|Home | About | Journals | Submit | Contact Us | Français|
Conceived and designed the experiments: MAK CG. Performed the experiments: MAK AP DA MAN CD PT HA NT CP CG. Analyzed the data: MAK NK AP DA CCDS DEL CP GVD MT AWE CG. Contributed reagents/materials/analysis tools: CCDS DEL CP GVD MT AWE CG. Wrote the paper: MAK NK AWE CG.
Legionella pneumophila is the primary etiologic agent of legionellosis, a potentially fatal respiratory illness. Amongst the sixteen described L. pneumophila serogroups, a majority of the clinical infections diagnosed using standard methods are serogroup 1 (Sg1). This high clinical prevalence of Sg1 is hypothesized to be linked to environmental specific advantages and/or to increased virulence of strains belonging to Sg1. The genetic determinants for this prevalence remain unknown primarily due to the limited genomic information available for non-Sg1 clinical strains. Through a systematic attempt to culture Legionella from patient respiratory samples, we have previously reported that 34% of all culture confirmed legionellosis cases in Ontario (n=351) are caused by non-Sg1 Legionella. Phylogenetic analysis combining multiple-locus variable number tandem repeat analysis and sequence based typing profiles of all non-Sg1 identified that L. pneumophila clinical strains (n=73) belonging to the two most prevalent molecular types were Sg6. We conducted whole genome sequencing of two strains representative of these sequence types and one distant neighbour. Comparative genomics of the three L. pneumophila Sg6 genomes reported here with published L. pneumophila serogroup 1 genomes identified genetic differences in the O-antigen biosynthetic cluster. Comparative optical mapping analysis between Sg6 and Sg1 further corroborated this finding. We confirmed an altered O-antigen profile of Sg6, and tested its possible effects on growth and replication in in vitro biological models and experimental murine infections. Our data indicates that while clinical Sg1 might not be better suited than Sg6 in colonizing environmental niches, increased bloodstream dissemination through resistance to the alternative pathway of complement mediated killing in the human host may explain its higher prevalence.
Legionellosis is a potentially fatal infectious disease caused by Gram-negative, aerobic bacteria belonging to the genus Legionella , . Among 54 known Legionella species, L. pneumophila is the major cause of outbreaks (91.5%), and was the etiological agent of the first recognized outbreak in 1976 during a convention of the American Legion in Philadelphia , , , . The severity of this disease ranges from a mild respiratory illness to a rapidly fatal pneumonia , . The case fatality rate of legionellosis is between 40–80% in untreated immuno-suppressed patients but can be reduced to 5–30% with appropriate case management . Legionellosis is a major public health concern in industrialized nations , , . From 2000 to 2009, a 217% increase in legionellosis cases was reported in the United States , .
Legionella spp. are found worldwide and can be detected in up to 80% of man-made freshwater sites , , , , , , . More recently, a study showed that Legionella spp can persistently colonize aquifers over several years . In natural and in man-made water systems, L. pneumophila may exist as planktonic cells or as biofilms , . The bacteria can also be isolated from different protozoa in the environment , , . This is an essential step in L. pneumophila’s life cycle, since the bacteria are able to replicate in the environment within the protozoa host , . The human infection results from the inhalation of aerosols contaminated with either infected protozoa or free floating L. pneumophila , . The general model predicts, that upon inhalation, the infection starts with the invasion and replication in macrophage and epithelial cells of the lung , . Once inside the cell, Legionella replicate with in a non-acidified vacuolar environment , . While many genes required for surviving in amoeba are yet to be described, some like the dot/icm genes ,  have shown to be essential for survival in both the amoeba and the human host.
L. pneumophila have been classified into sixteen serogroups based on reactivity to specific monoclonal antibodies , . While these serogroups have been reported worldwide, Sg1 has historically been identified in most clinical cases (84.2%) . The most widely used clinical diagnostic method for all suspected legionellosis cases are rapid, non-invasive, urine detection tests that recognize the presence of soluble antigen . These clinical tests provide high sensitivity for Sg1 infection diagnosis but they are unable to detect most non-Sg1 strains , . Recently, we reported that 34% of culture confirmed cases of legionellosis in Ontario (n=351) were caused by non-Sg1 L. pneumophila species in the last 3 decades . Interestingly such a high percentage of non-Sg1 was also reported in some Scandinavian countries . In a comparative clinical and environmental distribution analysis study, Sg1 was shown to constitute 28.2% of the Legionella spp. bacteria isolated from man-made water systems (n=2,747) but 95.4% of clinical samples (n=259) . This suggests that the clinical prevalence of Sg1 might be independent of predominance in the environment, but rather to fitness or virulence advantages in the human host.
The underlying reasons for the clinical prevalence of Sg1 strains are a subject of active debate. A large, microarray multigenome analysis of 217 L. pneumophila isolates previously suggested that the clinical predominance of Sg1 may be linked to its specific lipopolysaccharide . In this study, thorough population-based molecular epidemiology, we first identified the most clinically prevalent Sg6 strain among non-Sg1 clinical isolates in Ontario. Next, by combining next-generation sequencing with in vitro and in vivo experimental models, we performed a comparative analysis with the closest identified phylogenetic Sg1 (strain Philadelphia). Our results support a model in which increased resistance to serum complement, rather than variation in environmental fitness, explains the relative clinical prevalence of Sg1.
