of two or three independent experiments. The error bars represent the standard deviation, but some were too small to clearly appear in the figure.
metabolic pathways and the ribosomal machinery. In addition, the Dot/Icm type IVB secretion system (T4BSS) as well as the Lsp type II secretion system (T2SS), both indispensable for intracellular replication, also belong to the core genome of this genus. The chromosomal organization of the Dot/Icm and the Lsp secretion system is also conserved, except for the genes icmD and icmC, which are duplicated in L. fallonii. Interestingly, the degree of conservation of the different Dot/Icm proteins is very variable, ranging from >90% for DotB to proteins without any homology such as IcmR. Surprisingly, DotA, an integral inner membrane protein [27] indispensable for intracellular growth [28], is one of the least conserved proteins of the Dot/Icm T4SS (Additional file 2). Unexpectedly, the sequenced L. hackeliae strain (ATCC35250) had a stop codon in the gene coding for DotA, splitting it into 984 and 2,040 nucleotide fragments. Resequencing of the dotA gene confirmed the presence of the stop codon. As this strain was not able to replicate in A. castellanii, we thought that this might be due to the mutated dotA gene leading to a nonfunctional T4SS. To verify if this mutation was specific for the sequenced strain, we analyzed the dotA gene in a second L. hackeliae strain (ATCC35999). In this strain  Another particularity of the Dot/Icm system is the icmR gene. Indeed, similar to what was reported for L. hackeliae and L. micdadei where icmR was replaced by a nonhomologous gene with functional equivalence [31,32], a gene encoding a protein with no similarity to any previously described protein is present in the position of icmR in L. fallonii, possibly serving as a functional equivalent of icmR of L. pneumophila. Other variable genes include icmX and icmG. IcmG has been described as a component that interacts with the effector proteins [33], which may explain the high variability in different species. In contrast, the components dotB, icmS, icmW and icmP are highly conserved. Indeed, these four genes can functionally replace their homologues in C. burnetii [34].
The L. micdadei, L. hackeliae and L. fallonii genomes encode surprising functions L. fallonii is able to synthesize cellulose Enzymes degrading cellulose have been described in L. longbeachae and were also found in L. fallonii. However, in addition the L. fallonii genome encodes a complete machinery for the synthesis of cellulose ( Figure 4A). Although the bacterial need for cellulose may be surprising, cellulose has been reported as a common component of biofilms of several bacterial species such as Salmonella enterica or Escherichia coli [35]. The bacterial genes for cellulose synthesis are called bcsABZC. In S. enterica and E. coli a second operon necessary for cellulose biosynthesis named bcsEFG is present [35,36]. Both clusters (from lfa3354 to lfa3363 and lfa2987 to lfa2988) are present in L. fallonii, although with some differences in organization ( Figure 4A). To analyze whether L. fallonii is able to synthesize cellulose, we used agar plates containing calcofluor, which binds cellulose and leads to fluorescence under UV radiation. Indeed, L. fallonii showed strong fluorescence under long-wave UV light, in contrast to L. pneumophila ( Figure 4B), demonstrating cellulose biosynthesis in the genus Legionella for the first time. A blast search identified genes homologous to the L. fallonii cellulose operon (except bcsE and bcsF) also in the draft genome sequences of L. anisa and L. dumoffii ( Figure 4A). This suggests that several Legionella species have the capacity to synthesize cellulose.
L. fallonii possesses genes encoding hopanoid biosynthesis and antibiotic resistance L. fallonii encodes genes for hopanoid biosynthesis currently not found in any other Legionella species. About 10% of all sequenced bacteria contain genes for hopanoid synthesis, in particular cyanobacteria, acetobacter, streptomycetes, methylotrophs and purple non-sulfur bacteria. Hopanoids have been proposed to enhance membrane stability and to decrease membrane permeability [37], similar to sterols in eukaryotic cell membranes [38]. In Burkholderia cenocepacia these genes are involved in sensitivity to low pH, detergent and antibiotics, and are related to motility [39]. In Streptomyces coelicolor, this cluster has been well studied. Although not all genes of the S. coelicolor cluster are conserved in L. fallonii (Additional file 3), to date all bacteria carrying the gene for squelene-hopene-cyclase produce hopanoids [39]. As L. fallonii also carries this gene, we expect this species is able to synthesize hopanoids, although their function in this species remains unknown.
