Prediction of whole-genome DNA-DNA similarity, determination of G+C content and phylogenetic analysis within the family Pasteurellaceae by multilocus sequence analysis (MLSA)

The family Pasteurellaceae (Pohl, 1981) contains bacterial species that colonize the mucous membranes of the alimentary, genital or respiratory tract of vertebrates, including man and other mammals, birds and reptiles. Most of these bacteria are commensals, but a few are important pathogens, such as: Haemophilus influenzae, causing otitis media and meningitis in man; Pasteurella multocida, the aetiological agent of fowl cholera, haemorrhagic septicaemia in cattle and water buffaloes, and atrophic rhinitis in swine; Mannheimia haemolytica, causing shipping fever in cattle; and Actinobacillus pleuropneumoniae, responsible for porcine pleuropneumonia (Donachie et al., 1995). For a long time, the family only included the old genera Pasteurella, Actinobacillus and Haemophilus. Currently, it contains the additional eight genera Mannheimia, Histophilus, Phocoenobacter, Lonepinella, Gallibacterium, Volucribacter, Avibacterium and Nicoletella. More than 60 validated species are members of the family (). Many more species, as well as genera, are expected to be described in the near future.

The taxonomy of the Pasteurellaceae is complex and at present incompletely resolved. A number of organisms, mainly (but unfortunately not only) those classified in the past, need to be reclassified. Moreover, many taxa await final description and classification, e.g. several genomospecies and numerous Bisgaard taxa (Christensen et al., 2003b). Based on monophyletic clusters observed with 16S rRNA gene (rrs) phylogeny, a minimum of 30 genera are expected for the family. Due to many closely related taxa, phenotypic identification and separation are problematic in certain cases. The requirement for phenotypic characteristics that separate new species from related taxa is often an obstacle for new species descriptions, leaving diagnostic laboratories with unnamed genomospecies. Further complicating a precise classification of new taxa is the need for DNADNA hybridization, which is still regarded as the gold standard for species definition and description. However, this technique is cumbersome, if not impossible, to perform with some taxa. In addition, the method shows a very high variation between experiments and between laboratories. Due to these difficulties and their labour-intensive nature, full matrix hybridizations are rarely carried out, which weakens the significance of the results obtained. Moreover, for each new taxon suggested, cross-hybridization with all the relevant taxa is needed, and the use of reference strains is a prerequisite. Finally, the method only allows investigation at the species level, since its use for investigating genera is less trustworthy, due to its technical limitations and low resolution. Therefore, the ad hoc committee for the re-evaluation of species definition in bacteriology has suggested the development and validation of alternative techniques to DNADNA hybridization (Stackebrandt et al., 2002). However, phylogenetic analysis by rrs sequences, used successfully for years, has its limitations for classification. It does not allow analysis of closely related taxa, and in certain cases contradicts other classification markers. The exploitation of alternative housekeeping genes and their use in multilocus sequence analysis (MLSA) should therefore be investigated (Gevers et al., 2005). Several studies using housekeeping genes have been done within the Pasteurellaceae. They include the use of infB for the genera Haemophilus and Actinobacillus (Hedegaard et al., 2001; Norskov-Lauritsen et al., 2004), the rpoB gene for the entire family (Korczak et al., 2004), and sodA for the genus Pasteurella (Gautier et al., 2005). Investigations have also included comparisons of rrs, rpoB, infB and atpD (Christensen et al., 2004b). These studies have shown that these genes represent useful phylogenetic markers which can be applied in taxonomy, but they are not suitable for the prediction of whole-genome relatedness as it is assessed by DNADNA hybridization. Recently, Zeigler (2003) has shown that a selection of three genes can be used to predict whole-genome relatedness. Comparing whole-genome sequences of 49 bacterial species, he concluded that out of 32 candidate genes investigated, similarities of the genes recN, thdF and rpoA, used in a formula, can be applied to predict confidently the whole-genome similarities among the respective taxa. Based on these findings, we developed a sequencing strategy for recN (encoding a DNA repair protein), thdF (encoding a GTPase) and rpoA (encoding the alpha subunit of the RNA polymerase) within the Pasteurellaceae, and validated their use in MLSA to predict the whole-genome DNADNA similarity, G+C content and phylogeny of selected taxa.