To identify prevalent population based clusters of non-Sg1 L. pneumophila clinical strains in Ontario, we conducted a phylogenetic analysis on 73 clinical isolates recovered between 1980 and 2009 using sequence based typing (SBT) and multiple-locus variable number tandem repeat analysis (MLVA). SBT based phylogenetic comparisons between different strains is accomplished by observing sequence variations of 7 allelic markers including house-keeping genes and known virulence factors , . MLVA takes advantage of the polymorphisms of 9 tandemly repeated DNA sequences . The UPGMA analysis based on a matrix of pairwise allelic differences between all the sequence types (ST) of our dataset identified a large cluster (n=40) comprising most of the Sg6 isolates of our repository. Isolates of this large cluster showed less phylogenetic diversity compared with other clusters and it encompassed 2 two main subclusters (Figure S1). The phylogenetic subcluster 1 included 17 isolates of Sg6 and ST 187 (Figure S1), while the phylogenetic subcluster 2 included 9 isolates who were all identified as Sg6 ST 68 (Figure S1). MLVA typing of this dataset, identified a large phylogenetic cluster as well, comprising two subclusters, where cluster 1 grouped 20 isolates, and included Sg6 ST 187 and Sg6 ST 68, along with Sg4, Sg8 and Sg10 strains (Figure S2). Interestingly, cluster 2 had 10 isolates, out of which 8 were Sg6 ST 187, 1 was a Sg6 ST242 and 1 was a Sg3 ST93 (Figure S2). To further increase the discriminatory power of the molecular typing methods , we next constructed a phylogenetic tree combining SBT and MLVA profiles (Figure 1). This combined analysis confirmed that the largest phylogenetic cluster of non-sg1 clinical isolates comprised mainly of Sg6 ST187 and Sg6 ST68. In contrast to several other molecular types such Sg6 ST407 which had only been isolated over a period of 5 years, Sg6 ST187 and Sg6 ST68 strains seemed clonally stable, since isolates with identical sequence types have been obtained for 25 to 30 year periods of our repository (initiated in 1980). For the purposes of further analysis we selected strain Thunder Bay, Sudbury and Mississauga isolates as the representative ST187, ST68 and ST407, respectively. The high-prevalence of Sg6 among non-Sg1s was also confirmed through an inspection of the European Working Group for Legionella Infections (EWGLI) database (http://www.ewgli.org/) and has been reported independently elsewhere .
Optical mapping is an emerging technology used for accurate genome assemblies  and for conducting comparative chromosomal analysis . It allows for rapid identification of divergent regions, chromosomal rearrangement, and inversions , , . We compared the relatedness of Sg6 str. Thunder Bay with other non-Sg1 and Sg1 strains, whose genomic sequences are available. First, we constructed an optical map of Sg6 str. Thunder Bay and compared it to in silico maps of Sg1 str. Philadelphia and Sg12 str. 570-CO-H (Figure 2A). The optical maps of Sg6 str. Thunder Bay and Sg1 str. Philadelphia only showed divergence in a 40 kB segment. This segment corresponded to genes involved in O-antigen biosynthesis. In contrast, Sg12 str. 570-CO-H showed a similar restriction pattern with Sg6 str. Thunder Bay. Secondly, we constructed an evolutionary tree based on optical maps of Sg6 str. Thunder Bay, Sg12 str. 570-CO-H and all sequenced and fully assembled Sg1 strains (Figure 2B). This analysis confirmed that the clinical non-Sg1 strains, Sg6 str. Thunder Bay and Sg12 str. 570-CO-H share a high degree of homology, and are phylogenetic neighbours of Sg1 str. Philadelphia.
The comparative optical map analyses identified possible serogroup specific genomic parameters. In order to investigate if these differences result in possible serogroup specific advantages between Sg1 str. Philadelphia and Sg6 str. Thunder Bay, we conducted comparative biological analysis at different stages of the L. pneumophila life-cycle. A strategy used by L. pneumophila to survive in anthropogenic aqueous environment is to generate biofilms . The ability to generate higher amounts of biofilm can be directly correlated with persistence in the environment . We determined that Sg6 str. Thunder Bay produces significantly higher levels of biofilm compared to Sg1 str. Philadelphia by 48 hours (P<0.05; Figure 3A). Furthermore, this difference further increased by 72 hours (P<0.05). This suggests that while Sg1 are able to generate significant amounts of biofilm, the Sg6 might possess an advantage that could explain its higher reported prevalence in some environmental surveys .
In order to determine a possible advantage that either Sg1 or Sg6 may possess in infecting humans and surviving in the environment, we compared the replication rates of Sg1 str. Philadelphia and Sg6 str. Thunder Bay over 48 hours in the human lung epithelial cell line H292, U937 macrophages and Acanthamoeba castellani amoeba models , , . While no differences were observed in the lung epithelial cells (Figure 3B), significant increase of Sg6 str. Thunder Bay counts over Sg1 str. Philadelphia was seen by 48 hours in the macrophage model (Figure 3C). Similarly in amoeba, while no discernable difference was observed between the Sg1 and Sg6 strains one hour after uptake, by 24 hours, counts for the Sg6 strain were 28-fold higher compared to Sg1 str. Philadelphia (Figure 3D). At 48 hours, this difference had increased to 168-fold, suggesting that Sg6 str. Thunder Bay is able to infect and replicate more effectively than Sg1 str. Philadelphia in A. castellani model. Taken together, these results suggest that the Sg6 str. Thunder Bay replicates more efficiently than Sg1 str. Philadelphia within amoebae and macrophages. Thus Sg6 str. Thunder Bay might possess a fitness advantage in the environmental niche over the Sg1 str. Philadelphia, which is an unlikely explanation for the observed differences in clinical prevalence.
To identify the factors that may be contributing to the population structure of L. pneumophila in clinical settings, we performed whole genome sequencing of Sg6 str. Thunder Bay. The L. pneumophila Sg6 str. Thunder Bay (accession number: CP003730) genome consists of a 3455167-bp chromosome with an average GC content of 38.24% (Table S1). A total of 3052 protein-encoding genes are predicted, 2299 (75.3%) of which have been assigned a putative function. A BLAST map analysis of Thunder Bay against five reference Sg1 strains showed several conserved regions (Figure 4A). Furthermore, high genomic strand asymmetry identified through the GC skew suggested no abnormalities in the assembly of the genome  (Figure 4A). To determine plasticity of the Sg6 genomes, we selected two additional strains from our clinical repository, Sg6 str. Sudbury (Accession number PRJNA170020) and Sg6 str. Mississauga (Accession number PRJNA170014), and sequenced their genome using Illumina technology (paired-end). The genomes were de novo assembled into 29 and 37 contigs, respectively. By including these strains in our analysis, we report genomic data on most sequence types that were identified as Sg6 in our database. The genome size of Sg6 str. Sudbury (3351893 bp) and Sg6 str. Mississauga (3384301 bp) were comparable to that of the Sg6 str. Thunder Bay. Reference assembly of raw Illumina reads to the finished Sg6 str. Thunder Bay genome also indicated that Sg6 str. Sudbury was more closely related to Sg6 str. Thunder Bay (~4800 SNPs) than Sg6 str. Mississauga (~79,000 SNPs) (data not shown). Taking these SNP variations into consideration, the overall polymorphism between Sg6 str. Thunder Bay and Sg6 str. Sudbury only represents 0.13% of the Sg6 str. Thunder Bay genome while the polymorphism between Sg6 str. Mississauga and Sg6 str. Thunder Bay represents 2.28% of Sg6 str. Thunder Bay genome.
