Identification of lactobacilli by pheS and rpoA gene sequence analyses

Lactic acid bacteria (LAB) belonging to the genus Lactobacillus comprise the largest group of Gram-positive, rod-shaped and catalase-negative organisms (Hammes & Vogel, 1995) with Lactobacillus delbrueckii as the type species (Kandler & Weiss, 1986). Species of the genus Lactobacillus form part of the normal flora of the gastrointestinal tract, vagina and oral cavity of humans and animals (Hammes & Vogel, 1995; Klein et al., 1998). Lactobacilli are of great economic importance for the dairy and other fermented food industries, where they are used as starter cultures for fermenting raw materials of vegetable or animal origin. Lactobacillus species are claimed to have health-promoting (probiotic) properties and some pharmaceutical preparations contain viable Lactobacillus strains (Holzapfel et al., 2001; Reid, 1999; Stiles & Holzapfel, 1997). In this context, the accurate identification of members of the genus Lactobacillus remains a point of crucial importance.

Several methods have been used for the identification of lactobacilli to the species level, e.g. SDS-PAGE of whole-cell proteins, randomly amplified polymorphic DNA (RAPD), amplified fragment length polymorphism (AFLP), rep-PCR and ribotyping (Daud Khaled et al., 1997; Gancheva et al., 1999; Gevers et al., 2001; Massi et al., 2004; Pot et al., 1993; Yansanjav et al., 2003). Although useful, there are some pitfalls associated with the use of these methods concerning portability, inter-laboratory reproducibility and time efficacy. Informational genes such as the 16S rRNA gene are commonly considered as reliable phylogenetic markers for assigning evolutionary relationships among species of the genus Lactobacillus (Schleifer & Ludwig, 1995). However, 16S rRNA gene sequence data do not allow the identification of closely related species. The use of housekeeping genes is emerging as an alternative to overcome these problems (Santos & Ochman, 2004; Stackebrandt et al., 2002). Recent in silico studies based on complete genomes have provided the basis for establishing sets of housekeeping genes that can accurately predict genome relatedness and improve the accuracy of species identification. The need for alternative genomic markers that provide higher levels of discrimination than the 16S rRNA gene has led to a more systematic sequencing of housekeeping genes (Coenye et al., 2005; Gevers et al., 2005; Konstantinidis & Tiedje, 2005; Naser et al., 2005a, b; Thompson et al., 2005; Zeigler, 2003).

To be useful for species discrimination, genes must ideally be present in a single copy, evolve more rapidly than rRNA genes and be widely distributed among bacterial genomes. Those genes in which recombination might confer a selective advantage, or closely linked genes, should be avoided. Furthermore, these genes should be informative with an adequate degree of resolution and provide sufficient variability to differentiate species of a particular genus (Zeigler, 2003).

The use of the housekeeping genes that code for the α-subunit of bacterial phenylalanyl-tRNA synthase (pheS) and the α-subunit of RNA polymerase (rpoA) has proven to be a robust system for the identification of all the recognized species of the genus Enterococcus (Naser et al., 2005b). As it is our intention to extend the application of these protein-coding loci to all other LAB genera, the present study was aimed at evaluating the usefulness of pheS and rpoA gene sequences as alternative genomic tools for the identification of species of the genus Lactobacillus. We compared the sequence data of the pheS and rpoA genes with the available 16S rRNA gene sequences. In addition, a software tool, named TaxonGap, was developed during this study to enable a straightforward evaluation of the discriminatory power of the individual genes in the Lactobacillus identification scheme.

Two hundred and one well-characterized Lactobacillus strains representing 98 species and 17 subspecies of the genus Lactobacillus isolated from humans, animals or food products were analysed in this study (Table 1). Strains were grown on MRS agar media (Oxoid) at 37 °C for 48 h. All strains included in this study have been deposited in the BCCM/LMG Bacteria Collection at Ghent University (Ghent, Belgium). Bacterial genomic DNA was extracted as described by Gevers et al. (2001) or DNA alkaline extract was used (Niemann et al., 1997). The amplification and sequencing of pheS and rpoA genes were as described by Naser et al. (2005a, b) with the following modifications: where an amplicon was not obtained with the referred conditions, the primer combination rpoA-21-F/rpoA-22-R (5'-ATGATYGARTTTGAAAAACC-3'/5'-ACYTTVATCATNTCWGVYTC-3') was used for the amplification of the rpoA gene and/or the Failsafe PCR system (Epicenter).