Another peculiarity of L. fallonii is that it contains several antibiotic resistance genes not previously described in Legionella, including one encoding a chloramphenicol acetyltransferase (lfa0269) that is predicted to catalyze the acetyl-CoA-dependent acetylation of chloramphenicol. Furthermore, we identified a gene likely involved in erythromycin resistance, ereA (lfa1884) that is present also in L. drancourtii and L. dumoffii. This gene is located in gene clusters related to DNA mobility, such as integrases or prophage-related genes, and are rich in ELPs and repeats. These features indicate that these regions are putative genomic islands (Additional file 4).
L. hackeliae and L. fallonii encode chitin deacetylase activity L. hackeliae and L. fallonii each contain a different gene coding for a chitin deacetylase (lha3256/lfa0697), an enzyme involved in deacetylation of chitin. An in vitro test described by Vadake [40] suggests that L. fallonii does have chitin deacetylase activity whereas it was not possible to demonstrate this clearly for L. hackeliae (Additional file 5). Chitin, a homopolymer of N-acetylglucosamine, is one of the most abundant polymers in the Earth's biomass, especially in marine environments. Interestingly it is also a component of the cyst wall of Entamoeba invadens, and enzymes responsible for chitin synthesis have been found in Entamoeba genomes [41].
The presence of chitin or chitin synthases has not been described in other protozoan genomes, but very few genomes of this group have been sequenced yet. Thus, chitin may be a common component of protozoa that are able to encyst. Although the other Legionella genomes analyzed here do not encode chitin deacetylase activity, all Legionella genomes encode chitinases. Chitinases are chitin-degrading enzymes leading to low molecular weight chito-oligomers whereas chitin decetylase degrades chitin to chitosan. Both products are of interest for industry and there is growing interest in organisms that produce chitosan. Legionella may be a new possible source of chitosan production.
L. micdadei contains the first putative complete prophage identified in a Legionella genome The analysis of the unique genes from L. micdadei identified a specific region encoding 73 proteins, at least 16 of which are phage-associated proteins that represent a putative complete prophage (Additional file 6). This region contains genes encoding the phage capsid tail and replication proteins. Complete prophages have never been described in Legionella despite the frequent presence of phage-related proteins scattered in their genomes. Most attempts to isolate prophages that exclusively infect Legionella have also failed, until recently when two groups isolated Legionella bacteriophages [42,43] from environmental water samples and organs of guinea pigs. Thus, Legionella do have phages, but they seem to be rare.

L. fallonii and L. micdadei contain additional flagella operons
The comparison of the L. pneumophila and L. longbeachae genomes revealed that L. longbeachae does not contain genes allowing flagella biosynthesis [9]. As recognition of flagellin by Naip5 initiates host immune responses that control L. pneumophila infection in certain eukaryotic cells [44,45], the presence or absence of flagella is important for intracellular replication of Legionella. L. hackeliae, L. fallonii and L. micdadei also contain three flagella operons homologous to those described in L. pneumophila ( Figure S5A-C in Additional file 7). Interestingly L. fallonii and L. micdadei encode a fourth region not previously described in any sequenced Legionella species that might also code for flagella ( Figure 5).

A highly dynamic mobilome characterizes the Legionella genomes
Genomic elements like plasmids, genomic islands or transposons constitute the mobilome of a genome. All Legionella species analyzed contain many of these mobile elements. For example, L. hackeliae possesses a plasmid of 129.88 kb whereas L. fallonii (LLAP10) contains two plasmids of 238.76 kb and 14.57 kb, respectively (Table 1). Furthermore, the plasmid present in L. hackeliae is identical to the L. pneumophila strain Paris plasmid (100% nucleotide identity over the entire length except for two transposases in the strain Paris plasmid; Additional file 8). This suggests that this plasmid has recently moved horizontally between both species, which is a new example of the high rate of gene transfer among Legionella genomes [46,47].