The in silico optical maps deducted from concatenated whole genome sequencing contigs of Sg6 str. Sudbury and Sg6 str. Mississauga were next compared to the optical maps of Sg6 str. Thunder Bay, Sg1 str. Philadelphia and Sg12 str. 570-CO-H (Figure 5A). Similarly to Sg12 str. 570-CO-H and Sg6 str. Thunder Bay, Sg6 str. Sudbury showed high similarity with Sg1 str. Philadelphia with the exception of the O-antigen biosynthesis segment. In contrast, differences between optical maps of Sg6 str. Mississauga and other strains were distributed around the chromosome whereas the O-antigen biosynthesis segment was only partially conserved. An evolutionary tree based on optical maps of all sequenced Sg6 strains of our study, Sg12 str. 570-CO-H and all sequenced and fully assembled Sg1 strains was next constructed (Figure 5B). This analysis identified a cluster of strains with a relative low diversity compared to other genomes, this cluster included Sg6 str. Thunder Bay, Sg6 str. Sudbury, and Sg12 str. 570-CO-H and Sg1 str. Philadelphia. In contrast, the low prevalent Sg6 str. Mississauga did not cluster with Sg6 str. Thunder Bay and was relatively distant from its closest phylogenetic neighbours including Sg1 str. Alcoy and Sg1 str. Corby.
In order to identify genes that might confer differential prevalence of L. pneumophila serogroups, we compared the 2486 Sg1 proteins that have been previously described as the Sg1 core  for amino acid sequence homology with the Sg6 str. Thunder Bay predicted proteins. Forty three Sg1 core proteins shared less than 65% homology and were defined as divergent (Table S2, Figure 4B). While 35 amongst these were described as hypothetical proteins (HP), there was a cluster (~40 kB) of 11 predicted proteins that corresponded to O-antigen biosynthesis (lpg0738 to lpg0786). This result is in agreement with the optical map based comparative analysis between Sg1 str. Philadelphia and Sg6 str. Thunder Bay, which identified the O-antigen biosynthetic cluster of Sg1 as divergent. This is also in agreement with a previous study revealing that Sg1 LPS gene cluster spaning lpp0827 (galE) to lpp0843 of L. pneumophila Sg1 str. Paris is highly divergent from those of Sg6, Sg10, Sg12, Sg13 and Sg14 . Some other predicted proteins that were outside of this cluster included MreD (lpg0813) and LvrB (lpg1258). The role of LvrB has been characterized in several bacterial pathogens and has been shown to be essential in the formation of the Type IVA (lvh) secretion system , , , . MreD has also been described in several bacteria, and has been shown to be involved in controlling cell wall biosynthesis and bacterial morphology , . In our study, no morphological difference was observed between the Sg1 and Sg6 using differential interference contrast microscopy (data not shown).
In a reverse comparison, 581 Sg6 str. Thunder Bay proteins were defined as divergent compared to the Sg1 core. Amongst these only 41 were considered divergent to all L. pneumophila in the Genbank database (Table 1, Figure 4C). Twenty three of these Sg6-specific proteins were identified from the O-antigen biosynthetic cluster, and included proteins like Wzm (lp6_749), Wzt (lp6_750), GalE (lp6_758) and WecA (lp6_759), whose role in O-antigen biosynthesis is well described , , , . Notably a previous study failed to identify the presence of many of the genes coding for O – antigen biosynthesis proteins by DNA microarray in all non-Sg1 tested . Our results suggest that this may have been due to low sequence homology of these genes with their Sg1 orthologs. Interestingly, the Thunder Bay O-antigen cluster was fully conserved in the Sg6 str. Sudbury, but showed low homology in Wzt and N-actyltransferase when compared to Sg6 str. Mississauga.
Four Sg6 str. Thunder Bay proteins were defined as unique based on sequence homology with predicted proteins from all Sg1 genomic sequences (Table 1). Amongst these, three predicted proteins are coded by genes laying outside the O-antigen cluster and were classified as hypothetical proteins and XerD (lp6_1224), a site specific recombinase (Blakely and Sherratt, 1994). The remaining two proteins are coded by genes situated within the O-antigen cluster and included PseG (lp6_763), whose homologs have been shown to play a role in pseudaminic acid biosynthesis in Campylobacter jejuni and Pseudomonas aeruginosa , . Even though pseudaminic acid has not been detected in L. pneumophila, a nine-carbon alpha-keto acid called legionaminic acid which belongs to the sialic acid family is generated by NeuA, NeuB and NeuC (Glaze et al., 2008). This acid is known to be a structural component of the L. pneumophila O – antigen. In agreement with a previous comparative analysis of L. pneumophila serogroup LPS clusters , this divergent O – antigen biosynthesis cluster were 100% homologous with corresponding L. pneumophila Sg12 str. 570-CO-H proteins (Table 2). At the DNA level, the region spanning Lp6_758 (galE) to Lp6_782 of Sg6 str. Thunder Bay showed respectively a 99% and 100% identity with the previously reported regions of Sg6 str. ATCC33215 and Sg12 str. ATCC43290 .
We next defined a non-Sg1 core of 2946 proteins, by comparing Sg6 str. Thunder Bay and Sg12 str. 570-CO-H predicted proteins  (Table S3). Upon comparing with the Sg1 core, 435 proteins were defined as divergent from the non-Sg1 core. This included the O-antigen biosynthetic cluster, flagellar assembly, T4SS substrates and some metabolic enzymes. Interestingly, amongst the 2486 Sg1 core proteins, only 79 were defined as divergent compared to the non-Sg1 core (Table S4), and included genes from O-antigen biosynthetic cluster and the Lvh region of the type IV-A secretion system . These differences might reflect the discrepancy in growth and replication in the environment or the human host.
Comparative genomics and optical mapping analysis of the Sg6 str. Thunder Bay genome revealed that the O-antigen biosynthetic proteins present major variations with respect to the Sg1str. Philadelphia (Figure 6A). Therefore, we predicted that there might be changes in the O-antigen generated by Sg1 str. Philadelphia and Sg6 str. Thunder Bay. To verify this hypothesis, purified lipopolysaccharide (LPS) from these strains were visualized through SDS-PAGE. Sg1 displayed a classic ladder pattern (Tsai and Frasch, 1982), which suggests that the O-antigen is highly decorated . Conversely, the Sg6 O-antigen appears to be minimally decorated and lacked higher molecular weight species (Figure 6B). Our Sg6 str. Thunder Bay profile is in agreement with a previous study that analyzed LPS from Sg6 str. MP . Previously, differences in O – antigen profiles have been used to define distinct serogroups , , and are hypothesized to be associated to differences in virulence of L. pneumophila serogroups in humans .
