INTRODUCTION

Top Abstract Introduction Materials & Methods Results Discussion References

The genus Legionella comprises approximately 40 species, at least 7 of which have more than one serotype (3, 15, 31). Approximately half of the species have been associated with human disease (28). Legionella-like organisms isolated from clinical specimens, or from the environment during the course of an outbreak, need to be identified to elucidate the disease process and to identify the source. Legionellae have proved to be relatively unreactive when traditional biochemical tests are utilized, necessitating more complex identification methods (6, 7, 26, 41). Serologically based methods are widely used in clinical laboratories, but antigen cross-reactivity limits specificity and restricts their confident use to a few frequently isolated species (38). This is especially true for countries where legionellosis caused by species other than L. pneumophila is common (12). More complex classification schemes have been proposed (26, 38), the most successful being one based on the range and proportion of cellular fatty acids and ubiquinones (21, 22, 40, 43). As additional species have been characterized, this method has become less discriminating, since apparently unique patterns were proved to be shared by several species (43). The inclusion of hydroxylated fatty acids has improved discrimination, but it requires the analysis of both mono- and dihydroxylated fatty acids, and individual patterns are complex, making analysis difficult (21).

Gene sequence-based phylogenic (genotypic) schemes have become widely used for organisms which are difficult to classify, as more sequences have been determined and sequencing methods have become simpler, more widely available, and cost effective (11, 23, 24, 29, 32, 34). Genotypic schemes have the great advantage of being unaffected by colony age and growth conditions and, in contrast to chromatographic methods, are not subject to extraction and chromatographic conditions or constituent equipment. Additionally, because a gene sequence is essentially a long digital string, with each digit being one of only four nucleotides, genotypic schemes are less ambiguous and can utilize significantly more discriminatory data than phenotypic ones, and in a form that lends itself to widely available computer analysis software. Many genotypic schemes utilize variation in the 16S rRNA sequence (11, 23, 24, 32, 34), because of the ease with which regions can be amplified and sequenced with universal primers. The 16S rRNA sequences of Legionella species have been reported (18), as have the sequences of the mip gene (2, 12, 13, 31), which codes for an immunophilin of the FK506 binding protein (FKBP) class (14). This protein, which ranges in size from 232 to 251 amino acids, depending on the Legionella species (31), is an outer membrane protein important in the intracellular cycle of Legionella. While it is known to be involved with the survival of the bacterium immediately after uptake into phagocytic cells (9, 12, 28), its exact role is unclear. Additionally, analogs are found widely in both prokaryotes and eukaryotes and are likely to have a significant cellular role (14). Other gene sequences have been determined for Legionella (5, 17, 36), but only the rRNA sequences and the mip gene have been comprehensively determined for most species, an essential prerequisite for any gene to be the basis of a genotyping scheme. Ratcliff et al. (31) recently phylogenetically compared most Legionella species, using the species variation among both the 16S rRNA and mip genes, and found over twice the variation in the mip gene at the DNA level (56% of base sites) as in 16S rRNA (23% of base sites). A pairwise comparison of species reveals a mip gene variation of 3 to 31% (mean, 20%) between species pairs compared with 1 to 10% (mean, 6%) for 16S rRNA. For the mip gene, interspecies nucleotide variation occurred throughout the gene but especially within a hypervariable insert of up to 51 bases immediately adjacent to the region coding for the signal sequence, at redundant third codon sites, and in sequences coding for either single or small regions of variable amino acids interspersed among regions coding for total or near-total amino acid homology, especially toward the 3' end of the gene (31). These last regions are known to encode the active portions of the protein's enzymatic peptidylprolyl cis-trans-isomerase (PPIase) activity (31).

Additionally the mip gene appeared to be relatively stable genetically, with no evidence of homologous recombination, in that identical or near-identical sequences were not found for the mip genes from phenotypically divergent species. With respect to genetic stability, the mip gene may therefore behave like housekeeping genes, which are known to be more stable than other gene classes (1). Homologous recombination would severely compromise a sequence-based classification scheme (1), and it is a theoretical possibility at least for rRNA targets (37). Thus, the genetic stability and greater mutational variation of the mip gene suggest that it is an ideal target for a classification scheme, with results likely to be more discriminating in identifying species and more resilient to clonal variation within each species. It may even be possible to discriminate between serogroups where these are present or to demonstrate distinct intraspecies clonal groups.