In addition to the plasmids identified and their evident exchange among strains and species, a hallmark of the Legionella mobilome is the presence of many different type IVA secretion system-encoding regions in the plasmids as well as in genomic island-like regions on the chromosome. Interestingly, these regions often encode tra-like genes with considerable homology among the different strains. However, each new strain analyzed contained new regions, underlining the high diversity of these systems in the Legionella genomes. Predominant are F-type and P-type IVA systems that code for conjugative pili that allow mating. F-type IVA secretion systems are present on all L. pneumophila plasmids, the L. hackeliae plasmid, the 238 kb L. fallonii plasmid (two systems) and on the chromosomes of L. pneumophila strain Philadelphia, L. longbeachae and L. fallonii (Additional file 9). Each encodes a homologue of the global regulator CsrA, named LvrC, which when present in the chromosome also encodes the lvrRAB gene cluster. This was recently described as being involved in the regulation of excision of the ICE Trb1 of L. pneumophila strain Corby [48]. Thus, conjugative exchange of DNA has an important role in Legionella and is one key factor enabling Legionella to rapidly adapt to changing conditions. The mobility and horizontal transfer of these different regions are further shown when studying the distribution of these systems. For example, the lvh cluster, a type IVA system involved in virulence under conditions mimicking the spread of Legionnaires' disease from environmental niches [49], is also present in L. micdadei, in one of the two completely sequenced L. longbeachae strains and in five of the completely sequenced L. pneumophila strains ( Table 2). In addition, the so-called GI-T4SS recently described in strain L. pneumophila 130b [13], and first recognized in Haemophilus influenzae as a T4SS involved in propagation of genomic islands [50], is believed to play an important role in the evolution and adaptation of Legionella [51]. GI-T4SS clusters were found to be conserved in L. pneumophila, with two clusters each in strains Corby, Paris, 130b and HL06041035, and one in each of Alcoy, Philadelphia, Lens and Lorraine [51], as well as in strains of L. longbeachae, L. hackeliae, L. micdadei and L. fallonii (Table 2). Thus, a heterogeneous distribution among species and strains testifies to the continuous exchange of these elements among Legionella, contributing to the plasticity and dynamic nature of their genomes.
L. micdadei strains from different geographical regions are highly similar except for their mobilome To investigate the genomic diversity of the species L. micdadei, we determined the draft genome sequence of a clinical isolate obtained from the Microbiological Diagnostic Unit Public Health Laboratory (MDU), Australia and compared it with the completely sequenced strain L. micdadei ATCC 33218. The genome size and GC content of the two L. micdadei strains were highly similar ( Figure 6). The main differences between the two L. micdadei strains were mobile genetic elements. Furthermore, the number of SNPs (1,985 SNPs) was very low, similar to serogroup 1 strains of L. longbeachae (1,611 SNPs) [9]. This is strikingly different to L. pneumophila where two different strains may contain more than 30 000 SNPs. This suggests that L. micdadei and L. longbeachae evolved more recently compared to the L. pneumophila. Three large regions of the L. micdadei ATCC 33218 genome are absent from the Australian isolate ( Figure 6). One is a genomic island encoding a GI-T4SS (36 kb), one is the predicted prophage we identified in this study, and another is a smaller cluster of approximately 9 kb that is flanked by three tRNA genes and that contains phage-related genes and a gene associated with abortive infection system ( Figure 6). Similarly, in the Australian isolate a cluster absent from the completely sequenced L. micdadei strain corresponds to a P-type IVA secretion system. Interestingly, the Lvh region, encoding a T4ASS that is highly conserved among all strains and species analyzed to date, is divergent in the two L. micdadei strains with a high number of SNPs (Additional file 10). Thus, the main genetic differences between these two closely related L. micdadei strains are mobile genetic elements, further underlining the large extent of horizontal gene transfer that is present in the genus Legionella.
The core set of Dot/Icm effectors is small with only 24 conserved substrates L. pneumophila encodes over 300 proteins that are translocated into the host cell by the Dot/Icm T4SS (Additional file 11). Their conservation is high among different L. pneumophila strains, as 77% of these substrates are present in all L. pneumophila strains sequenced to date. Interestingly, when the Dot/Icm substrates of L. pneumophila and L. longbeachae are compared, only 35% (101) are present in both species [9]. Interestingly, the L. longbeachae and L. pneumophila genomes contain the highest number of common substrates, although L. fallonii is phylogenetically closer to L. pneumophila than to L. longbeachae ( Figure 3). When investigating the presence of these substrates in five Legionella species by adding the L. hackeliae, L. micdadei and L. fallonii genomes, this revealed that their conservation is very low (Figure 3). With 33 conserved substrates, the lowest number is shared between L. micdadei and L. pneumophila. This result suggests that the shared substrates might relate to similar environmental niches or virulence properties (L. pneumophila and L. longbeachae) than to a closer phylogenetic relationship.