Variations in the outer-membrane O – antigen segment of the LPS can provide resistance to complement mediated serum killing in pathogenic bacteria , , . The differences observed in the O – antigen structure of Sg1 str. Philadelphia and Sg6 str. Thunder Bay indicated that there might be differential sensitivity to serum complement. Indeed when the Sg6 was treated with Non-immune Human Serum (NHS), a 35% and a 73% decrease in colony forming units (CFUs) was seen by 45 and 120 min, respectively (Figure 6C). However, no effect was observed for Sg1. Sg6 and Sg1 viability was not affected by decomplemented (heat inactivated) NHS, consistent with a Sg6-specific killing by heat-sensitive complement. These results suggest that Sg6 O-antigen is unable to provide any appreciable resistance to serum complement, whereas Sg1 O-antigen may allow the bacteria to escape complement mediated killing by naive human serum.
We next measured the sensitivity of Sg1 str. Philadelphia and Sg6 str. Thunder Bay to the classical complement pathway using sera obtained from patients diagnosed with either Sg1 or Sg6 infections (Figure 7A). Presence of either Sg1 or Sg6 specific antibodies was previously confirmed through IFA (data not shown). The Sg1 str. Philadelphia isolate showed a 72.5% decrease in CFU after a 60 min incubation in sera containing Sg1 antibodies, an effect that was not observed in decomplemented patient sera. In presence of Sg6 specific antibodies, the Sg1 str. Philadelphia CFUs were decreased by 31.5% at 60 min. The Sg6 str. Thunder Bay counts, when incubated with Sg1 antibodies were decreased by 98.8%, and similarly decreased by 99.3% in Sg6 patient sera. Taken together, these results suggest that while Sg1 str. Philadelphia is resistant to the alternate complement system, it is highly sensitive to the classical pathway of complement mediated killing. Furthermore, Sg1 are more sensitive to this pathway in the presence of Sg1 specific antibodies, while the Sg6 can be targeted by both Sg1 and Sg6 antibodies.
The above experiments suggested that the high prevalence of Sg1 in patient cases might be related to higher resistance to serum complement. We hypothesized that we might observe a similar correlation between clinical prevalence and resistance to the alternative complement pathway amongst the non-Sg1 strains in our repository. From this set, we selected the most prevalent sequence type for each serogroup, and measured their survival in NHS (Figure 7B). The highest CFU/ml counts were obtained for Sg6 strains Thunder Bay and Sudbury, which account for 20% and 10.6% of all non-Sg1 infections, respectively. Interestingly Sg6 str. Mississauga, which account for only 2.35% of all clinical cases, showed a marked reduction in survivability compared to Thunder Bay and Sudbury strains (Figure 7C). All other serogroups that shared similar prevalence as Sg6 str. Mississauga, exhibited high sensitivities to serum complement. These results strongly indicate that while all non-Sg1’s are sensitive to serum complement, the degree of susceptibility correlates strongly with clinical prevalence.
Serogroup specific differences in complement resistance may be reflected in the efficiency to avoid the formation of membrane attack complex (MAC) , . MAC forms transmembrane channels in the bacterial membrane, leading to depolarization and cell death . We compared the binding of complement proteins C3 and C9 to the surfaces of Sg1 str. Philadelphia and Sg6 str. Thunder Bay by indirect fluorescent antibody (IFA). C3 protein is able to bind to the LPS of bacteria and provides the scaffold for the membrane attack complex , . The binding of the C9 protein completes the membrane attack complex, which leads to bacterial death , , . Our experiments indicate that while only residual amounts of C3 bind Sg1 (Figure 7D i), significant quantities were detected on the Sg6 surface (Figure 7D iii). This subsequently led to the detection of C9 on Sg6 (Figure 7E iii), but absent from Sg1 surfaces (Figure 7E i). This result suggests that the membrane attack complex can be assembled in Sg6 but not in Sg1, highlighting Sg6’s sensitivity to the alternative complement pathway.
The different serum resistance profiles of Sg1 and Sg6 suggested that these serogroups might differ in their ability to cause infection in the lungs and/or dissemination to other organs. In order to test this hypothesis, we infected A/J mice intratracheally with Sg1 str. Philadelphia or Sg6 str. Thunder Bay and determined the severity of the disease 48 hours post intra-tracheal inoculation , . Forty eight hours post infection, mice infected (n=5) with Sg1 showed symptoms of distress as indicated by ruffled fur, loss of appetite and lethargy. Conversely, mice infected with Sg6 (n=5) displayed no deviation from normal behaviour and appearance. Furthermore, while no significant differences were seen in bacterial counts in the lungs, a 100-fold increase was noted for Sg1 in blood. Similarly, 100-fold and 10-fold increases were also seen for Sg1 in kidneys and liver, respectively (Figure 8). Surprisingly, no differences were observed in the spleens. This data suggests that while both Sg1 and Sg6 cause similar level of infections in the lung, they differ in their ability to induce bacteraemia and disseminate to other organs. Severe complications associated with Legionnaire’s disease are well documented and include septicemia and multi-organ failure , . Therefore, our data are in agreement with the clinical manifestations of Legionnaire’s disease, and suggests that bloodstream dissemination of L. pneumophila is a key determining factor of the severity of legionellosis.
The human host provides a challenging environment for the survival of pathogens. Given that no human to human transmission has been reported for L. pneumophila, the clinical prevalence of L. pneumophila Sg1 over other serogroups poses a puzzling problem, since any Sg1 specific advantage in the human host must be a secondary consequence to an environmental benefit. Here, we thoroughly investigated this question by reporting a comprehensive study comparing the genomes, the clinical impact, the environmental fitness and the pathogenicity of a clinically prevalent Sg6 strain with a phylogenetically related Sg1 strain.