The present study reports the use of the mip gene to develop a sequence-based classification scheme for Legionella, the first proposed for this genus. Further, it reports the comparison of sequences from species which have additional serogroups, to determine if serogroups can be discriminated. Similarly, it reports the comparison of sequences from wild strains isolated on several continents, for which there is confirmatory phenotypic or DNA hybridization identification data, to test the robustness of the scheme for variation within strains of the same species. Lastly, isolates which appear phenotypically or from DNA hybridization studies to be different from currently characterized species were tested to determine if a sequence-based classification scheme can clarify their identities. Some of these unusual isolates have been previously reported (43).

	MATERIALS AND METHODS

Top Abstract Introduction Materials & Methods Results Discussion References

Bacterial strains, culture conditions, and DNA preparation. The following type strains, with their corresponding American Type Culture Collection (ATCC) designations and GenBank, EMBL, and/or DDBJ accession numbers, were used to construct the classification scheme: L. adelaidensis (ATCC 49625; U91606); L. anisa (ATCC 35292; U91607); L. birminghamensis (ATCC 43702; U91608); L. bozemanii serogroup 1 (ATCC 33217; U91609); L. brunensis (ATCC 43878; U92227); L. cherrii (ATCC 35252; U91635); L. cincinnatiensis (ATCC 43753; U91636); L. dumoffii (ATCC 33279; U91637); L. erythra (ATCC 35303; U92203); L. fairfieldensis (ATCC 49588; U92204); L. feeleii serogroup 1 (ATCC 35072; U92205); L. geestiana (ATCC 49504); L. gormanii (ATCC 33297; U91638); L. gratiana (ATCC 49413; U92206); L. hackeliae serogroup 1 (ATCC 35250; U92207); L. israelensis (ATCC 43119; U92208); L. jamestowniensis (ATCC 35298; U92228); L. jordanis (ATCC 33623; U92209); L. lansingensis (ATCC 49751; U92210); L. londiniensis (ATCC 49505; U92229); L. longbeachae serogroup 1 (ATCC 33462; X83036); L. longbeachae serogroup 2 (ATCC 33484; AF000958); L. maceachernii (ATCC 35300; U92211); L. micdadei (ATCC 33218; S62141); L. moravica (ATCC 43877; U92212); L. nautarum (ATCC 49506; U92213); L. oakridgensis (ATCC 33761; U92214); L. parisiensis (ATCC 35299; U92215); L. pneumophila serogroup 1 Philadelphia-1 (ATCC 33152; S42595); L. quateirensis (ATCC 49507; U92216); L. quinlivanii (ATCC 43830; U92217); L. rubrilucens (ATCC 35304; U92218); L. sainthelensi serogroup 1 (ATCC 35248; U92219); L. santicrucis (ATCC 35301; U92220); L. shakespearei (ATCC 49655; U92221); L. spiritensis (ATCC 35249; U92222); L. steigerwaltii (ATCC 35302; U92223); L. tucsonensis (ATCC 49180; U92224); L. wadsworthii (ATCC 33877; U92225); and L. worsleiensis (ATCC 49508; U92226). Table 1 presents additional reference type strains and wild strains, used to test the robustness of the classification scheme, and the GenBank, EMBL, and DDBJ accession numbers of sequences determined from those strains during the study. All sequences which were not identical in the region tested to the type strain sequences listed above were submitted. Wild strains consisted of 102 isolates obtained from clinical or environmental specimens from a variety of laboratories within Australia over a period of several years and stored at 70°C and 52 isolates obtained from Europe, Israel, Singapore and the United States. For this latter group of strains, the identification determined by the originating laboratory was used in this study. For some strains, identification was based solely on colony morphology and serological agglutination or immunofluorescence with polyvalent antisera, although cross-reactivity caused some ambiguity. These isolates were still included in the study because they were the only wild strain of a species or the only isolate from a region or because they produced a variant sequence for a species. In-house storage numbers are used to identify strains from which unique sequences were obtained and submitted to GenBank, EMBL, and DDBJ.

Sequencing strategy. From an alignment of the mip sequences of 40 type strains, forward and reverse primers were chosen to target the ribosomal binding site or open reading frame region and the PPIase site, respectively, of the mip gene to amplify a fragment of 661 to 715 bases, depending on the presence and size of the hypervariable region, which is species dependent (31). This equates to nearly 90% of the gene from the 5' end. While the inferred amino acid sequence is highly conserved in the targeted regions, the primers required redundancies to account for variation at the third codon site.