The Dot/Icm substrates conserved in all Legionella species are probably those indispensable for intracellular L. longbeachae NSW-150 replication and are important players in host-pathogen interactions. Most surprisingly, only 24 of the 300 described substrates of L. pneumophila are present in all five Legionella species and most of these are of yet unknown function (Table 3). However, a third of the conserved substrates contain eukaryotic motifs like ankyrin or Sel-1 domains or TPR repeats. Others were previously defined as ELPs, such as the sphingomyelinase-like phosphodiesterase. Among the substrates that have been investigated further are VipF, which causes growth defects in S. cerevisae, and several of the ankyrin repeat motif proteins. VipF inhibits lysosomal protein trafficking [52] and AnkH was shown to play a role in intracellular replication of L. pneumophila in macrophages and protozoa and in intrapulmonary proliferation in mice [53]. The function of MavBFNQ and RavC is not known, but they have been recovered in screens for vacuolar localization and have been shown to co-localize with SidC at the L. pneumophila vacuole [54]. SdhA, a L. pneumophila effector that is necessary for full virulence of this species, is a particular case. It is present in all Legionella analyzed but the similarity with L. longbeachae is small and thus below the cutoff established for our orthologous search (at least 65% of the length of the compared protein). However, given that homologues with a significant similarity are present in all species in synteny (except for L. hackeliae), and coiled-coil motifs are detected in all, SdhA was also defined as a core effector. Moreover, SdhA has been shown to be necessary for infection of mice and in Galleria mellonella [55,56]. Surprisingly, the effector SidJ is not part of the core set of Legionella substrates, although its deletion led to a strong replication defect in eukaryotic cells. However, SidJ is present in L. pneumophila and L. longbeachae, the major human pathogens.
Interestingly, the growth defect of strains lacking SdhA and SidJ seems more important in mice and human macrophages than in amoeba. Replication of the sdhA mutant is severely impaired in mouse bone marrow-derived macrophages but less in the amoeba Dictyostelium discoideum [56]. Similarly, a ΔsidJ strain shows significant growth defects in both macrophages and amoebae, but replication in macrophages is affected from the start of the infection, whereas the growth defect in amoebae is evident only after 72 h of infection and was less pronounced [57]. These data may suggest that effectors important in human infection are not necessarily essential in the protozoan hosts and thus certain effectors might be important for human infection even though no growth defect in protozoan infection is detectable.
Eukaryotic-like proteins are a specific feature of the genus Legionella One feature shared by many of the substrates of the Dot/ Icm secretion system is the presence of eukaryotic motifs (EMs). Indeed, of 55 proteins of L. pneumophila Philadelphia encoding EMs, 45 (82%) are confirmed substrates of the Dot/Icm secretion system (Additional file 12). Thus, we searched for proteins containing EMs in all sequenced genomes. In the five Legionella species we identified 218 proteins with eukaryotic domains (Additional file 13). The genomes of L. longbeachae and L. fallonii contain nearly twice as many proteins with EMs as the other genomes, probably due to their larger genome size. The ankyrin motif is the most frequent one, followed by long coiledcoil domains. Some EMs that were described remain specific for L. longbeachae, such as the PPR repeats, PAM2 domain or the phosphatidylinositol-4-phosphate 5-kinase, indicating that they are probably related to its particular habitat in soil [9]. In contrast, proteins with tubulin- Figure 6 Genome comparison of two L. micdadei strains. The complete genome sequences of the two L. micdadei strains included in this study were aligned using the software Mauve. The two strains align perfectly with the exception of three mobile genetic elements specifically present in strain L. micdadei ATCC33218 and one specifically present in the Victorian isolate. The specific regions of each genome are indicated. The 'Lvh region' is indicated, as this region is, with a high number of SNPs, quite divergent between the two isolates.
tyrosine ligase domains (LLo2200), probably involved in the posttranslational modification of tubulin [58], are absent only from L. pneumophila. With the aim to analyze whether additional eukaryotic motifs not yet identified are present in the Legionella genomes, we developed a strategy allowing for a comprehensive scan of all genomes. First we searched the Interpro database for all motifs, which occur in at least 85% of proteins from eukaryotic genomes and only 15% or less in proteins from prokaryotic genomes. Using this criterion, we obtained 8,329 motifs that were considered as eukaryotic (see Materials and methods). All predicted Legionella proteins were scanned for these motifs. This approach allowed us to identify 10 EMs not described before in Legionella, including thaumatin, RhoGTPase and DM9 domains (Table 4). Interestingly, thaumatin-like proteins accumulate in plants in response to infection by pathogens and possess antifungal activity [59,60] and a Drosophila DM9-containing protein is strongly up-regulated after infection of Drosphila larvae by Pseudomonas species [61]. Many of these new EMs are only present in the newly sequenced genomes, such as synaptobrevin, an intrinsic membrane protein of small synaptic vesicles [62] or the clathrin/coatomer adaptinelike domain that is associated with transport between the endoplasmic reticulum and Golgi [63]. Given their function in eukaryotic organisms, these protein domains might indeed be important in host-pathogen interactions.