By combining MLVA and SBT analyses of a population based repository of non-Sg1 clinical isolates, we could identify a large cluster of prevalent molecular types presenting a low diversity. At the phylogenetic level, isolates from this cluster seem clonally stable as they have been collected for over 30 years. Based on this finding, a member of this cluster, Sg6 str. Thunder Bay, was selected for further analyses. Given the high plasticity of previously reported Legionella genomes, the optical maps of Sg6 str. Thunder Bay and Sg1 str. Philadelphia showed a high degree of similarity in genomic structure and restriction pattern, except for one large divergent region that includes several genes involved in O-antigen biosynthesis . Furthermore, a comparison of Sg6 str. Thunder Bay and Sg12 str. 570-CO-H optical maps identified the O-antigen region to be conserved between these two sequenced non-Sg1 strains. Comparative genomics analysis was consistent with our optical mapping experiments, where a majority of genetic differences between Sg1 str. Philadelphia and Sg6 str. Thunder Bay were identified in the O-antigen biosynthetic cluster. The potential role of the LPS cluster in the predominance of Sg1 was initially hypothesised in a large multigenome microarray analysis of L. pneumophila strains . More recently, the O-antigen regions of several serogroups were sequenced and compared using a PCR/Sanger sequencing approach. Using this approach, O-antigen biosynthetic genes wzt, wzm and lag1 were only identified in Sg1 and it was postulated that these genes might be absent in other serogroups . Here, while at the protein level they only share low homology with the Sg1 and discrepancies with previous reports may be explained by the use of different comparative genomics cut-off values, next generation sequencing approaches allowed us to identify distant homologues of wzm, wzt and galE genes in Sg6. Together Wzm and Wzt form the inner membrane O-antigen transporter ,  and GalE is a cytoplasmic epimerase that converts UDP-Galactose to UDP-Glucose upon which the remaining O-antigen sugars are assembled . Lag1 was previously described as an O-acetyltransferase in some Sg1 strains , shared less than 50% sequence identity with a predicted AlgL an alginate acetyltransferase in all three Sg6 strains and Sg12 str. 570-CO-H. Although, mutations in lag-1 have been associated with lack of the 8-O-acetyl groups in legionaminic acid leading to polymorphic changes in the LPS, these modifications were shown to have no effect on serum resistance or uptake in amoeba . Interestingly, it has been proposed that the 8-O-acetylation of first three O-antigen sugars, which are located closest to the core sugars, might occur via a Lag1 independent pathway .
The chemical structure of Sg1 O-antigen is very unusual due to its atypical hydrophobic nature, owing to the presence of a homopolymer of legionaminic acid on its surface . Previous analysis of the LPS from non-Sg1 suggested that L. pneumophila O-antigens are structurally conserved, since legionaminic acid was identified in all serogroups, although they lacked O-acetylation . This finding was further substantiated when O-antigens from different L. pneumophila serogroups were visualized through SDS-PAGE, and polymorphic changes compared to Sg1 were reported . Our analysis indicates that while Sg6 does generate an O-antigen, it might lack high molecular weight species.
The structural differences in the O-antigen leading to the increased clinical prevalence of Sg1 may be the result of a selective pressure from the natural bacterial reservoir. In other pathogens, O-antigen plays a role in biofilms architecture  and predation by environmental amoebae . Furthermore, GalE, which showed serogroup specific differences in our study, has previously been shown in Porphyromonas gingivalis to play a significant role both in O-antigen biosynthesis and biofilm production . In light of these studies, we hypothesized that the O-antigen differences between Sg1 and Sg6 might confer an advantage for Sg1 in surviving in the environment. Comparative analysis showed that Sg6 str. Thunder Bay generates increased levels of biofilm compared to Sg1 str. Philadelphia, suggesting that the Sg6 might possess an advantage when it comes to colonizing water systems. A second strategy employed by L. pneumophila to survive within environmental water systems is to replicate within the amoeba host. Although our analysis was limited to one reference amoebae species model , Sg1 str. Philadelphia and Sg6 str. Thunder Bay were equally effective in infecting A. castellani, while Sg6 str. Thunder Bay is able to replicate better within these amoebae. These findings, in combination with the conclusions drawn from our biofilm analyses, suggest that Sg1 may not possess environmental advantage upon other serogroups. Furthermore, the increased efficiency of Sg6 in replicating in amoebae should most likely permit its transmission in larger doses. This is in agreement with previous studies that have either detected similar levels of Sg1 and non-Sg1 in environmental surveys, or have reported the environmental prevalence of Sg6 over all serogroups , , , . More specifically, studies report that Sg6 strains are more abundant in urban water sources in Ontario , and are also more prevalent than Sg1 strains in environmental niches, like the Great Lakes . More so, in absence of an obvious demonstrated environmental advantage for Sg1 strains, it remains unknown why certain LPS structures have been selected, but we speculate it may protect Sg1 strains from being compromised by exposure to environmental stress factors.
In several pathogens, highly decorated O-antigens structures provide protection against serum complement , , . Therefore, we hypothesized that different O-antigen profiles of Sg1 and Sg6 might confer different levels of resistance to serum complement. Our data indicates that Sg1 str. Philadelphia is highly resistant to the alternative complement system, while both Sg1 str. Philadelphia and Sg6 str. Thunder Bay are susceptible to the classical complement pathway. Interestingly, differences in serum resistance amongst different serogroups correlated to their clinical prevalence, since Sg6 str. Thunder Bay and Sudbury, the two most predominant non-Sg1 strains in Ontario exhibited the least sensitivity to the alternative complement system. L. pneumophila’s susceptibility to the classical complement pathway has previously been reported . Unlike the classical complement pathway where complement proteins bind the surface of bacteria opsonised with IgG/IgM , in the alternative pathway the spontaneous hydrolysis of C3 into C3a and C3b fragments triggers the activation of the complement cascade . C3 was previously shown to bind MOMP on L. pneumophila surfaces in small quantities, which may lead to the activation of the alternative complement system . Perhaps, the highly decorated LPS of Sg1 interferes with this interaction, and therefore, would lead to a limited binding of C3, resulting in increased resistance to serum complement . This is in agreement with previous evidence that suggests that certain strains of Sg1 might be resistant to alternative complement pathway as well . A previous study also suggested that C3 cannot be detected on Sg1 surfaces using IFA in the absence of serum antibodies . Similarly, we could not detect C3 on Sg1 str. Philadelphia surface, but it was present on the surface of Sg6 str. Thunder Bay. We also detected C9 on the surface Sg6 str. Thunder Bay indicating the activation of the complement cascade.