Nucleotide sequence accession numbers. The GenBank, EMBL, and DDBJ accession numbers of sequences determined during this study are listed in Table 1.

	RESULTS

Top Abstract Introduction Materials & Methods Results Discussion References

With Legmip_f and Legmip_r, all type strains gave a single amplification product of the predicted size (661 to 715 bp, depending on the species and the presence of the hypervariable insert), with the exception of L. geestiana, which failed to produce any product. The sequence of each amplicon was determined for the complete length of the fragment, with both Legmip_r and Legmip_fs reliably priming the sequencing reaction to produce unambiguous sequence after approximately 15 bases (to the end of the amplicon). The combination of Legmip_f, as the forward amplifying primer, with Legmip_r to produce the amplicon and Legmip_fs as the forward sequencing primer consistently produced the best results. During the course of this study all amplicons were sequenced in both forward and reverse directions. Amplicon sequences from the type strains were identical to the published sequences (2, 12, 13, 31) and contained sufficient variation (3.6 to 30.5%) to uniquely identify each species.

To test the specificities of the primers, DNA extracted from the following organisms was tested: Staphylococcus aureus, Streptococcus pneumoniae, Escherichia coli, Haemophilus influenzae, Salmonella (multiple serovars), Shigella flexneri, Proteus mirabilis, Aeromonas hydrophila, Pseudomonas aeruginosa, Yersinia enterocolitica, Campylobacter (multiple species), Helicobacter pylori, Corynebacterium diphtheriae, Mycobacterium abscessus, Vibrio parahaemolyticus, and Nocardia asteroides. No amplification was detected, despite the presence of genes coding for Mip-like analogs in some species (19, 20, 30, 44).

Heat lysis proved adequate for extracting DNA for amplification, although boiling or steaming gave slightly more consistent results than microwave heating. Prolonged boiling or steaming for 15 min did not appear to degrade the DNA, as measured by the amount of amplicon produced by subsequent amplification.

Table 1 presents the results for additional reference strains, as well as the comparison of results from the wild strains. This includes both those strains which conform to type strains and those for which speciation cannot be determined, which may represent new species. For a few of these latter strains, amplification produced only limited product which sequenced poorly. Lowering the annealing temperature to 50°C produced more product, which in turn produced longer unambiguous sequence. Sequencing was also attempted after amplification at a below-optimal annealing temperature to produce multiple products of different sizes, and excellent sequence was still obtained with Legmip_fs but not with Legmip_r, as would be predicted. Isolate LC4381 was not able to be amplified, even with the annealing temperature reduced to 40°C.

Figure 1 is a UPGMA phylogenic dendrogram of the inter- and intraspecies similarities.

	DISCUSSION

Top Abstract Introduction Materials & Methods Results Discussion References

For a sequence-based classification scheme to be successful, the gene amplification needs to be specific for the genus but universal for the individual species and the subsequent interspecies sequence variation needs to be sufficient to discriminate clearly between species, even after allowing for the intraspecies sequence variation found among wild strains.

The primers reported here produced single amplicons of the correct size, and in sufficient quantity to enable accurate sequence determination, for all isolates tested, whether they were type strains or wild strains, with the exception of L. geestiana and one unusual, serologically nonreactive, nonidentifiable European wild strain, LC4381. Neither this last isolate nor the type strain or a European wild strain of L. geestiana produced an amplicon of the correct size, even after the annealing temperature was lowered to 40°C. The mip gene sequence for L. geestiana has not been published, and no sequence was determined with the various combinations of primers used by Ratcliff et al. to amplify and sequence this gene in 35 other Legionella species (31). However, a Mip analog has been detected in L. geestiana by using Mip-specific monoclonal antibody (16). Consequently, the lack of amplification for L. geestiana, and LC4381, is likely to be due to sequence variation in the sites targeted by the amplification primers. This was no impediment in our hands, as no locally derived wild strain tested failed to amplify. Similarly, other laboratories should have little problem, as such isolates are likely to be very rare.