Table 1. Details of the Lactobacillus species and strains that were analysed in this study

Consensus sequences were determined as described by Naser et al. (2005a, b). The CLUSTAL_X program was used for multiple sequence alignment. Consequently, the aligned sequences were imported into BioNumerics software version 4.5 (Applied Maths) for the calculation of similarity matrices and neighbour-joining trees (Saitou & Nei, 1987). The reliability of hierarchical clustering was determined by using the bootstrapping method with 1000 resamplings. The 16S rRNA gene sequence data of the Lactobacillus type strains were obtained from EMBL.

TaxonGap software tool.
When evaluating multiple genes as candidate biomarkers for the identification of different operational taxonomic units (OTUs) (Sneath & Sokal, 1973), one is intuitively looking for molecular markers that show the least amount of heterogeneity within OTUs and also result in maximal separation between the different OTUs. The first requirement must guarantee that members of the same OTU have the same (or at least similar) biomarkers, so that they can easily be grouped together based on those markers. The second requirement is that members of different OTUs must have sufficiently different biomarkers so that an evaluation of these markers cannot erroneously suggest assignment of the members to the same OTU. The TaxonGap software tool was specially designed to produce a compact representation of the resolution of the biomarkers within and between taxonomic units, allowing easy and reliable inspection of the data for evaluations across the different OTUs and the different biomarkers.

For a given set of OTUs O₁, O₂, . . ., O_n, the s-heterogeneity within the taxon O_i (i=1, . . ., n) is defined as max_{x,y∈Oi, x≠y} d_s (x, y). Herein, d_s (x, y) represents the distance between the (different) members x and y of the taxon O_i as measured from the biomarker s. Likewise, the s-separability of the taxon O_i (i=1, . . ., n) is defined as min_{x∈Oi,y ∉ Oi} d_s (x, y). The taxon containing y, for which the minimum distance is reached during the calculation of the s-separability, is called the closest neighbour of the taxon O_i. Note, however, that the closest neighbour relationship is not necessarily symmetric; given that O_i is the closest neighbour of O_j, it does not automatically follow that O_j is also the closest neighbour of O_i. The calculation of the s-heterogeneity and the s-separability are schematically represented in Fig. 1 for a taxon A and its closest neighbouring taxon B.

(12K): Fig. 1. TaxonGap versus intraspecies diversity. A schematic representation of diversity within and between two species A and B. Dots represent operational taxonomic units (OTUs) in evolutionary space, with the distance between dots relative to the distance derived from sequence information. The intraspecies diversity is indicated by the light grey arrows and represents the maximum sequence distance between strains of the species (corresponds to the light grey bars in Fig. 2). The TaxonGap, indicated by the dark grey arrow, represents the distance between species A and its closest neighbouring species, species B (corresponds to the dark grey bars in Fig. 2).

The TaxonGap software tool calculates the matrix of s-heterogeneity and s-separability values with the different OTUs as the matrix rows and the different biomarkers as the matrix columns. Headers are placed to the left and on top of the matrix. The rows and columns of the matrix can be placed in any order. However, to improve interpretability of the resulting representation, we have included the option to present the OTUs according to their position in a phylogenetic tree as an alternative to listing them in alphabetical order. Again, with the aim of improving the visual inspection and interpretation of the data, the TaxonGap software tool presents the s-heterogeneity and s-separability values as light grey and dark grey horizontal bars, respectively. The same scaling is used for plotting the s-heterogeneity and s-separability bars for the individual biomarkers in order to support optimal comparability of the values across the biomarkers. The name of the closest neighbour is attached to the right side of the dark grey bar. Light grey bars are printed on top of the dark grey bars and are made slightly thinner than the dark grey bars to improve visualization even when the light bars grow larger than the dark bars. The latter only occurs in the rare occasion when, for a given biomarker, members in a taxon are more distant to each other than a member of the taxon is to a member of another taxon. Although not a strict requirement, it is advised that the same OTUs are used for the evaluation of different biomarkers. Missing biomarker data for a given OTU leads to holes in the TaxonGap output matrix. There is no requirement to use the same OTU members for measuring different biomarkers.