Many eukaryotic proteins are indeed transferred horizontally from eukaryotes
Not all proteins we defined as ELPs possess EMs, but certain ones are also considered eukaryotic-like as they show a high homology to eukaryotic proteins over their whole length. One of the best known examples of this type of ELP is the sphingosine-1-phosphate lyase (encoded by the gene lpp2128), an enzyme that in eukaryotes catalyzes the irreversible cleavage of sphingosine-1-phosphate, and that has most likely been transferred horizontally from eukaryotes [47,64,65]. With the aim to detect proteins with higher similarity to eukaryotic proteins than to prokaryotic ones and for which we can suggest a eukaryotic origin through phylogenetic analysis, we have developed a pipeline that automatically extracts those proteins from the Legionella pan-genome with high similarity to eukaryotic proteins (for details see Materials and methods). Using this pipeline we identified 465 proteins as putative ELPs. For each of these proteins we constructed a phylogenetic tree that was curated and analyzed manually. However, for many of the ELPs a phylogenetic reconstruction did not allow clear demonstration of eukaryotic origin. Some aligned too poorly with their eukaryotic homologues or on just a small domain. This might be due to the fact that genomes of ciliated protozoa and amoeba, the known hosts of Legionella from which these ELPs are most likely acquired, are underrepresented in current databases. However, for 40 of the 465 proteins that are suggested to be of eukaryotic origin, the phylogenetic reconstruction clearly showed that they had been acquired by Legionella through horizontal gene transfer from eukaryotes (Table 5; Figure  S9A-C in Additional file 14). Among these proteins 27 had not been described before and 15 were identified in the newly sequenced species. A clear case of horizontal gene transfer from eukaryotes is GamA (Lpp0489), a glucoamylase that allows Legionella to degrade glycogen during intracellular replication in A. castellanii [66]. In addition to already characterized proteins, we identified promising candidates for host-pathogen interactions in this study -for example, a L. longbeachae protein containing a tubulin-tyrosine ligase domain (Llo2200; Figure S9A in Additional file 14), a motif involved in addition of a carboxy-terminal tyrosine to α-tubulin as part of a tyrosination-detyrosination cycle that is present in most eukaryotic cells. This tyrosination process regulates the recruitment of microtubule-interacting proteins [67]. It is thus tempting to assume that Legionella is able to interfere with or to modulate the recruitment  of microtubule-interacting proteins in the host. Another example is the serine carboxypeptidase S28 family protein (Llo0042/Lfa0022; Figure 7). These proteins have been identified exclusively in eukaryotes and are active at low pH, suggesting a function in the phagosome [68]. Taken together, each Legionella genome contains many different ELPs and proteins carrying eukaryotic domains that help Legionella to establish its intracellular niche. Some of these proteins are specific to one or other Legionella species but most are present in all of them, although these proteins are rarely real orthologues. This suggests that the acquisition of these proteins is important for Legionella to manipulate the host but that their horizontal acquisition has taken place on multiple occasions.

Linking virulence properties and gene content
When using THP-1 cells as a model for infection of human macrophages, not all Legionella species were able to infect and replicate ( Figure 1A). These results correlated with the epidemiology of legionellosis where only certain Legionella species are isolated from human disease. With the aim of identifying the genetic bases conferring these differences, we searched for genes that were present in the strains that cause disease but absent in the ones that had not been isolated from humans. This comparative analysis showed that L. pneumophila, L. longbeachae and L. micdadei share 40 genes that are not present in any of the other species. Among those we identified the hyp operon (hypABFCDE -lpg2171-75), necessary for hydrogenase activity in E. coli and the cyanobacterium Synechocystis [69]. Legionella has additional downstream genes encoding for hydrogenases that are unique to these three species. This region is flanked by tRNA genes in L. micdadei and L. longbeachae, suggesting its acquisition by horizontal gene transfer.