Figure 1 shows that all of the remaining species could be easily discriminated and that the wild-strain isolates grouped within the same species as determined by serological, chromatographical, or molecular identification techniques, except for a few where the phenotypic identification was based solely on serology and cross-reactivity prevented unambiguous speciation. Many of these strains had been retained and stored at 70°C because of these nontypical reactions. For instance, LC2763, believed to be a L. bozemanii isolate, also showed strong cross-reactivity with L. parisiensis-specific antiserum. LC3936 reacted equally well with antisera specific for L. cincinnatiensis, L. santicrucis, and both serogroups of L. sainthelensi and L. longbeachae. LC4042 reacted weakly with only L. spiritensis-specific antiserum. However, this isolate was red autofluorescent, suggesting the isolate is likely to be related to either L. erythra or L. rubrilucens, the only red-autofluorescent species so far described. These results demonstrate how ambiguous serological identification can be for some species. The genotypic identification of other strains from the same laboratory, where the serological identification had been confirmed by ribotyping and/or DNA hybridization, was identical in every case.

View larger version (38K): [in this window] [in a new window]   FIG. 1.   UPGMA phylogenetic dendrogram of sequence similarities found among type strains and wild strains of Legionella. The vertical bar joining two isolates or clusters indicates the level of similarity. sg, serogroup; a, Wadsworth 130b, Benidom 030 E, and Bellingham-1 strains; b, Knoxville-1 and Allentown strains; c, Philadelphia-1, Camperdown-1, Heysham 1, OLDA, and Oxford 4032 E strains.

The Legionella genomospecies clustered with L. quinlivanii, with which it has 69% DNA homology (4). With a difference of 3.6%, L. bozemanii and L. tucsonensis showed the least interspecies sequence variation. However, the four L. bozemanii wild strains and L. bozemanii serogroup 2, with an intraspecies variation of 1.3% or less, were easily discriminated from L. tucsonensis. Similarly, L. cincinnatiensis wild strains (maximum intraspecies variation, 1.6%) were easily discriminated from L. santicrucis strains (interspecies variation, 4.1%). Some species, such as L. anisa, L. longbeachae, L. micdadei, and L. londiniensis, for which several wild strains were tested, showed very little sequence difference, suggesting a single homogeneous clonal population, compared to L. jamestowniensis, L. rubrilucens, and L. quinlivanii, which appeared to demonstrate multiple divergent clonal populations. This is especially interesting given that the L. longbeachae strains were isolated on several continents whereas all but one of the L. quinlivanii isolates, including the type strain, came from one region within Australia. The monophyletic nature of L. longbeachae has been demonstrated by other workers (8). L. jamestowniensis, L. rubrilucens, and L. quinlivanii were generally quite divergent from their nearest neighbors, so the greater sequence variation among wild strains did not compromise the resolution of their speciation. This contrasts with chromatographic identification schemes, where ubiquinone and fatty acid profiles do not discriminate easily between some species, for example, L. erythra and L. rubrilucens (43). In fact, it was this lack of discrimination with such methods which was the motivation for this study.

LC0777C is a very interesting European strain. It reacted poorly serologically, and while there was moderate homology with L. gormanii, with which it was genotyped, there was also moderate homology with L. tucsonensis and L. feeleii serogroup 2. With 4.2% sequence variation from L. gormanii, LC0777C demonstrates more genotypic variation from L. gormanii than is typical for wild-strain variation in sister species. This is consistent with its reduced serological reactivity, and it may represent an additional serogroup, a new genotype clone, or even a new species.

However, the multiple serogroups of species where they have been determined could not be confidently discriminated by this scheme in every circumstance. For example, the sequences for L. hackeliae serogroups 1 and 2 are identical. Similarly, many of the serogroups of L. pneumophila are identical to each other, and as a species seem to mainly cluster into two closely related clonal populations, with serogroup 1 strains represented in both. L. pneumophila serogroup 5 and two nonserogrouped wild strains fall outside these two clonal populations. These two strains may represent a new serogroup, since they are both serologically and genotypically different from other L. pneumophila strains. IMVS-D1/77, a serogroup 13 strain, grouped with the serogroup 13 type strain. For L. bozemanii, IMVS-A8E7, a serogroup 2 isolate, and IMVS-A5F7, a serogroup 1 isolate, both grouped with the serogroup 2 type strain. Two other serogroup 2 isolates, IMVS-D2/7 and IMVS-K7B4, produced similar sequences, but they were different from that of the serogroup 2 type strain. Similarly, some serogroup 1 strains clustered uniquely together. Consequently, while these latter genotype groups may be indicative of a particular serogroup, as may some of the unique L. pneumophila serogroup sequences, the differences are small, and it is possible that such differences may be within the normal clonal variation found in the wild-strain population independent of a specific serogroup. The testing of more strains of these species would be necessary to determine if these few differences are characteristic of specific serogroups.