Distances used for the calculation of the s-heterogeneity and s-separability values were determined using pairwise nucleotide sequence alignments with the Needleman-Wunsch algorithm as implemented in the BioNumerics 4.5 software package.

Application of TaxonGap for the evaluation of pheS and rpoA gene sequences as biomarkers for species identification
Fig. 2 shows the TaxonGap output for the Lactobacillus identification scheme discussed in the present study. The OTUs subjected to the TaxonGap analysis were the different species of the genus Lactobacillus. Cases where species synonymy has been reported in the literature were regarded as a single species during the TaxonGap analysis. The biomarkers were the pheS, rpoA and 16S rRNA genes. The s-heterogeneity is a measure of the heterogeneity observed in the biomarker s among the different strains of the same Lactobacillus species (subsequently referred to as intraspecies heterogeneity). The s-separability is a measure of the divergence between the different Lactobacillus species (subsequently referred to as interspecies divergence). Subspecies were not taken into account during this analysis as it was evident from the data that few subspecies could be separated by the biomarkers studied. Where a given gene was able to make clear separation between subspecies, it is indicated in the discussion of the different phylogenetic groups below.

(64K): Fig. 2. Representation of the discriminatory power of the genes for species identification of the genus Lactobacillus. The left panel shows a neighbour-joining tree of the complete 16S rRNA gene sequences of the Lactobacillus type strains, including species groups. EMBL accession numbers of the 16S rRNA gene sequences are indicated in parentheses. For each of the species in the phylogenetic tree, the right panel depicts the intraspecies variability for pheS and rpoA genes and interspecies variability for the 16S rRNA, pheS and rpoA genes as horizontal light grey and dark grey bars, respectively. Overall distance gaps between species are represented in the graphic as lines. The right panel also contains the names of the closest relatives as estimated from the different loci.

The members of the genus Lactobacillus were ordered according to their phylogenetic positioning in a neighbour-joining tree calculated from the 16S rRNA gene sequences of their type strains. The different Lactobacillus species groups are delineated on the left of the neighbour-joining tree. Although heterogeneity could not be estimated for the 16S rRNA gene as sequence data were only available for the type strains, the separability of the Lactobacillus species based on the 16S rRNA gene was added as the first column of the TaxonGap output matrix. This allows better evaluation of the discriminatory power of the 16S rRNA gene for species identification when compared with the other genes included in the identification scheme. The pheS and rpoA genes formed the second and third biomarker columns in the TaxonGap output matrix. In order to guide the readership in the interpretation of the TaxonGap output in the following discussion, we focus on the first row of Fig. 2. From this row, we can determine the s-heterogeneity and s-separability values for the L. agilis species. For this species, the observed pheS-heterogeneity was 1.5 % (see light grey bar), whereas the rpoA heterogeneity only reached 0.3 % for the same species. Likewise, one can see that the closest neighbour of the L. agilis species is estimated differently for the 16S rRNA gene (L. equi; 4.9 %), the pheS gene (L. animalis; 17.3 %) and the rpoA gene (L. acidipiscis; 15.7 %). However, it should be noted that all of these species belong to the same L. salivarius species group. This is an example of the general trend observed in the dataset: that when species have different closest neighbours for the genes in the identification scheme, these species all belong to the same species group.

The representation produced by the TaxonGap software tool offers a number of advantages over comparing individual trees for the different gene sequences included in polygenic identification studies. First of all, a separate row is reserved in the TaxonGap output matrix for the heterogeneity and separability values of the different genes for each species, which is not the case when comparing phylogenetic trees. Even after the tedious process of swapping branches, it is not always possible to draw phylogenetic trees in a way that enables clear visual comparisons to be made. This is especially the case when trees for multiple genes need to be compared. In addition, TaxonGap uses the same scaling for depicting the distance values based on the different gene sequences. Few software tools for drawing phylogenetic trees allow precise control over the scaling. Both placement and scaling improve the comparability of the heterogeneity and separability for individual species. Secondly, we want to point out that phylogenetic trees present approximations of the underlying distance values whereas the TaxonGap filters out original similarity values instead of approximations by using minimum and maximum as aggregation operators. This is important when comparing s-heterogeneity and s-separability for all species for a given gene s. To underscore the overall success rate of the individual genes to discriminate between species of the genus Lactobacillus, we have depicted the overall heterogeneity (light grey) and separability (dark grey) per species as vertical lines for each gene in Fig. 2. Finally, the graphical output of TaxonGap remains compact, even for datasets where the number of OTU members grows large. This is because the software has a built-in aggregation based on the individual OTUs. Representing phylogenetic trees with over a few hundred entries would be almost impossible in printed format.