Furthermore, a gene encoding a truncated hemoglobin (lpp2601) of group I called trHbN was identified as specific to the human pathogenic strains. Truncated hemoglobins are a family of small oxygen-binding heme proteins [70] that are ubiquitous in plants and present in many pathogenic bacteria such as Mycobacterium tuberculosis. Mycbacteria missing trHbNs are severely impaired for nitric oxide detoxification [71], and the expression of this gene is required for M. tuberculosis during macrophage infection [72]. The proteins of M. tuberculosis and L. pneumophila share 30% identity and the important TrHbN residues are conserved in both, indicating a similar biochemical function. Furthermore, the M. tuberculosis trHbN shows 40% identity to its eukaryotic homologue in Tetrahymena thermophila and the Legionella protein 44% to the T. thermophila and 46% to the Paramecium tetraurelia protein. However, according to an in-depth phylogenetic analyses of truncated hemoglobins in prokaryotic and eukaryotic organisms, it seems that trHbNs are of prokaryotic origin and might have been transferred to eukaryotes [73]. Interestingly, the Lvh system is not part of the genes unique to L. pneumophila, L. longbeachae and L. micdadei as not all L. pneumophila strains contain it, but it is uniquely present only in these three species. Finally, of the more than 300 proteins described as translocated by the Dot/ Icm secretion system, only two, CegC4 (lpp2150/ lpg2200) and Lem25 (lpp2487/lpg2422), are exclusive to the three species found in human disease, but their function is not known yet.
Comparing L. pneumophila and L. longbeachae, the two species responsible for over 95% of human infections, to all other Legionella species, showed that 124 genes are specific to these human pathogenic Legionella. Among them are 38 substrates of the Dot/Icm secretion system, including RalF (lpp1932/lpg1950), SidJ (lpp2094/lpg2155), SidI (lpp2572/lpg2504), SdeC (lpp2092/lpg2153), SidE (lpp2572/lpg2504), SdcA (lpp2578/lpg2510) and CegC7 (lpp0286/lpg0227). In addition to the secreted substrates, iron availability seems to be important for the human pathogens as among the specific proteins several are related to iron scavenging or iron storage. These are homologues of PvcA and PvcB (lpp0236-lpp0237), the siderophore pyoverdine that is involved in virulence and biofilm formation in the cystic fibrosis pathogen Pseudomonas aeuroginosa [74]. In Legionella these genes are highly expressed in sessile cells, suggesting their involvement in sessile growth [75]. Furthermore, a bacterioferritin (lpp2460) that is present also in L. micdadei but highly divergent is specific for the human pathogenic Legionella.
Bacterioferritin plays a role in iron storage and is involved in protecting cellular components from oxidative damage, thereby playing a role in oxidative stress relief [76,77]. Furthermore, a gene coding for a homologue of the Yersinia pestis plasminogen activator (lpp2452) that was shown to create transient plasmin activity [78] and the phospholipase C (lpp1411) implicated in host killing in a G. mellonella model [79] are specific to L. pneumophila and L. longbeachae.

Conclusions
The first comprehensive analyses of five species of the genus Legionella and the comparison of the genomes of human disease-related strains with non-disease-related strains have provided new insights into the genomic specificities related to adaptation and host-pathogen interactions of this fascinating intracellular bacterium and have identified specific features of the major human pathogenic Legionella. Highly dynamic genomes that evolve through frequent horizontal gene transfer, mediated by many and diverse T4SSs and acquisition of different eukaryotic proteins and protein domains at multiple times and stages of their evolution that allow host subversion are a hallmark of this amoeba-associated bacterial genus. The major human-related Legionella species, L. pneumophila and L. longbeachae, contain a set of genes that seems to increase their successful infection of mammalian cells. The key to their success may be a better capacity to subvert host functions to establish a protective niche for intracellular replication due to a specific set of secreted effectors and a higher ability to acquire iron and to resist oxidative damage. The analysis of additional Legionella genomes and other intracellular pathogens may allow the future definition of the major common strategies used by intracellular pathogens to cause disease and to understand how environmental pathogens may evolve to become human pathogens.