Amplification and sequencing of housekeeping genes.
Amplification was carried out in 50 µl volumes containing 20 pmol each primer, 1 mM dNTP, 1x reaction buffer B (supplied with FIREPol DNA polymerase), 2.5 mM MgCl₂ and 2.5 U FIREPol polymerase (Solis BioDyne). Approximately 100 ng template was added as genomic DNA or as lysate. Cycling conditions were 3 min denaturation at 94 °C, followed by 35 cycles at 94 °C for 30 s, 54 °C for 30 s and 72 °C for 1 min. A final extension step for 7 min at 72 °C was included. PCR products were purified with the High Pure PCR Purification kit (Roche Applied Science). In cases where non-specific bands were observed, the specific band was purified from agarose gels. Finally, about 30 ng purified PCR product was used for sequencing with the BigDye Terminator cycle sequencing kit (Applied Biosystems). Sequences were analysed on an ABI Prism 3100 Genetic Analyser (Applied Biosystems) and then edited using the Sequencher software (GeneCodes).

Sequence determination of rrs, rpoB and infB genes of Pasteurellaceae was done as previously published (Kuhnert et al., 2002, 2004; Korczak et al., 2004).

Sequence analysis.
Sequence similarities of the three genes from different species were calculated using the CLUSTALX program generating a distance matrix (Thompson et al., 1997). Values were converted to a similarity matrix in Microsoft Excel and then used in the formula described by Zeigler (2003):

where SI is sequence identity. The mol% G+C content was calculated from the base composition of the three genes. Phylogenetic analysis was carried out using Bionumerics version 4.0 (Applied Maths). Primer selection and sequence determination
Based on available genome sequences of Pasteurellaceae species, consensus primers were selected for rpoA, recN and thdF. First of all, they had to be as universal as possible for the entire family in order to minimize numbers of primers used for sequence determination. Secondly, in order to obtain optimal information for the calculation of genome relatedness, primers located as close as possible to the 5' and 3' ends of the genes had to be chosen. Finally, the same primers had to be used for PCR and sequencing in order to obtain a simple and straightforward approach. A list of primers used, and their sequences and location, is given in Table 1. With these primers, almost the full length of the rpoA (1 kb) and thdF (1.4 kb) genes, as well as 1.4 kb of the 1.7 kb recN gene, could be obtained. PCR worked very well for all species with the rpoA gene, which is the most conserved of the three genes used. With the less-conserved thdF and recN genes, multiple bands were observed, mainly in the genus Actinobacillus. In these cases, internal sequencing primers were employed, or specific bands were excised and purified from agarose gels. For certain species, optimization of PCR and sequencing primers was necessary. Compared to the problems associated with conventional hybridization, a one-off selection of additional primers and optimization of temperatures for certain taxa represents only a minor problem.

Table 1. Primers used for amplification and sequencing

For recN, in most cases a mixture of the four primers recN-L, recN-L2, recN-R and recN-R2 was used for PCR and for sequencing. Exceptions to this were [Actinobacillus] indolicus and [Haemophilus] parasuis, for which recN-L2/recN-R2 alone gave a PCR product. Primer recN_Vp-R was used instead of the two R primers (recN-R and recN-R2) for Volucribacter psittacicida, Avibacterium gallinarum und Mannheimia ruminalis. For Gallibacterium anatis, internal sequencing primers recN_Ga-1 and recN_Ga-2 were necessary to obtain the full sequence.

For thdF, the primers thdF_first-L2 and thdF_first-R2 were generally used as PCR and sequencing primers. Primers thdF-L2 and thdF-R2 were used as alternative primers during the study, although unlike with recN, never in combination with the other two. Primers thdF-1 and thdF-2 were used as internal sequencing primers for all strains, except for Actinobacillus hominis, Actinobacillus suis, Actinobacillus equuli subsp. haemolyticus, Actinobacillus arthritidis, Actinobacillus genomospecies 1 and Pasteurella multocida subsp. gallicida, for which internal primers thdF-3 and thdF-4 were used. For A. gallinarum and [Actinobacillus] minor, primer thdF_MP-L was used instead of thdF_first-L2 for PCR and sequencing.