The TaxonGap software tool thus allows for a more straightforward evaluation of the discriminatory power of the individual genes in the Lactobacillus species identification scheme, as opposed to the need to compare separate gene trees drawn for each of the genes in the scheme.

Robustness of pheS and rpoA partial gene sequences for Lactobacillus species identification
The success of any bacterial species identification system depends on accuracy. Accuracy allows the distinction between intraspecific variation and interspecific divergence in the selected loci. The less overlap there is between genetic variation within species and divergence from species, the more effective the system becomes (Meyer & Paulay, 2005).

Both the pheS (382–455 nt) and rpoA (402–694 nt) partial gene sequences were applied as alternative genomic markers for the identification of Lactobacillus at the species level. Two hundred and one well-characterized Lactobacillus strains representing 98 species and 17 subspecies of the genus Lactobacillus from different origins were analysed in this study (Table 1). The strains were selected on the basis of previous polyphasic classification using AFLP, RAPD-PCR and SDS-PAGE of whole-cell proteins and represent the known heterogeneity of Lactobacillus species. In order to evaluate the pheS and rpoA gene sequence variations at the intraspecies level, we included several representative strains for each Lactobacillus species. In general, the pheS and rpoA gene sequences showed intraspecies variations up to 3 % and 2 %, respectively (Fig. 2).

The differentiating power of the pheS and rpoA partial gene sequences was examined for Lactobacillus species at the subspecies level. In general, the subspecies of Lactobacillus were highly related, having 98–100 % pheS and rpoA gene sequence similarities. This shows that the discriminatory power of the investigated loci to differentiate between the subspecies of most lactobacilli is low. However, pheS gene sequences could differentiate between the subspecies of L. sakei and L. plantarum (see below).

The analysis of pheS and rpoA partial gene sequences clearly differentiates the members of the genus Lactobacillus (see also Supplementary Figs S1 and S2 available in IJSEM Online). In comparison with the 16S rRNA gene, our data clearly indicate that pheS and rpoA genes provide higher resolution for differentiating Lactobacillus species. As shown in Fig. 2, both pheS and rpoA partial gene sequences provide alternative reliable genomic markers to differentiate the members of the genus Lactobacillus. However, it should be mentioned here that both pheS and rpoA partial gene sequences showed a variable discriminatory power for identifying different species of the genus Lactobacillus. An example that illustrates the variation of the pheS and rpoA partial gene sequences in their degree of resolution is shown in Fig. 2 between the type strains of L. acidifarinae and L. zymae (L. buchneri group).

The pheS gene sequence analysis provided the highest discrimination for the identification of different species of lactobacilli. The case of L. antri and L. oris (L. reuteri group) is an exception here where the rpoA gene provided more resolution than the pheS gene in differentiating the two species. The pheS gene sequence analysis provided an interspecies gap, which normally exceeds 10 % divergence and an intraspecies variation up to 3 %. The rpoA gene sequences revealed a somewhat lower resolution with an interspecies gap normally exceeding 5 % and an intraspecies variation up to 2 %.

It should be mentioned that the variation of the investigated genes in their discriminatory power, together with the fact that different genes might provide different closest neighbours or topologies without hampering their use to unambiguously circumscribe bacterial species, validated the necessity for the simultaneous analysis of several protein-coding loci for a robust taxonomic analysis at the species and genus levels.

Species groups based on 16S rRNA gene similarity
The currently recognized phylogenetic relationships within the genus Lactobacillus have been determined by comparative analysis of their 16S rRNA gene sequences (Schleifer & Ludwig, 1995). Based on these data, different phylogenetic species groups have been distinguished: the L. acidophilus, L. reuteri, L. buchneri, L. alimentarius, L. plantarum, L. sakei, L. casei and L. salivarius species groups.