Annotation and genome comparison
The newly sequenced genomes of L. fallonii, L. hackeliae and L. micdadei were integrated into the MicroScope platform [81] to perform automatic and expert annotation of the genes, and comparative analysis with the already sequenced and integrated L. pneumophila strains. Micros-Scope annotation is based on a number of integrated bioinformatic tools: Blast on UniProt and specialized genomic data, InterPro, COG, PRIAM, synteny group computation using the complete bacterial genomes available at NCBI RefSeq, and so on (for more details see [82]). Orthologous groups were established using the program PanOCT [83] with the following parameters: e-value 1e-5, percent identity ≥30, and length of match ≥65. The programs Easyfig and BRIG [84,85] were used for graphical representation of genome regions compared using BLAST. MAUVE [86] was used for aligning and comparing the L. micdadei genomes.

A. castellanii and THP1 infection assays
In For THP-1 infection, cells were seeded into 24-well tissue culture trays (Falcon, BD lab ware, Altrincham, Manchester, United Kingdom, England) at a density of 1.5 × 10 5 cells/well and pretreated with 10 −8 M phorbol 12-myristate 13-acetate (PMA) for 72 h in 5% CO 2 at 37°C to induce differentiation into macrophage-like adherent cells. Stationary phase Legionella were resuspended in RPMI 1640 serum free medium and added to THP-1 cell monolayers at an MOI of 10. After 1 h of incubation cells were treated with 100 μg ml −1 gentamycin for 1 h to kill extracellular bacteria. Infected cells were then washed with phosphate-buffered saline (PBS) before incubation with serum-free medium. At 24, 48 and 72 h THP-1 cells were lysed with 0.1% TritonX-100. The amount of Legionella was monitored by counting the number of CFU determined by plating on BCYE agar. The infections were carried out in triplicate.

Cyclase translocation assay
The vector containing RalF-CyaA [29] was transformed into L. micdadei, L. hackeliae and L. fallonii and strain Paris wild type and its isogenic ΔdotA::Km mutant were used as positive and negative controls. Transformant strains were used to infect THP-1 cells previously plated at 1 × 10 5 cells/well in 24-well tissue culture dishes and pre-treated with 10 −8 M PMA. After 1 h and 30 minutes following infection cells were washed three times with cold PBS and lysed in 50 mM HCl, 0.1% Triton X-100. Lysates were boiled 5 minutes and neutralized with 0.5 M NaOH. We then added 95% cold ethanol and samples were spun for 5 minutes at maximum speed in a microcentrifuge. Supernatants were transferred in new 1.5 ml tubes and vacuum dried, and cAMP concentrations were measured using the cAMP Biotrak Enzyme immunoassay System (Amersham, United Kingdom, England). Each value was calculated as means of two independent infections ± standard deviations.

Amoebae plate test
Samples of suspended amoeba were applied to BCYE agar plates as described previously [19]. Stationary-phase bacterial cultures (OD600 > 4.5) were adjusted to an identical OD600 (2.5), series of 10-fold dilutions in sterile H 2 O were prepared and 3 μl of each dilution were spotted onto CYE plates both with amoeba and without amoeba (control plates) and incubated for 3 to 5 days at 30°C or 37°C.

Detection of new eukaryotic motifs in Legionella proteins
To better define the term 'eukaryotic motifs' we searched for the already known EMs in all proteins present in the Pfam database and calculated their occurrence in eukaryotic proteins or prokaryotic proteins. The previously described EMs in Legionella showed an occurrence of about 99% in eukaryotic proteins and only 1% in prokaryotic ones, with the ankyrin repeats being the less restricted to eukaryotic proteins (85%). The only exception is Sel-1 domains, which were considered as EMs. Sel-1 domains have now been shown to be highly present also in prokaryotes. However, since this domain is present in many substrates of the Dot/Icm system and it was shown to be implicated in host-pathogen interactions [87], it was taken into account. Based on the frequencies of the typical EMs present in Legionella we searched the Interpro database for all motifs that occur in eukaryotes at least to 85%. Using this criterion we obtained 8,329 motifs that can be considered as eukaryotic. These motifs were searched in all proteins predicted in the different Legionella genomes. This approach identified 10 eukaryotic motifs previously not described in Legionella proteins.