The standard annealing temperature for PCR was 54 °C, but in certain cases this had to be lowered to 48 °C (e.g. recN with G. anatis) or increased to 58 °C (e.g. thdF with A. pleuropneumoniae and Mannheimia species).

Genome relatedness of the genera
The three genes rpoA, recN and thdF of 43 strains (37 species and subspecies, including all type species of named genera) from the family Pasteurellaceae were amplified and sequenced. A list of all the strains included and the accession numbers of the sequences are given in Table 2. Based on the obtained sequences, whole-genome relatedness between the species was calculated and is given as similarity values in Table 3. For the first time, cross-comparisons among species and genera were possible. With the classical hybridization method of Brenner et al. (1982), the DNADNA relatedness at species level was set at 70 % (Wayne et al., 1987). Within the family Pasteurellaceae, data have mainly been generated with the spectrophotometric method, and Mutters et al. (1989) have shown that a limit of 85 % for species separation is reasonable. Moreover, a homology above 55 % is set for genus separation with the Pasteurellaceae. However, the use of DNADNA hybridization to investigate less-related organisms (e.g. two different genera) does not provide reproducible results at this level. Application of the presented sequence method has no limitations, since similarity values are calculated and are not dependent on experimental settings. Therefore, genomes of less-related species can be compared and the calculated values can be used for taxonomic purposes. Based upon the data in the lower left of Table 3, preliminary threshold values can be set at around 0.85 for species separation, while the limit for genus separation is about 0.4, i.e. strains showing similarity values below 0.85 almost certainly belong to two different species, whereas strains showing similarity values below 0.4 most likely belong to two different genera. Strains of the same species show similarity values of 0.9 and higher, and species of the same genus have similarity values above 0.4. Similarity values between 0.85 and 0.9 are intermediate, and might indicate different species or subspecies, as observed for the two species M. haemolytica/Mannheimia glucosida (0.88) or with the two subspecies of A. equuli (0.87). These limits for species and genus delineation are preliminary and generalized suggestions, and might change as new data accumulate, or might have to be adapted for certain groups within the family. As DNADNA similarity is used as a part of a polyphasic approach, criteria other than sequence-based prediction of genome similarity should be included for proper classification of such strains. The highest similarity value that can be obtained with 100 % identical sequences in the formula is 0.94. The fact that this value is not 1.0 somehow reflects the possibility of mobile genetic elements changing the overall similarity of genomes. Phages, plasmids and integrons can change a single strain and lead to strain variation, even though the rest of the genome is identical. Thereby, the mathematical model respects the biological fact of genome plasticity within species and even within strains (D. R. Zeigler, personal communication).

Table 2. Strains used in this study and accession numbers of genes

Table 3. Whole-genome DNADNA similarity value matrix based on calculation using the three genes recN, rpoA and thdF (lower left) or recN alone (upper right)

Comparison of the conventional and novel methods within Mannheimia
In order to compare the sequence-based method using the three genes to predict genome similarity with data from classical DNADNA hybridization, we selected the genus Mannheimia (Angen et al., 1999), a recently described genus which has been clearly circumscribed based on a polyphasic approach and contains several species. Moreover, most data have been generated in the same laboratory under the same methodological and experimental conditions, thereby giving the most reproducible values and avoiding the inherent variability observed between different methods and laboratories. Table 4 shows the values published and calculated from the sequences. Regression analysis showed a very good correlation between published and calculated values (r=0.980). Nevertheless, similarity values obtained with the sequence-based method should not be compared directly with those derived from DNADNA hybridization, since each method represents a comparative system of its own. For example, with the hybridization method, M. haemolytica showed the same value of 63 % with M. glucosida and Mannheimia ruminalis, even though the sequence-derived similarity values were 0.88 and 0.74, respectively. All the Mannheimia species showed similarity values above the threshold for genus separation of 0.4. [Pasteurella] trehalosi and Actinobacillus capsulatus had similarity values with M. haemolytica clearly below this limit. Moreover, based on the values published by Angen et al. (1999), Mannheimia granulomatis showed 34 % DNADNA similarity by hybridization with M. ruminalis and A. capsulatus, although the sequence similarity values were 0.48 and 0.34, respectively (Table 3, lower left), placing the two Mannheimia species in the same genus and clearly separating A. capsulatus from the genus Mannheimia. According to the present investigation, the recently described [Mannheimia] succiniciproducens has been misclassified and consequently misnamed. Since it did not show significant similarity values with other genera and also phylogenetically is outside the genus Mannheimia (Table 3, lower left), it should be excluded from Mannheimia sensu stricto. Looking at the species separation based on predicted genome similarity, only the similarity value of 0.88 that M. glucosida shows with M. haemolytica needs additional interpretation. The similarity value was between 0.85 and 0.9, which indicates that the two are very closely related, as previously shown by Angen et al. (1999), and species separation is not obvious according to the sequence-based method alone. M. haemolytica and M. glucosida are also phylogenetically very close (Korczak et al., 2004). However, additional strains of other species should be sequenced to investigate more precisely the genomic similarity at the species level in comparison with previous DNADNA hybridizations.