We next compared the virulence of Sg1 str. Philadelphia and Sg6 str. Thunder Bay using in vitro and in vivo experimental models. Both strains showed equal ability to infect and replicate within human lung epithelial cells. Unexpectedly, Sg6 str. Thunder Bay showed a high propensity to replicate inside macrophages compared to Sg1 str. Philadelphia. In intratracheally infected naïve mice, our findings indicate that both strains have comparable ability to colonize lungs, suggesting that complement mediated killing via the alternative pathway does not play a significant role in the immune response against L. pneumophila within lungs. Previously, the alternative complement cascade was shown to be limited within lungs . From the lung alveoli, the bacteria may disseminate to other organ systems by crossing the lung endothelium as either free bacteria or inside macrophages . Strikingly, Sg1 str. Philadelphia and Sg6 str. Thunder Bay differed in their potential to disseminate to other organs and to induce bacteraemia. In blood, where complement mediated lysis is most active, the Sg6 counts were 100-fold lower than Sg1. Significantly lower Sg6 counts were also seen in the liver and kidney. This is presumably due to the blood processing roles of these organs. The lower counts of Sg6 str. Thunder Bay in the blood suggests that after the first cycle of replication in circulating macrophages, the bacteria lyse the cells for a second round of infection, and in doing such exposes itself to the complement system. Surprisingly, no serogroup specific difference was seen in the spleen. Dissemination to kidneys and liver requires transit through blood, but both lymphatic and the circulatory system are directed to the spleen. Taking this into account, the data suggests that Sg6 and possibly L. pneumophila use the lymphatic system for dissemination beyond the lungs. All together our in vitro and in vivo results support that resistance to the alternative complement system may play a crucial role in the high clinical prevalence and severity of L. pneumophila Sg 1 infections.
The analysis of the role of O – antigen would be better examined by swapping the large O – antigen biosynthetic cluster between L. pneumophila Sg1 and Sg6. However, genetic tools to achieve this exchange have yet not been developed. Furthermore, the region in question has previously been shown to be an essential element in the genome of L. pneumophila , therefore limiting the likelihood of obtaining knockout mutants. Thus, the data presented in this comprehensive study combining long term molecular epidemiology, comparative genomics and pathogenicity analyses provide a unique perspective on the role of serogroups in L. pneumophila infections. The ability of Sg1 strains to tolerate the alternative complement cascade may allow it to disseminate to other organs via transit through blood, which is in agreement with the septicaemia and bacteraemia associated with the disease , . Because there appears to be no immediate benefit of these differences within biofilms or a natural amoebal host, future investigations explaining why this specific trait is sequestered amongst the Sg1 strains will provide critical insight into the environmental factors and hosts that shape the evolution of this important pathogen’s virulence.
The collection of control human blood from healthy volunteers used in this study was approved by the Research Ethics Board of the Mount Sinai Hospital (Toronto, Canada). As approved by the Research Ethics Board of the Mount Sinai Hospital, a written informed consent to participate in this study was provided by all participants. The Keenan Research Centre of the Li Ka Shing Knowledge Institute, Animal Care and Use Committee reviewed and approved our study protocol for infecting and sampling mice with Legionella pneumophila. All work in our study was conducted adhering to the institution’s guidelines for animal husbandry, and followed the guidelines Canadian council on animal care for the care and use of Laboratory Animals.
Legionellosis is a notifiable disease in Ontario (population 13 million persons). Since 1978, the diagnosis of Legionella infections has been centralized at the Ontario Public Health Laboratory. This laboratory serves as the Legionella reference laboratory and performs all testing for outbreak investigations and most testing of clinical specimens. Therefore, isolates analyzed in this study are representative of the non-Sg1 strains isolated in Ontario in the past 3 decades. The proportion of culture-confirmed case-patients with L. pneumophila infection remained stable during the period of analysis, and 34% of the isolates were non-SG1 . Species and serogroups were confirmed by direct immunofluorescent antibody assay and slide-agglutination , . Isolates (n=81) were stored at –80°C in trypticase soy broth supplemented with 5% horse blood. Eight isolates could not be resuscitated and were omitted from our analysis. With the exception of 2 Sg6 isolates collected at the same date in the same hospital, none of the isolates were epidemiologically related. L. pneumophila ST 39 (Sg2–3 culture confirmed cases), ST 68 (str. Sudbury (CG346); Sg6–9 culture confirmed cases), ST 187 (str. Thunder Bay (CG315); Sg6–17 culture confirmed cases), ST 378 (Sg4–2 culture confirmed cases), ST 407 (str. Mississauga (CG331); Sg6–2 culture confirmed cases), ST 465 (Sg3–5 culture confirmed cases), ST 466 (Sg8–1 culture confirmed case), ST 470 (Sg10–1 culture confirmed case), ST 471 (Sg5–1 culture confirmed case) were selected among the 73 typable isolates for further analyses. Reference clinical strain L. pneumophila subspecies pneumophila str. Philadelphia-1 (ST 36) was obtained from ATCC (#33152).
The initial molecular typing of L. pneumophila strains was conducted using Legionella Sequence based Typing (SBT) scheme , , . Molecular typing was also concurrently done through the Multiple-Locus VNTR Analysis Typing Scheme (MLVA), as formerly reported . A combined MLST-MLVA dendrogram was also prepared according to a recently described method . The serogroups of patient isolated strains was determined using a method previously described .
The Sg6 str. Thunder Bay optical map was generated according to the manufacturer’s protocol (OpGen). In short, the strain was grown for four days on buffered charcoal yeast extract (BCYE) agar at 37°C with 5% CO2. Genomic DNA was extracted using the OpGen DNA extraction kit from a single colony. DNA purification via this method generated concentrated, non-sheared DNA. The DNA was allowed to relax overnight and its quality was checked via a Q-Card. The DNA was next applied to a special cover slip using a wide-bore pipette, and as a result of capillary action the DNA was stretched in channels. The DNA was stained with a fluorescent dye and in the presence of an antifade solution, and was visualized through a laser microscope. The DNA concentration was adjusted to achieve 20–30 DNA fragments/frame, with a minimum size of 150 kB. The diluted DNA was applied to the M-card, and was treated with NheI for 30 min. The digested pieces of DNA were visualized via the Optical mapping array and fragment length was assigned by the onboard software (MapManager). These fragments were aligned to generate a circularized map with a minimum coverage of 65 X. Optical maps for reference strains were constructed in silico by providing genome sequences from public databases. The optical maps were used to construct an evolutionary tree using Mapsolver (OpGen) based on the UPGMA algorithm.