Detection of genes transferred from eukaryotes to Legionella
To detect genes with putative eukaryotic origin we developed a pipeline based on several step filters. This pipeline was applied to one protein of each of the orthologous groups of the pan-proteome of the five studied species to avoid redundancy in the detection process with proteins of the same orthologous group. The first step consisted of discarding the protein families without significant similarity to eukaryotic sequences. This was achieved by a homology search using Blastp with an evalue cutoff of ≤10e −4 and a BLOSUM62 matrix with a representative protein of each group of orthologous families of the Legionella pan-genome against a database containing 83 genomes representative of all major eukaryotic phyla and certain viruses. In particular, members of Amoebozoa and other protist lineages that may be hosts for Legionella were included in this database. The results of the first filter led to the recovery of 2,669 proteins of the Legionella pan-genome with significant homology to eukaryotic sequences in the database. Then, among these 2,669 protein families those that have closer homologues in bacteria were discarded by searching for homologues against a database containing both eukaryotic and prokaryotic sequences using the same criteria. Only those that had at least a hit against a eukaryotic sequence among the first 25 hits were further selected. This step led to the selection of 465 protein families of the Legionella pan-genome representing ELP candidates. Finally, we carried out automatic phylogenetic reconstruction of these 465 proteins and their bacterial and eukaryotic homologues. The different steps of the pipeline were: (1) for each selected putative ELP the corresponding orthologues in other Legionella species analyzed where added if present; (2) each group of homologous sequences was aligned with MUSCLE [88]; (3) unambiguously aligned positions were automatically selected using the multiple alignment trimming program BMGE with low stringency parameters [89]; (4) preliminary maximum likelihood trees were obtained using FastTree [90]. We applied a strict filter to select only very likely ELPs. Then each of the 465 trees was manually inspected to select those where the Legionella sequences were branching within eukaryotes or were closer to eukaryotic sequences than to prokaryotic ones. This allowed identification of 40 Legionella proteins that aligned well with their eukaryotic homologues. For those having a sufficient number of eukaryotic homologues and a sufficient number of positions that could be selected after trimming, we proceeded to phylogenetic analysis by maximum likelihood using LG +4 gamma as the evolutionary model. Then, we selected those trees where the Legionella sequences were branching within eukaryotes or were closer to eukaryotic sequences than to prokaryotes. Finally, in order to verify the eventual existence of closer bacterial homologues or additional eukaryotic homologues from representatives not present in our local database, we performed a Blast on the non-redundant database at the NCBI. Alignments were obtained and trimmed, and trees reconstructed as described above.

Test for chitinase degradating activity
According to Vadake [40], Whatman filter paper strips were cut to 5 cm × 1 cm. These strips were immersed and air-dried in a solution of p-nitroacetanilide (5 g in 100 ml of ethanol 100%). The procedure was repeated three times to impregnate well the strips with pnitroacetanilide. L. fallonii and L. pneumophila (used as negative control) were grown in liquid medium for 24 h and 2 ml of these cultures were transferred to a new sterile tube containing 2 ml of fresh liquid media and the diagnostic strips. These cultures were grown for 2 days at 30°C for L. fallonii and 37°C for L. pneumophila. After 2 days the development of yellow color on the strip indicated the presence of deacetylase in the corresponding bacterial culture.

Cellulose detection assays
To visualize the production of cellulose, plates containing Legionella BCYE medium supplemented with calcofluor (5%; fluorescent brightener 28; Sigma-Aldrich, Oakville, Ontario, Canada) were prepared. Drops of 5 μl of liquid media containing L. fallonii grown for 72 h were spread on the plates and incubated at 30°C for 48 h. The same procedure was carried out for L. pneumophila at 37°C as negative control. After incubation plates were visualized under a UV light source.

Additional files
Additional file 1: Figure S1. The ability of L. micdadei to replicate intracellularly in A. castellanii is dependent on the temperature. L. pneumophila strain Paris wild type (wt) and ΔdotA were used as positive and negative controls, respectively. Intracellular replication of L. micdadei at 30°C was determined by recording the number of colony-forming units (CFU) through plating on BCYE agar. Blue, wild-type L. pneumophila strain Paris; red, ΔdotA; orange, L. micdadei. Results are expressed as Log10 ratio CFU Tn/T0 and each point represents the mean ± standard deviation of two independent experiments. The error bars represent standard deviation, but some were too small to clearly appear in the figure.
Additional file 2: Figure S2. The Dot/Icm-encoding genes are present in all Legionella species with variable homology. Graphical representation of Blastp comparisons of the genes encoding the Dot/Icm T4BSS in all strains/species used in this study. L. pneumophila strain Philadelphia genes were taken as query. Each color ring represents a species/strain, and each segment in the ring a different gene. The intensity of the color indicates the percentage of amino acid identity. The percentages of identity and their correlation with the color intensity is given in the left panel. Gene names are given in the outside circle. Gene names in red indicate 100% amino acid identity.