Table 4. Comparison of sequencing method with hybridization method within Mannheimia and related species NA, Not available.

Analysis of the genus Actinobacillus sensu stricto
To investigate the power of the new approach at genus and species level, the genus Actinobacillus was chosen as another example. This genus is less well defined than Mannheimia, and it is unclear whether certain species, such as A. capsulatus, should be considered to be inside or outside the genus, based mainly on rrs sequences (Olsen & Moller, 2005). In addition, the type species of the genus, Actinobacillus lignieresii, is almost identical to A. pleuropneumoniae (Christensen & Bisgaard, 2004).

Strains belonging to Actinobacillus sensu stricto showed very high similarity values when compared with each other (Table 3, lower left). This included A. capsulatus, which showed a similarity value of 0.74 with the type species A. lignieresii. Based on this figure, A. capsulatus should be classified as a true member of Actinobacillus sensu stricto. This is further supported by earlier DNADNA hybridization studies, as well as rpoB and infB phylogeny (Mutters et al., 1989; Korczak et al., 2004; Norskov-Lauritsen et al., 2004). Further studies that include more strains than the type strain alone need to be carried out in order to finally classify A. capsulatus. [Actinobacillus] minor and [Actinobacillus] porcitonsillarum are closely related to the Actinobacillus sensu stricto cluster, though forming a small branch of their own in rrs as well as rpoB trees (Korczak et al., 2004). In Table 3 (lower left), these two taxa showed similarity values below 0.4 with all members of Actinobacillus sensu stricto. However, between [A.] minor and [A.] porcitonsillarum, the similarity value was 0.79, which would combine them in their own genus, although separating them as two species.

The other two porcine Actinobacillus species [A.] indolicus and [Actinobacillus] porcinus included in our study were clearly outside the Actinobacillus sensu stricto cluster, and did not show genetic relatedness with each other. However, [A.] indolicus showed a high similarity value of 0.6 with [H.] parasuis, indicating that these two species form a genus of their own.

Variation within species
In order to examine intraspecies variation we investigated a few serotypes of A. pleuropneumoniae. As seen in Table 3 (lower left), the obtained similarity values were clearly above 0.9. A. pleuropneumoniae and A. lignieresii also showed a similarity value above 0.9 and could be classified as subspecies of A. lignieresii, corresponding to major differences in disease manifestations. Intraspecies variability can also be seen with P. multocida and its subspecies. Whereas P. multocida subsp. gallicida showed an even higher similarity value to the type species P. multocida subsp. multocida than strain Pm70, P. multocida subsp. septica had a similarity value below 0.9 but above 0.85, which leaves it open whether it is a subspecies or a species of its own, as argued by several recent papers (Kuhnert et al., 2000; Davies, 2004).

recN alone
Sequence determination of all three genes might be cumbersome. For that reason we investigated whether the recN sequence alone might be representative of all three genes, as also proposed by Zeigler (2003). The similarity values are shown in the upper-right panel of Table 3. Negative values were obtained in this approach. This mathematical artifact has no further meaning and only indicates a rather low genetic relatedness between the corresponding taxa. When performing a regression analysis of all similarity values calculated using the three genes versus the values calculated using recN alone we found a very good correlation (r=0.978). Threshold similarity values for genus and species boundaries are comparable, with very few discrepancies (Fig. 1). Therefore, recN could be used as an initial means of comparing genomes of the Pasteurellaceae although, as demonstrated by Zeigler (2003), with a lower confidence. For identifying homogeneous species or separation of distantly related organisms it might be sufficient, but for better resolution, the sequence of all three genes would be necessary.