Biofilm assays were conducted according to a previously published protocol . Briefly, single clones grown for 4 days on BCYE agar at 37°C and 5% CO2 were used to inoculate broth cultures, which were allowed to reach an OD of 2.0 and diluted in fresh medium to give a final OD at 600 nm of 0.2. In a 96 well plate, 200 µl of this suspension was dispensed, and the plates were incubated at 37°C and 5% CO2 for 2 and 4 days. Biofilms were stained with 40 µl 0.25% crystal violet in each well for 15 min and washed 3 times in deionized water, and the crystal violet stain was solubilized in 95% ethanol. Absorbance was read at 600 nm. Three independent experiments were performed using eight replicates.
A. castellani was obtained from ATCC (50739) and was cultured in peptone yeast glucose (PYG) broth (2% Peptone, 0.1% Yeast Extract, 0.1 M Glucose, 4 mM MgSO4, 0.4 M CaCl2, 0.1% HOC(COONa)(CH2COONa)2 · 2H2O, 0.05 mM Fe(NH4)2(SO4)2-6H2O, 2.5 mM NaH2PO3, 2.5 mM K2HPO3, pH 6.5) as previously described . U937 monocytes were obtained from ATCC (CRL-1593.2) and cultured in Roswell Park Memorial Institute (RPMI) media, supplemented with Glutamine and 10% heat inactivated fetal bovine serum. All cells were passaged 3 times before inoculation assays were carried out. U937 cells were differentiated into adherent macrophages by treatment with 1.0×10−3 mg/ml 12-O-Tetradecanoylphorbol-13-Acetate for 48 hours. L. pneumophila Sg1 and Sg6 patches were grown on BCYE agar for 4 days at 37°C with 5% CO2 and suspended in PBS. In a 24 well plate, 105 amoebae/macrophages were incubated with 106 bacteria for a multiplicity of infection (MOI) of 10. The incubation was carried out for 2 hours at 37°C with 5% CO2, after which free bacterial cells were removed by washing the wells 3 times with AC buffer (4 mM MgSO4, 0.4 M CaCl2, 0.1% HOC(COONa)(CH2COONa)2 · 2H2O, 0.05 mM Fe(NH4)2(SO4)2-6H2O, 2.5 mM NaH2PO3, 2.5 mM K2HPO3, pH 6.5) or PBS. Next the infected cells were treated with gentamicin (100 µg/ml) for 1 hour to kill extracellular bacteria and synchronize the infections. The amoebae/macrophages were washed 3 times with AC buffer/PBS and incubated for 0, 24 and 48 hours in AC buffer/PBS at 37°C with 5% CO2. At the end of the incubation period the cells were lysed by incubating them for 5 min in ice-cold filter sterilized tap water, followed by successive passes through a 27.5 gauge needle. Serial dilutions of lysate and supernatant mixture were plated on BCYE agar and incubated for 4 days at 37°C with 5% CO2 for CFU enumeration.
Binding and invasion assays of NCI-H292 (ATCC, CRL-1848) cells from human lung mucoepidermoid carcinoma with L. pneumophila str. Philadelphia and Thunder Bay were conducted as previously described , but with the following modifications: Bacteria were resuspended in PBS from 4 day old plates that had been incubated at 37°C with 5% CO2, and lung epithelial cells were inoculated at an MOI of 10 and were lysed by incubation in filter sterilized tap water for 5 minutes and passed through a 27.5 gauge needle.
LPS were purified from L. pneumophila str. Philadelphia and Thunder Bay as previously described . In brief, bacterial suspensions (OD600 2.0) were centrifuged at 10,000×g for 5 min. The pellets were washed twice in 5 ml PBS (pH 7.2) containing 0.15 mM CaCl2 and 0.5 mM MgCl2. Pellets were then resuspended in 10 ml PBS (pH 7.2) and sonicated at 4 W three times at an output setting of 0.5 (Misonix S3000) for 10 min on ice. In order to eliminate contaminating protein and nucleic acids, samples were treated with proteinase K (100 µg/mL) at 65°C for 2 hours, followed by DNaseI (20 µg/mL) and RNase (40 µg/mL) treatments at 37°C overnight. The extraction of LPS was initiated by adding an equal volume of hot (65–70°C) 90% phenol with vigorous shaking at 65–70°C for 2 hours. Suspensions were cooled on ice, transferred to 1.5 mL polypropylene tubes and centrifuged at 8500×g for 15 min. Supernatants were transferred into 15 mL conical centrifuge tubes and phenol phases were re-extracted using 300 µL dH2O. Sodium acetate at 0.5 M final concentration followed by 10 volumes of 95% ethanol were added to the extracts and samples were stored at −20°C overnight in order to precipitate LPS. Tubes were centrifuged at 2000×g, 4°C for 10 min and the pellets were resuspended in 1 ml distilled water. Extensive dialysis against distilled water at 4°C was carried out to remove any residual phenol. Purified LPS was solubilized in Laemmli buffer and boiled for 5 min to achieve a final concentration of 1 mg/ml. 20 µL/well from each sample was separated on 15% SDS gel with a 4% stacking gel under reducing condition at 100 mA for 2 hrs, using a mini-PROTEAN electrophoresis instrument (Bio-Rad Laboratories). Silver staining of the gels was performed according to the manufacturer’s protocol (Bio-Rad Laboratories).
The genome of L. pneumophila str. Thunder Bay was sequenced using a combination of Sanger, Roche-454 and Illumina sequencing platforms. All general aspects of Roche-454 library construction and sequencing were conducted according to the manufacturer’s directions. Preparation of Illumina libraries and sequencing were conducted as previously described . Sequencing using the Roche-454 platform generated individual reads at a minimum 19× coverage, while Illumina paired-end reads were obtained at 99× coverage. Initial genome assembly of 454 reads via the Newbler software package (Roche) generated 36 large contigs. The order of the contigs was determined via optical mapping and gaps between these contigs were closed via primer walking to generate a single molecule. A total of 440 Sanger finishing reads were produced to close gaps, while Illumina reads were used to resolve repetitive regions, and raise the quality of the finished sequence. Sequencing errors were corrected by mapping Illumina reads to the assembled L. pneumophila str. Thunder Bay genome using CLC Genomics Workbench (Ver 5.1.1). single 454 contig. The closed genome was annotated through the GenDB platform. All gene annotations calls were checked manually and corrected based on greater than 95% homology with other L. pneumophila homologs, Pfam domains, TIGRFAM domains and SignalP predictions. Raw reads of Sg6 str. Sudbury (ST 68) and Mississauga (ST 407) were de novo assembled using CLC Genomics Workbench (Ver 5.1.1), and were transitively annotated against L. pneumophila str. Thunder Bay using the GenDB platform. The L. pneumophila str. Sudbury and L. pneumophila str. Mississauga genomes were compared with L. pneumophila str. Thunder Bay using GenDB (ver 5.1.1) for Single Nucleotide Polymorphism (SNP) analysis and gene content.