On the basis of pheS gene sequence analysis, members of the L. reuteri, L. alimentarius, L. plantarum, L. sakei and L. casei species groups clustered together in clades corresponding with the 16S rRNA gene based phylogeny (see Supplementary Fig. S1 in IJSEM Online), whereas members of the L. acidophilus, L. buchneri, and L. salivarius species groups are clustered in two separate clades. On the basis of rpoA gene sequence analysis, the L. acidophilus, L. reuteri, L. alimentarius, L. plantarum, L. sakei, L. casei and L. salivarius species groups clustered together in clades corresponding with the 16S rRNA gene based phylogeny whereas the L. buchneri species group clustered in two separate clades (see Supplementary Fig. S2 in IJSEM Online).

In subsequent sections, we will discuss and compare our data and the data from the literature for all species of the genus Lactobacillus on the basis of the species groups delineated by the 16S rRNA gene phylogeny.

L. acidophilus species group
Within the L. acidophilus species group, the pheS and rpoA gene sequence data clearly differentiate the members of the L. acidophilus group with a maximum of 94 % and 98 % pheS and rpoA gene sequence similarities, respectively, except for L. kitasatonis and L. amylovorus (with 98.5 % and 99 % pheS and rpoA gene sequence similarities, respectively). At the intraspecies level, strains of same species were highly related (>98 % pheS and rpoA gene sequence similarities). However, as an exception, the neighbour-joining tree based on pheS gene sequences revealed distinct subclusters among strains of the species L. gasseri (8 strains) having 95 % pheS gene sequence similarity and among strains of the species L. johnsonii (9 strains) having 96 % pheS gene sequence similarity (results not shown). The heterogeneity within L. gasseri strains was also observed by comparing the fluorescent amplified fragment length polymorphism (FAFLP) fingerprints of these strains with reference profiles of lactic acid bacteria taxa (unpublished data).

The neighbour-joining trees derived from the pheS and rpoA gene sequences revealed close relatedness between L. helveticus and L. suntoryeus, with at least 99.5 % pheS and rpoA gene sequence similarities (see Supplementary Figs S1 and S2). In addition, sequence analysis of the gene that codes for the α-subunit of ATP synthase (atpA) also showed a high relatedness between the two species. Further genomic data derived from DNA–DNA hybridization unambiguously demonstrated that L. suntoryeus is a later synonym of L. helveticus (Naser et al., 2006a).

The pheS and rpoA partial gene sequences revealed heterogeneity among culture collection strains of L. amylophilus described by Nakamura & Crowell (1979). Strains LMG 11400 and NRRL B-4435 represent a separate lineage that is distantly related to the type strain of L. amylophilus LMG 6900^T and to three other strains of the species (NRRL B-4438, NRRL B-4439 and NRRL B-4440). The pheS and rpoA gene sequence data showed that strains LMG 11400 and NRRL B-4435 constituted a distinct cluster, showing 100 % pheS and rpoA gene sequence similarities. The other reference strains clustered together with the type strain of L. amylophilus LMG 6900^T and were clearly differentiated from strains LMG 11400 and NRRL B-4435 (80 % and 89 % pheS and rpoA gene sequence similarities, respectively). Further phenotypic and genotypic research confirmed that both strains represent a novel taxon, for which the name Lactobacillus amylotrophicus has been proposed (Naser et al., 2006b).

L. alimentarius species group
Within the L. alimentarius group, the pheS gene sequence similarity between L. kimchii and L. paralimentarius is 92 %, whereas on the basis of rpoA gene sequences, the two species show high relatedness, having 98.5 % rpoA gene sequence similarity. The pheS gene reflects a fast-evolving evolutionary clock that shows a finer resolution than the rpoA gene at both the intraspecies and interspecies levels in most cases. In support of the distinct genomic relatedness between L. kimchii LMG 19822^T and L. paralimentarius LMG 19152^T, De Vuyst et al. (2002) reported a DNA–DNA reassociation value of 68 %. Such a hybridization value is considered to be at the borderline for species delineation. The pheS gene sequence data indicates that L. kimchii and L. paralimentarius are separate species.

L. buchneri species group
Both pheS and rpoA gene sequence analyses showed that the members of L. buchneri species group are clustered in two subclades (see Supplementary Figs S1 and S2). An interesting relationship confirmed by the simultaneous analysis of pheS and rpoA gene sequences is the high genomic relatedness of L. parabuchneri LMG 11457^T and L. ferintoshensis LMG 22038^T (100 % pheS and rpoA gene sequence similarities). Recently published data are in complete accordance with the pheS and rpoA gene sequence data. Vancanneyt et al. (2005) confirmed this finding and demonstrated that these taxa are synonymous species, based on a polyphasic study.