In contrast, the differences in sequences from the two serogroups of L. sainthelensi do appear discriminatory, with two wild strains producing sequences identical to that of the serogroup 2 type strain. Strain LC4261 has been serologically confirmed as L. sainthelensi serogroup 2, although it also demonstrated serological cross-reactivity with L. santicrusis and, to a lesser extent, L. cincinnatiensis and L. longbeachae. Similarly, although the L. longbeachae serogroup 2 type strain differs from the serogroup 1 type strain by only 2 bases, the 26 serogroup 1 strains and one serogroup 2 strain all produced sequences identical to those of their type strains. Therefore, further testing of additional serogroup 2 wild strains may prove this small difference between the two serogroups to be definitive.

Wilkinson et al. (43) reported the existence of strains representing six uncharacterized species, designated species A to F. These strains were amplified, and the products were sequenced to determine how well the genotyping scheme would identify them. Species F is the type strain of L. adelaidensis (3). Species A and B give unique sequences, supporting the ubiquinone and fatty acid profile data indicating that they are indeed new species. Species C, D, and E cluster closely with L. londiniensis, L. waltersii, and L. brunensis, respectively. However, either these three species were not characterized or isolates were not available at the time of testing (4, 10, 42). A comparison of the ubiquinone and fatty acid profiles for species C, D, and E and L. londiniensis, L. waltersii, and L. brunensis confirmed that the scheme had grouped them correctly. The speciation of species C and E has been independently confirmed with DNA-DNA hybridization (2a). In addition to these uncharacterized species, other isolates, stored because they possessed unusual or unique ubiquinone and fatty acid profiles, were also tested. Isolates IMVS-911 and IMVS-960, designated species H, produced sequences identical to each other but unique from all characterized species. Similarly, IMVS-959, designated species I, produced a unique sequence. These three results are in complete agreement with the ubiquinone and fatty acid profiles, and these strains are likely to represent two new undescribed species. Species G isolates (IMVS-823 and IMVS-895) produced identical sequences but grouped moderately close to L. shakespearei (5.4% variation). These two isolates produce ubiquinone and fatty acid profiles similar to those of L. shakespearei, a species which had not been described at the time of their storage (39). The degree of difference from L. shakespearei, however, is greater than the intraspecies variation found in other species, so their true identity needs to be further elucidated. Species J (isolate IMVS-933) has grouped with L. feeleii. While the isolate produces a ubiquinone profile similar to that of L. feeleii (27), the fatty acid profile shows some differences. The significance of this result is unclear, and the identity of this isolate also needs to be further elucidated. Five European wild strains, similarly uncharacterized, were also tested. LC3043 and LC3044, serologically similar to each other, were found to also have sequences identical to each other and identical to those of the two species G strains. Interestingly they demonstrated weak serological homology only with L. shakespearei. The remaining three isolates, designated species L (LC1863), species M (LC4046), and species N (LC4048), were all found to have unique sequences, consistent with the phenotypic classification.

In conclusion, the scheme was able to unambiguously discriminate among 39 of 40 species and correctly group 26 additional serogroups or reference strains within those species. Additionally, 102 wild strains isolated within Australia and 50 wild strains from Europe, the United States, Singapore, Israel, and Kenya, including strains which are thought to represent currently uncharacterized species, were grouped consistently with their phenotypic identification. Two isolates grouped genotypically differently from their phenotypic classification, but it is probable that DNA hybridization would support the genotypic classification. There were no regional clonal variations detected which would indicate that laboratories in other countries would be troubled with ambiguous classification. The scheme is technically simple for a laboratory with even basic molecular capabilities and equipment, if access to a sequencing laboratory is available, especially given that heat lysis is quite adequate for extracting DNA suitable for amplification. Although all wild-strain amplification products were sequenced in both directions during this study, routinely only one direction would be necessary. While both sequencing primers performed extremely well, Legmip_fs would be the preferred sequencing primer, as it proved to be unaffected by nonspecific amplification.

There was no evidence of genetic recombination horizontally across species in the sequences from the 220 strains used in this study, and while they are still theoretically possible, such events would be unlikely to affect the classification of wild strains in practice. However, unusual results or critical isolates could be confirmed by other phenotypic methods or rRNA gene sequences.