The genus Brucella has been divided into species and biovars for a long time, but this classification has been discussed controversially since DNA-DNA hybridization has been applied. The genus proved to be highly monomorphic with a level of relatedness among all species higher than 90% . This homogeneity complicated the development of molecular assays able to efficiently recognise the species-specific entities. This finding led to the proposal of a monospecies genus, i.e. B. melitensis. The classical species would be considered as biovars only. However, most bacteriologists did not accept this concept which has recently been rejected by the subcommittee of taxonomy . The purpose of the present study was firstly to investigate the polymorphism of tandem repeat loci predicted to be polymorphic by comparing the data of the three different Brucella strains already sequenced and secondly to evaluate to which extend tandem repeat typing and classical biotyping clustering fit together. We evaluated most of these loci with a repeat unit of 5 bp or more.
Polymorphism has been confirmed at 71 loci. DNA was amplified at every locus from all 21 reference strains, including the 3 marine mammal strains (except for Bruce04 in the B. melitensis bv 3 reference strain Ether and Bruce01 in the B.ovis reference strain BOW63/290) confirming the very high genetic homogeneity of the genus Brucella.
A MLVA typing assay depends on the selection of markers which individually would not provide a relevant clustering. Taken separately, the TR markers are either not informative enough, or too variable or show a high level of homoplasy. However, the combination of well selected independent loci may be highly discriminatory and to some extend phylogenetically relevant, as shown previously for other species , and demonstrated here for Brucella. We propose a selection of 15 markers to be used in a Brucella MLVA assay consisting of two complementary panels, panel 1 (8 markers) and panel 2 (7 markers). The fifteen markers are a combination of moderately variable (minisatellites, panel 1) and highly discriminant (microsatellites, panel 2) loci (Table 2).
The strain clustering achieved is consistent with well-established phenotypic and molecular characteristics (Figure 3, 4 and 5). The biovars 1, 2 and 4 of B. abortus are gathered in agreement with (i) the sensitivity to thionin and (ii) the PCR-RFLP pattern of the omp2a genes specific for these biovars . B. abortus biovar 3 strains are found in a separate group except for 2 strains originated from Africa (BCCN 93-26 and the reference strain Tulya). Strains isolated in Africa often show distinct phenotypes  and thus, it is not surprising to find these two strains separated. The two strains do not require CO2 for growing. Their MLVA closest neighbours are two B. abortus biovar 6 strains also isolated in Africa. Assignment to biovar 3 or 6 reflects the H2S production which is the unique phenotypical criteria to differentiate these two biovars. The MLVA assay confirms that some African strains significantly differ from isolates of other origin and that B. abortus biovar 3 is a heterogeneous group.
The B. melitensis group is very heterogeneous using either panel 1 or both panels (MLVA-15), and comprises four main subgroups. Biovar 2 and 3 strains are mixed in two groups, together with a few biovar 1 strains. The other biovar 1 isolates form 2 groups, one including the 16M reference strain, and the other (genotypes 173 and 174, Figure 5) comprising 3 isolates from the United Arab Emirates. B. melitensis BCCN 84-3 strain (MLVA-15 genotype 20) is an isolate from a dog in Costa Rica, which was biotyped as B. melitensis biovar 2, but appears to be distantly related to other B. melitensis strains. This strain is smooth as observed by the agglutination with anti-A serum, and the profile obtained in oxidative metabolism is typical of B. melitensis. Panel 1 analysis (not shown) does associate this strain with B. melitensis, but the full MLVA-15 analysis suggests a position closer to the B. canis group (Figure 3).
B. suis strains are clearly differentiated in three groups (Figures 3 and 4). A first group includes all biovar 1, 3, and 4 strains, and a second group all biovar 2 strains. The two rare biovar 5 strains are very distantly related. The correlation with biovars is good with some interesting exceptions. The five B. suis biovar 3 isolates from Croatia have the same genotype (MLVA-15 genotype 36, Figure 3 [see Additional file 1]), and cluster with B. suis biovar 1 strains but not with the reference B. suis biovar 3 strain. More B. suis strains phenotypically identified as biovar 3 from other geographic origins are required. This may suggest that the biovar 3 phenotype may have appeared independently more than once. Biovar 1 and biovar 3 strains are distinguished by sensitivity to fuchsine and ability to produce H2S. Atypical fuchsine-resistant biovar 1 strains have already been described , as well as atypical fuchsine-sensitive B. melitensis strains [29, 30]. So both the fuchsine sensitivity, and the H2S production (as suggested above for B. abortus) may appear to be phylogenetically weak markers with some degree of homoplasy. Among biovar 2, strains isolated from Spain and Portugal are related and can be distinguished from other European strains investigated. Biovar 4 strains can be found right beside B. canis. Meyer  has previously proposed a model for evolutionary derivation of Brucella organisms on the basis of phenotypic characteristics and proposed a close relationship between B. suis biovar 3/4, and B. canis. PCR-RFLP analyses of the porin genes are in agreement with this finding .
Three classical vaccine strains were included, Rev.1 (genotype 201), S19 (genotype 161) and RB51 (genotype 159). Six other isolates, from Israel, share genotype 201. These streptomycin resistant isolates were confirmed as Rev.1 vaccine strains using the previously described assay  (data not shown). This is not unexpected since vaccination is used in this country, and simply illustrates the stability of the MLVA assay in the present case.
Strains clustering together frequently have a close or identical geographic origin, e.g. MLVA-15 genotype 16 comprises 2 B. ovis isolates, coming from the same region of France "Provence-Côte d'Azur" (departments 06 and 13). In almost all such instances where the MLVA genotype of two isolates is identical, the available epidemiological data is indeed compatible with a common source of infection. The rare exceptions would then suggest that some strains travel efficiently. MLVA-15 genotype 132 was observed in Germany in 1972 and in the centre of France (department 87) in 1994. MLVA-15 genotype 1 (B. canis) was observed in Greece and Germany. More epidemiological data will be needed in order to draw precise conclusions on the circulation of the strains.
The MLVA-15 results support the current classification of the genus Brucella. In addition, differences found by phenotypic identification and/or by molecular studies are also detected by MLVA. One major advantage of MLVA is the ease of data exchanges. The data itself can be summarized by a very simple flat text file containing the repeat copy numbers for each locus and each strain. This data can also be made accessible and queried across the internet as shown [21, 24].
Another advantage is that MLVA typing only depends on the measurement of DNA amplicon sizes, so that a number of electrophoretic techniques can be used, ranging from manual, low-cost, agarose gels, to high-throughput capillary electrophoresis sequencing machines.
In the near future, it is tempting to speculate that international databases containing MLVA data of thousands of strains will be produced, and MLVA will become a routine assay for any new isolate. We believe that the MLVA-15 assay will be one step in this direction. A first use of the assay for a clinical application was recently described .