The single chromosome of L. pneumophila str. Thunder Bay was annotated using a modified version of the GenDB platform as previously described , . In short, ORF prediction was done in GenDB using REGANOR , which uses CRITICA  and Glimmer  in addition to RBSFinder  to make ribosomal binding site and gene predictions. Ribosomal predictions are done using tRNA-Scan-SE  and RNAmmer . All intergenic regions were elongated with a 25 base pair overlap on each end and run through a pipeline for identification of potential frameshifts and short genes overlooked by GenDB regional prediction pipeline. Regions found with significant BLASTX  hits to EMBL database  were identified, added as new CDS regions in GenDB. All coding regions were ran through the functional annotation pipeline by running a BLAST of each region against the following databases: PFAM , TIGRFAM , KEGG , NCBI’s non-redundant protein database, TMHMM , and SignalP .
Comparative proteomics was conducted by first comparing the L. pneumophila str. Thunder Bay translated genes against Sg1 core, which has previously been described . Predicted proteins of L. pneumophila str. Thunder Bay that exhibited greater than 65% sequence identity were defined as the conserved group. These were then compared against the entire public database of known L. pneumophila proteins, in order to establish a list of unique proteins of L. pneumophila str. Thunder Bay. Furthermore, a list of unique proteins of the Sg1 core was compiled, by including proteins which showed less than 65% homology to L. pneumophila str. Thunder Bay predicted proteins. The pangenome of L. pneumophila was created using GView ver 1.6 .
Human blood was collected from 4 healthy volunteers and allowed to coagulate at room temperature for 30 min. The plasma was collected by centrifuging the tubes for 10 min at 1000×g and aspirating the supernatant. The plasma was analyzed via immunofluorescence assays (IFA) to confirm absence of L. pneumophila antibodies against Sg1 and Sg6 , and then was flash frozen in liquid nitrogen for storage at −80°C. Similarly, IFA was used to confirm the presence of Sg1 and Sg6 specific antibodies in patients diagnosed with Legionnaire’s disease.
L. pneumophila str. Philadelphia and Thunder Bay were grown for 4 days on BCYE agar at 37°C with 5% CO2. The bacteria were suspended in PBS to a final concentration of 106 CFU/ml. The suspensions were diluted 10-fold in 90% non- immune human serum (NHS) and incubated at 37°C for 45 min and 90 min. In the experiments comparing serum resistance of several serogroups, a single incubation time of 60 min was selected. Similarly, experiments involving immune human serum (IHS) were carried out for 60 min. As controls, bacteria were also incubated with heat inactivated serum for the corresponding periods. Next bacteria were pelleted and washed 3 X with PBS, and CFU were determined by plating several serial dilutions on BCYE agar and incubated as before. These CFU values were compared to the inoculums and the data is reported as % survival.
L. pneumophila str. Philadelphia and Thunder Bay were grown for 4 days on BCYE agar at 37°C with 5% CO2, were resuspended in 1× PBS and adjusted to OD 2.0. Bacteria were incubated in NHS for 25 min at 37°C and washed 2× with PBS. Bacterial suspensions were smeared onto glass slides and allowed to air dry. The smears were then heat fixed, followed by fixation in 4% paraformaldehyde for 15 min. Cells were blocked in 5% fetal calf serum in PBS for 1 hour at room temperature, followed by a 1 hour incubation with C3, C5 and C9 primary antibodies diluted (110) in the blocking solution (Hycult Biotech). Images were acquired using Quorum tech Canada Optigrid structure illumination system in a Leica DMI6000B stand with a with a Hamamatsu EM-1K, EMCCD camera.
Six to eight week old female A/J mice (Jackson Laboratories) were infected with 50 µl of 106 bacterial suspensions of either Sg1 str. Philadelphia or Sg6 str. Thunder Bay as previously described (Brieland et al., 1994). The bacterial suspensions were prepared in PBS from bacteria grown on BCYE agar plates for 4 days at 37°C with 5% CO2. The mice were infected via intra-tracheal inoculations. Forty eight hours post infection, the mice were sacrificed and organs were harvested. Homogenized organs and blood were plated on BCYE agar plates and grown for 4 days at 37°C with 5% CO2. The counts are reported as log10 CFU. The experiment was conducted in accordance with the approved protocol at the research vivarium of St. Michael’s hospital (Toronto, Ontario).
SBT typing of a population based clinical repository of L. pneumophila. Phylogenetic clusters were constructed based on UPGMA analysis of SBT distribution. The large cluster and the subclusters 1 and 2 are identified in the phylogenetic tree.
MLVA typing of a population based clinical repository of L. pneumophila. Phylogenetic clusters were constructed based on UPGMA analysis of MLVA based distribution. The large cluster and the subclusters 1 and 2 are identified in the phylogenetic tree.
General features of the sequenced Sg6 strains.
Comparative proteomics analysis between Sg1 core genes and Sg6 str. Thunder Bay. Proteins that share less than 65% sequence homology and 75% coverage were defined as divergent. HP identifies hypothetical proteins.
Genes that define the non-Sg1 core based on comparisons of Sg6 str. Thunder Bay and Sg12 str. 570-CO-H.
Comparison of non-Sg1 core with the Sg1 core.
This work was supported by the Ontario Agency for Health Protection and Promotion (http://www.oahpp.ca/) and the Canadian Institutes of Health Research (MOP-102514) (http://www.cihr-irsc.gc.ca/). AP and MT are supported by the Natural Sciences and Engineering Research Council of Canada (http://www.nserc-crsng.gc.ca/) and the Ontario Association of Medical Laboratories grants (http://www.oaml.com/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.