(10K): Fig. 1. Correlation of whole-genome similarities calculated by recN alone or by including all three genes recN/rpoA/thdF. The linear correlation coefficient (r) was 0.978. Circles indicate possible genus and species boundaries at similarity values of around 0.4 and 0.85, respectively.

mol% G+C content
The mol% G+C content should ideally be part of the description of a new taxon (Stackebrandt et al., 2002). Again, methods to determine this value can be different (e.g. denaturation temperature or HPLC), and obtained values may differ significantly not only between strains but also between experiments and laboratories. Therefore, published G+C content within a single species varies by up to 5 % for biological as well as methodological reasons (Goodfellow et al., 1997; Olsen et al., 2005). We calculated and summed up the G+C content of the three genes rpoA, recN and thdF and compared it to available published values (Table 5). A mean difference of 1.2 % was observed between published and calculated G+C contents. The difference between calculated and published values did not exceed 3 %, thereby being within the range of variation seen with conventional methods. The extremely high difference observed with [A.] indolicus (more than 7 %) was surprising (Moller et al., 1996). It should be noted that the published G+C content of only 35.5 % is very low compared to the other members of the family, and especially compared to [H.] parasuis (4142 %), to which [A.] indolicus is most closely related. Therefore, since the G+C content has until now only been determined from the type strain, this difference should be questioned. Interestingly, in the comparison with published genome sequences we saw that the calculated values were generally slightly (about 1 %) higher. One could therefore argue that a correction of 1 % should be considered. Employing housekeeping genes for G+C content calculation may reflect the true characteristics of the core genome of a species more accurately, since it is not affected by the variability that might result from horizontally transferred genomic elements, which often show variation from the overall G+C content. In addition to the data on whole-genome similarity, the G+C content calculation showed that the three genes rpoA, recN and thdF can be regarded as genome-representing genes. The other protein-encoding genes rpoB and infB, alone or in combination, could not be used for this purpose (data not shown).

Table 5. G+C content comparison between calculation based on the three genes and published values

Phylogeny based on six housekeeping genes
We have previously used the rrs, rpoB and infB genes for phylogenetic analysis within the family Pasteurellaceae (Kuhnert et al., 2004; Korczak et al., 2004). Combining these with the three additional genes we investigated the analysed strains in a multilocus sequence phylogeny (MLSP). Individual phylogenetic trees of each gene were constructed and are available as supplementary data. From the six trees, a consensus tree was generated in Bionumerics (Fig. 2).

(28K): Fig. 2. Consensus tree of six genes. Individual trees of rrs, rpoB, infB, rpoA, recN and thdF were used to generate a combined tree built in Bionumerics. JukesCantor correction was applied for the distance matrix and neighbour-joining for tree construction. Cophenetic correlations are given, indicating the reliability of the branching compared to the actual genetic relatedness of the taxa.

