INTRODUCTION

Top Abstract Introduction Materials & Methods Results Discussion References

Rhizobia, currently comprising the genera Azorhizobium, Bradyrhizobium, Mesorhizobium, Rhizobium, and Sinorhizobium, are widely distributed and abundant soil bacteria belonging to the alpha subclass of the Proteobacteria. Infective rhizobia are phenotypically well characterized by their ability to induce nitrogen-fixing nodules on leguminous plants (39). Rhizobia can be easily isolated from nodules, which are often occupied by single strains. Phylogenetically, these bacteria constitute a heterogeneous group, as revealed by 16S rDNA sequence analysis. Some of their members are more closely related to nonsymbiotic Proteobacteria, including denitrifying, pathogenic, and phototrophic strains of several genera (e.g., Afipia, Agrobacterium, Bartonella, Blastobacter, Brucella, Rochalimaea, and Rhodopseudomonas), than they are to other rhizobia (46). DNA sequence analysis of 16 rDNA regions has revealed a much greater diversity than previously recognized (23, 27, 48), leading to important revisions in the taxonomy and systematics of this group of bacteria and in the description of new genera and species (for recent reviews see references 22, 23, and 49).

Assessment of rhizobial genotypic diversity relevant to ecologically oriented studies requires a higher level of taxonomic resolution than can be achieved by 16S rDNA sequencing (7). Isolates of the same species, even of the same serotype, can significantly differ in their N₂-fixing efficiencies and in their abilities to occupy nodules in competition with other closely related strains (35, 37). PCR-based genomic fingerprinting methods provide a much finer taxonomic resolution than 16S ribosomal DNA (rDNA) sequencing. Genomic fingerprints generated with short arbitrary primers (randomly amplified polymorphic DNA-PCR) (44) or primers binding to interspersed repetitive sequences (repetitive extragenic palindromic PCR [rep-PCR]) (41) give the highest level of taxonomic resolution currently achievable by PCR methods (5, 20, 40). The high degree of reproducibility of the rep-PCR approach has recently been discussed (38).

Here we present an analysis of the genotypic diversity and phylogeny of a set of nine rhizobial strains isolated from nodules of plants of the Chamaecytisus proliferus taxonomic complex (10) and from other endemic woody Fabaceae of the Canary Islands (3). Tagasaste, the best known taxon of the C. proliferus complex, is the only nonornamental endemic species of the Canarian flora that has substantial agronomic relevance. This fodder-legume is cultivated not only in the Canaries but also in New Zealand and Australia due to its outstanding forage value and rapid growth (8, 36). The combination of high N₂ fixation efficiency (26) and responsiveness to mycorrhizal inoculation (42), renders this plant a very promising candidate for soil conservation programs and low-input agroforestry in regions with a Mediterranean climate.

León-Barrios and collaborators (21, 31) have phenotypically characterized several nodule isolates from tagasaste and other endemic Canarian legumes, including eight of the strains analyzed here. Each of these isolates was classified as a Bradyrhizobium sp. based exclusively on phenotypic traits, mainly the long generation time (above 6 h), alkalization of growth media, NAD-dependent phosphogluconate dehydrogenase activity, flagellation type, and symbiotic host range (21). Lipopolysaccharide electrophoretic patterns and serotyping revealed a high degree of phenotypic diversity within the strain collection (31).

The aim of this work was to determine the phylogenetic positions of the Canarian isolates and selected reference strains by means of amplified rDNA restriction analysis (ARDRA), a technique that is suitable to group strains at the genus or species level of taxonomic resolution (14). The results provided by ARDRA were confirmed by partial sequencing of the 16S rDNA region. To analyze the diversity of these strains at a finer taxonomic resolution intergenic spacer (IGS) PCR-restriction fragment length polymorphisms (PCR-RFLPs) and rep-PCR genomic fingerprints were generated and used to show that rDNA IGS restriction patterns and even very dissimilar rep-PCR patterns can be efficiently clustered into phylogenetic groupings when combined with 16S rDNA RFLPs by computer-assisted pattern analysis.