Representative strains of L. brevis, LMG 6906^T, LMG 11435, LMG 7761, LMG 11494 and LMG 11984, were investigated. The pheS gene sequence analysis showed that strains LMG 11494 and LMG 11984 constituted a distinct cluster separated from the type strain of L. brevis with a sequence similarity of less than 82 % (see Supplementary Figs S1 and S2). 16S rRNA gene sequence analysis showed that both strains belong to the L. buchneri group with nearest neighbours L. hammesii and L. brevis (sequence similarities of 99.2 and 98.1 %, respectively). Strains LMG 11494 and LMG 11984, isolated from cheese and wheat, respectively, showed 99.9 % pheS gene sequence similarity. It has recently been confirmed that both strains represent a novel taxon, for which the name L. parabrevis was proposed (Vancanneyt et al., 2006).

L. casei species group
Difficulties in the accurate identification of species belonging to the L. casei species group have been reported (Tynkkynen et al., 1999; Zhong et al., 1998). A study by Mori et al. (1997) found high 16S rRNA gene sequence similarity between the members of L. casei species group (>99 %). In the present study, L. rhamnosus, L. casei and L. paracasei were clearly distinguished on the basis of pheS and rpoA genes. Apart from L. casei and L. zeae (see below), these species have a maximum of 84 % and 95 % pheS and rpoA gene sequence similarities, respectively. This result further emphasizes the discriminatory power of the housekeeping genes investigated in this study.

Within the L. casei species group, the pheS gene sequence similarity between L. casei LMG 6904^T (=ATCC 393^T) and L. zeae LMG 17315^T (=ATCC 158520^T) was 93 %, whereas on the basis of rpoA gene sequences, the two species were more highly related, having 99 % gene sequence similarity. In addition, the sequence analysis of the gene that codes for the α-subunit of ATP synthase (atpA) also showed a high relatedness (96 %) between the two species (data not shown). Data from the literature were in complete accordance with the present data and supported the high relatedness found between these two taxa. Further genomic data derived from recA gene sequence analysis and high DNA–DNA reassociation values (80 %) demonstrated that both species are members of the same species (Dicks et al., 1996; Felis et al., 2001) and supported the reclassification of L. casei as L. zeae (Dellaglio et al., 2002). This example strongly supports the simultaneous use of multiple loci.

L. plantarum species group
16S rRNA gene sequences are not suitable for definitive differentiation of the members of L. plantarum species group due to the high gene sequence similarity (>99 %) between L. plantarum, L. paraplantarum and L. pentosus (Collins et al., 1991; Torriani et al., 2001). Our data clearly showed that pheS and rpoA gene sequences had a high discriminatory power in differentiating L. plantarum, L. paraplantarum and L. pentosus with a maximum 90 % and 98 % pheS and rpoA gene sequence similarities, respectively. At the subspecies level, the neighbour-joining tree based on the pheS gene sequences showed that L. plantarum subsp. plantarum and L. plantarum subsp. argentoratensis were clearly differentiated from each other (91 % pheS gene sequence similarity) (see Supplementary Fig. S1). L. plantarum LMG 6907^T and L. arizonensis LMG 19807^T were highly related with >99.5 % pheS and rpoA gene sequence similarity. Kostinek et al. (2005) showed that L. arizonensis is a later heterotypic synonym of L. plantarum because the type strain of L. arizonensis NRRL B-14768^T (=DSM 13273^T) is not distinguishable from the L. plantarum type strain DSM 20174^T on the basis of ribotyping patterns, rep-PCR fingerprinting patterns, 16S rRNA gene sequences or DNA–DNA hybridization data.

L. reuteri species group
Within this species group, high degrees of similarity exist between L. ingluviei LMG 20380^T and L. thermotolerans LMG 22056^T (99 % and 100 % pheS and rpoA gene sequence similarities, respectively) as well as between L. durianis LMG 19193^T and L. vaccinostercus LMG 9215^T (99 % and 98 % pheS and rpoA gene sequence similarities, respectively). A study recently conducted by Felis et al. (2006) confirmed that L. thermotolerans is a later synonym of L. ingluviei. Representative strains of L. durianis and L. vaccinostercus were further investigated. Genomic data derived from FAFLP and DNA–DNA hybridizations, respectively, has provided evidence for the reclassification of L. durianis as L. vaccinostercus (Dellaglio et al., 2006).