The clusters observed corresponded well with the current genera and indicated putative new genera that will have to be described in the future. The clusters also supported the DNADNA similarity values obtained from the sequencing method. For example, the species [M.] succiniciproducens was clearly outside the Mannheimia cluster. This is underlined by the very low similarity value of 0.15 of this species to the type species M. haemolytica (Table 3, lower left, and Table 4). Within the Mannheimia cluster M. granulomatis showed the lowest similarity value (0.48) with the type species. This is in agreement with the position in the tree. However, the two species M. haemolytica and M. glucosida, which showed a high similarity value of 0.88, clustered very closely, like the subspecies of A. equuli or P. multocida. The same was observed for A. linieresii and A. pleuropneumoniae, which also had a very high similarity value above the upper species threshold of 0.9. Moreover, the phylogenetic analysis using all six genes clustered A. capsulatus clearly inside Actinobacillus sensu stricto. On the other hand, [A.] minor and [A.] porcitonsillarum were outside this cluster, separated from it by [Haemophilus] ducreyi, which could already be assumed based on the low similarity values that these two species showed with the sensu stricto species (Table 3, lower left). Based not only on the similarity values but also on the relatively deep branching of the two, they should be classified as two different species, eventually forming their own genus. The analysis of more strains and species is necessary in order to reach a final classification of [A.] minor and [A.] porcitonsillarum. The tree also represented the close relatedness of [A.] indolicus and [H.] parasuis, supporting the observation from the similarity analysis which indicated that they belong to the same genus. The very deep branching of Pasteurella canis in relation to P. multocida is also worth mentioning, since this species showed a similarity value of 0.41, which is at the borderline for genus separation. The position of G. anatis was the most isolated in the tree, which is again reflected by the similarity values observed for this species with all the others analysed in our study. It is the species that showed the lowest similarity values, ranging from 0.06 to 0.17 (Table 3, lower left). All the other species had significantly higher similarity values on average. Therefore, the phylogenetic tree based on six genes and the similarity values calculated are in very good agreement.

The congruence between the individual trees (see supplementary data) was calculated by Pearson correlation in Bionumerics. Results are presented graphically as well as in tabular form in Fig. 3. From this it can be deduced that recN most properly represents the consensus tree (MLSP), indicated by the high correlation between the two (97.1 %). The other genes, rpoB, infB, rpoA and thdF, had lower values between 92 and 95 % with the consensus tree. Overall, the three genes recN, rpoA and thdF had tree topologies that were closest to that of the consensus tree. This is further evidence that these genes are highly representative for the entire genome. The genes rpoB and infB shared some common tree topology, as shown in Fig. 3. The most divergent tree, showing the lowest correlations to all the other trees, was the one derived from rrs. This might be explained by the different nature of this gene, since it is not protein coding. Nevertheless, this finding is also surprising, since rrs is regarded as the gold standard for phylogenetic analysis of bacteria. However, at least within the Pasteurellaceae, it seems as if this gene is less representative than any of the others used in our study. This has previously been observed within the family Pasteurellaceae as well as the genus Campylobacter from the fact that rpoB-based phylogeny is in better agreement with the results of DNADNA hybridization than rrs-based phylogeny (Korczak et al., 2004, 2006). Therefore, the rrs gene might be suitable for analysing distantly related taxa, but genes other than rrs might prove to be more informative for phylogeny and taxonomy in other bacterial families and genera.

(12K): Fig. 3. Congruence of trees. The matrix of congruence values (Pearson correlation coefficient) between the genetic distances derived from the six genes as well as the consensus tree (MLSP) is given. A dendrogram derived from this matrix is shown on the left.

Conclusion
From the data presented we conclude that MLSA using the three genes recN, rpoA and thdF represents a valid approach for investigating the taxonomy of Pasteurellaceae. The obtained results can be used to calculate whole-genome similarity, they give a very good estimate of the G+C content of a species, and, finally, the genes can be used alone or in combination as phylogenetic markers, allowing a comprehensive analysis of the phylogenetic position of a strain. The results obtained by the sequence-based method are congruent with data from conventional DNADNA hybridization at the species level with respect to whole-genome similarity as well as G+C content. We therefore propose to add the MLSA-based approach to those previously used, as recommended by leading taxonomists (Stackebrandt et al., 2002; Gevers et al., 2005). Complete matrix hybridizations between a new putative species and validated species and genera are no longer needed. A simple sequence analysis of selected strains of a new species and calculation of similarity with already-sequenced taxa is enough, and data are highly comparable between laboratories. This would also favour the creation of databases for this purpose, which will allow taxonomists to compare their new taxa at a remote location via the internet. The exchange of reference strains will no longer be necessary, and the new approach could boost the reorganization of the family Pasteurellaceae, which is important in the light of many misclassified species and unnamed taxa. Finally, the universal validity of the approach described should be tested with other bacterial families. We thank Magne Bisgaard for critical reading of the manuscript and valuable discussions. This work was supported by a grant from the Innovation Promotion Agency of Switzerland (KTI) (no. 6041.1 KTS).