	MATERIALS AND METHODS

Top Abstract Introduction Materials & Methods Results Discussion References

Isolation and cultivation of bacterial strains. Bacterial strains marked with an asterisk in Table 1 were isolated by one of the authors from mature, fan-shaped, indeterminate nodules present on the roots of plants belonging to different taxa of the C. proliferus taxonomic complex growing in natural or cultivated stands on several islands of the Canarian archipelago. Bacteroids were isolated from surface-sterilized nodules following a standard protocol (34). Isolates inducing nodules on tagasaste were stored as 25% (vol/vol) glycerol stock cultures at 70°C in yeast mannitol broth (YMB) (34). The origin of the isolates and the sources of the reference strains used in this study are listed in Table 1.

Isolation of DNA. Genomic DNA of the bacteria was prepared from liquid cultures harvested at late exponential phase by a standard, cetyltrimethylammonium bromide protocol (2). Purified DNA was dissolved in 10 mM Tris-HCl buffer containing 1 mM EDTA (pH 8.0), and its concentration was adjusted spectrophotometrically to 50 µg/ml.

PCR protocols. All primers used for the PCR experiments were synthesized by MWG-BIOTECH GmbH, Ebersberg, Germany. All PCRs were carried out with Taq polymerase and buffer (U.S. Biochemicals [USB]-Amersham International, Little Chalfont, England) in a Perkin-Elmer 2400 thermal cycler.

ARDRA. Primers fD1 and rD1 were used to amplify nearly full-length 16S rRNA genes (43). PCR was performed in 50-µl reaction mixtures containing 1× PCR buffer, 1.5 mM MgCl₂, 5% dimethylsulfoxide, 200 µM each nucleotide (Boehringer GmbH, Mannheim, Germany), 15 pmol of each primer, 1 U of Taq polymerase, and 50 ng of purified template DNA. The temperature profile was as follows: initial denaturation at 95°C for 3 min 30 s; 35 cycles of denaturation at 94°C for 1 min 10 s, annealing at 56°C for 40 s, and extension at 72°C for 2 min 10 s; and final extension at 72°C for 6 min 10 s. PCR products were purified with Promega's Wizard PCR-Prep columns (Promega, Mannheim, Germany) and digested with the tetrameric restriction endonucleases CfoI, DdeI, and MspI (Boehringer) and MboI (USB-Amersham International), as recommended by the manufacturers. The digests were resolved by electrophoresis with 7-cm-long 2% Metaphor agarose gels (Biozym, Hess. Oldendorf, Germany) in Tris-borate-EDTA (30) at 55 V for 3 h. A 100-bp ladder (GIBCO BRL, Eggenstein, Germany) was run at both sides and in the central lane of each gel.

16S-23S rDNA IGS RFLP analysis. The 16S-23S rDNA IGS region was amplified with primer pair FGPS1490/FGPL132' (20) by using the same PCR mixture described above for ARDRA and the following program: 95°C for 3 min 30 s, followed by 30 cycles at 93.5°C for 1 min, 55°C for 40 s, and 72°C for 1 min and a final extension at 72°C for 5 min.

rep-PCR genomic fingerprinting. The cycling programs and reaction mixture composition (25 µl) were as previously described (41), except that 50 ng of template DNA and 2 U of Taq polymerase were used. Six microliters of the reaction mixtures was loaded onto 18-cm-long 1.5% agarose gels and run at room temperature in Tris-acetate-EDTA (TAE) buffer for 4.2 h at 4 V/cm. A 1-kb ladder (GIBCO BRL) was included as a size reference, as described above.