On the other hand, the neighbour-joining tree based on pheS gene sequences revealed heterogeneity between strains of L. reuteri. As mentioned earlier, the pheS gene reflects a fast-evolving evolutionary clock that shows a finer resolution, in most cases, than the rpoA gene at both the intraspecies and interspecies levels.

L. sakei species group
L. sakei and L. curvatus have >99 % 16S rRNA gene sequence similarity; the corresponding pheS and rpoA gene sequence similarities were 88 % and 96 %. At the subspecies level, the neighbour-joining tree based on the pheS gene sequences showed that L. sakei subsp. sakei and L. sakei subsp. carnosus were clearly differentiated from each other (92 % pheS gene sequence similarity) (see Supplementary Fig. S1).

L. salivarius species group
The pheS neighbour-joining tree split this species group into two subclusters (see Supplementary Fig. S1). An interesting relationship detected by the simultaneous analysis of pheS and rpoA gene sequences is the high genomic relatedness of the L. cypricasei and L. acidipiscis type strains. L. acidipiscis strains (LMG 19820^T and LMG 23135) and the strains of L. cypricasei (LMG 21592^T, CCUG 42959, CCUG 42960 and CCUG 42962) revealed 99.8–100 % pheS and rpoA gene sequence similarities. Sequence analysis of the atpA gene also showed a high relatedness (>99 %) between the two species (data not shown). High DNA–DNA reassociation values confirmed that L. cypricasei is a later synonym of L. acidipiscis (Naser et al., 2006c).

In addition, whereas the type strains of L. animalis and L. murinus are separated by their 16S rRNA gene sequences, these two species are highly related on the basis of their pheS and rpoA gene sequences (Fig. 2). The type strains of L. animalis and L. murinus occupied a distinct subcluster having 98.5 % pheS and rpoA gene sequence similarities.

Other Lactobacillus species
The type strains of L. fructivorans and L. homohiochii showed a high degree of similarity (100 % pheS and rpoA gene sequence similarities). Further taxonomical studies are needed to clarify their relatedness.

Conclusions
It is now generally accepted that a correct classification should reflect the natural relationships as encoded in the DNA and consequently genotypic methods are considered of paramount importance to modern taxonomy. The use of several housekeeping genes in bacterial taxonomy is best suited for analysis at the species and genus levels as it integrates the information of different molecular clocks around the bacterial chromosome (Gevers et al., 2005; Stackebrandt et al., 2002; Zeigler, 2003).

Our data convincingly prove that the simultaneous analysis of pheS and rpoA partial gene sequences provide an alternative tool for the rapid and reliable identification of different species of the genus Lactobacillus. The analysis of pheS and rpoA gene sequences effectively allows closely related Lactobacillus species to be differentiated at a higher discrimination level than that possible with 16S rRNA gene sequence comparisons.

The fact that within species groups, different genes may yield different tree topologies does not hamper their use to unambiguously assign isolates to a particular species. Several factors account for the different topologies determined for different housekeeping genes, i.e. the level of the information content, the different rates of evolution due to different selection forces on various genes and the length of the partial sequences that are compared (Christensen et al., 2004). The variation in the discriminatory power of the investigated genes, together with the fact that different genes might provide different closest neighbours or tree topologies, has highlighted the necessity for simultaneous analysis of several protein-coding loci for a robust identification analysis.

We intend to contribute to the present identification system by the construction of a central, curated database in which data can be stored and accessed freely online. This is expected to contribute in the long run to the improvement of a better species definition for the genus Lactobacillus. The system is rapid, highly reproducible, portable and provides adequate resolution power. In addition, we further intend to extend this system to include all other genera of LAB.

S. M. N. acknowledges a PhD scholarship from the Ministry of Education and Higher Education, Palestine. J. S. and D. G. acknowledge grants from the Fund for Scientific Research (FWO), Belgium. We thank Leentje Christiaens and Marjan De Wachter for their technical assistance.