Computer-assisted analysis of rDNA restriction patterns and rep-PCR genomic fingerprints. Gel images were digitized with a charge-coupled device video camera (INTAS, Göttingen, Germany) and stored to disk as TIFF files. These were converted, normalized with the above-mentioned molecular size markers, and analyzed with GelCompar software (version 4.0; Applied Maths, Kortrijk, Belgium). The "rolling disk" background subtraction method was applied. For ARDRA and IGS RFLP analysis, a band-matching algorithm was selected to calculate pair-wise similarity matrices with the Dice coefficient (14). A band-matching tolerance of 0.75% was chosen. To analyze rep-PCR patterns, as well as combined 16S rDNA restriction patterns with rep fingerprints, similarity matrices of whole densitometric curves of the gel tracks were calculated by using the pair-wise Pearson's product-moment correlation coefficient (r value), an approach that compares the whole densitometric curves of the fingerprints (12, 28). For this type of analysis zones of gels containing primer bands or not fully restricted DNA bands were excluded by defining appropriate "active zones" on the digitized images to be included in the analysis. The rep-PCR genomic fingerprints from 200 bp to 12 kb were compared. 16S rDNA RFLP patterns were compared by using ranges from 70 bp to 1.4 kb for CfoI restrictions, 70 to 750 bp for DdeI restrictions, 70 bp to 1.3 kb for MboI restrictions, and 70 to 800 bp for MspI restrictions. Cluster analysis of similarity matrices was performed by the unweighted pair group method using arithmetic averages (UPGMA) (33).

Cloning of partial 16S rDNA sequences. Partial 16S rDNA sequences, corresponding to positions 44 to 337 in the Escherichia coli numbering system (4), were amplified with primers Y1ES (5'-ccgaattcgtcgacaacTGGCTCAGAACGAACGCTGGCGGC-3'; this study) and Y2HBX (5'-cccgggatccaagcttCCCACTGCTGCCTCCCGTAGGAGT-3'; this study). These primers are derivatives of Y1/Y2 (48) and have linker sequences at their 5' ends containing several restriction sites, permitting directional cloning of the PCR products. The compositions of the PCR mixtures were as described for the amplification of 16S rDNAs. PCR consisted of an initial denaturation at 94°C for 3 min, followed by 30 cycles of 94°C for 45 s, 62°C for 40 s, and 72°C for 2 min and a final extension at 72°C for 3 min 30 s. PCR products were restricted with EcoRI and BamHI. The restriction fragments were purified from low-gelling-point agarose (Biozym) gels with Promega's Wizard PCR-Prep columns (Promega) and ligated into pBluescript KS(+) (Stratagene, La Jolla, Calif.) with T4 ligase (USB-Amersham). The ligation mixture was used to transform CaCl₂-competent E. coli DH5 cells (30).

DNA sequencing of partial 16S rDNA fragments. DNA sequencing of partial 16S rDNA fragments was performed with a Li-COR automatic DNA sequencer, model 4000, by using the M13 universal infrared (IR) detectable primers M13fIR and M13rIR (24) and the Thermosequenase fluorescence-labeled primer cycle sequencing kit with 7-deaza-dGTP (Amersham International, Little Chalfont, England), as recommended by the manufacturers. Partial 16S rDNA sequences of both strands were obtained by direct sequencing of PCR products by a nested-PCR strategy previously described (40). Primers Y1 and Y2 (48) were used for the first PCR, the products being diluted 1/100 for the second nested PCR with primer combinations Y1M13f/Y2 and Y1/Y2M13f to obtain the templates for cycle sequencing, which were gel purified as described above. IR-detectable primer M13fIR was used for the cycle sequencing reaction. Both strands of the cloned fragments (two clones per strain) were sequenced with IR-detectable primers M13fIR and M13rIR. The following partial 16S rDNA sequences were retrieved from sequence databases to prepare Fig. 6: Bradyrhizobium liaoningensis (X86065), Bradyrhizobium sp. strain (Aeschynomene) BTAi1 (M55492), and Bradyrhizobium sp. strain (Lotus) NZP 2257 (M55486). For Bradyrhizobium japonicum USDA 6^T and USDA 110, the corresponding 264-bp fragment was derived from the sequences with accession no. U69638 and Z35330, respectively.

Plant material and nodulation assays. Tagasaste [C. proliferus subsp. proliferus var. palmensis (L.fil.) Link] seeds were collected from cultivated stands on La Palma, the island of its origin (9). Seeds of Acacia pendula, Glycine max cv. McCall and Peking, Lotus corniculatus, Macroptilium atropurpureum, and Vigna unguiculata cv. Red Caloona, kindly provided by S. G. Pueppke, were surface sterilized in a 3% sodium hypochlorite solution for 3 min, washed thoroughly with sterile water, imbibed for 1 h, and plated for germination on 1% (wt/vol) LN agar plates (45). Tagasaste seeds were similarly treated, but to overcome seed coat dormancy, embryos were excised from sterilized, scarified, and imbibed seeds (29) and incubated in darkness at 20°C on 1% (wt/vol) LN agar. Seedlings were transplanted for nodulation assays after 6 days. Leucaena leucocephala cv. Cunningham seeds (Commonwealth Scientific and Industrial Research Organisation, Canberra, Australia) were scarified for 30 min in concentrated H₂SO₄, thoroughly rinsed in sterile tap water, and plated on LN plates for germination. The nodulation assays were performed in plastic growth pouches or in Leonard jars with vermiculite-perlite (1:1 [vol/vol]) as the substrate and N-free LN nutrient solution buffered at pH 6.8 with 10 mM MES (morpholineethanesulfonic acid). The inoculum was derived from YMB cultures of the relevant rhizobial strains. Seedlings were briefly dipped into the inoculum suspension and transferred to Leonard jars (woody legumes) or growth pouches (herbaceous legumes) at a density of two seedlings per cultivation unit. Plants were grown in controlled-environment chambers (a cycle of 15 h of light and 9 h of darkness at 25 and 18°C, respectively, and 70% relative humidity) for 21 days, except for A. pendula, C. proliferus, L. leucocephala, and L. corniculatus plants, which were grown for 42, 35, 35, and 42 days, respectively.

Nucleotide sequence accession numbers. The sequences for strains BC-C1, BC-C2, BC-P5, BC-P6, BC-P7, BES-1, BGA-1, BRE-1, BTA-1, and CIAT 3101 were deposited in the GenBank sequence database under accession no. AF000550 to AF000559, respectively.

	RESULTS

Top Abstract Introduction Materials & Methods Results Discussion References

Analysis of combined CfoI, DdeI, MboI, and MspI 16S rDNA restriction patterns. Nearly full-length 16S rDNAs of nine nodule isolates from the Canaries and of six reference strains (Table 1) were amplified via PCR with universal primers rD1 and fD1 (43). The PCR products were restricted with enzymes CfoI, DdeI, MboI, and MspI. The individual RFLP patterns are shown in Fig. 1. The observed sizes of the restriction fragments for each type of pattern in the normalized gels are listed in Table 2, which also shows the expected sizes of the restriction fragments for strains USDA 6^T and USDA 76^T (the type strains for B. japonicum and Bradyrhizobium elkanii, respectively) and USDA 110spc4 based on the sequences with accession no. U69638, U35000, and Z35330, respectively. The average deviation between observed and expected sizes of the 19 restriction fragments detected for USDA 110spc4 with the four enzymes was 1.8 bp (Table 2).

View larger version (27K): [in this window] [in a new window]   FIG. 1.   Dice/UPGMA cluster analysis of combined CfoI, DdeI, MboI, and MspI restriction patterns of amplified 16S rDNA of nodule isolates of endemic woody legumes of the Canary Islands and reference strains (see Table 1). The individual RFLPs are shown as bands defined on the actual restriction fragments. The sizes of the restriction fragments for each type of pattern are presented in Table 2. Clusters A, B, and C contain all the Canarian isolates and the two reference Bradyrhizobium strains. R. etli, Rhizobium etli; R. tropici, Rhizobium tropici; A. tumefaciens, Agrobacterium tumefaciens; P. fluorescens, Pseudomonas fluorescens.

Analysis of combined rDNA IGS RFLPs. All the Canarian isolates and Bradyrhizobium sp. strain CIAT 3101 yielded single IGS PCR products of ~930 bp, while strain USDA 110spc4 yielded a slightly larger fragment of about 970 bp. The IGS PCR fragments of the other reference strains were substantially larger (data not shown) and were not included in the RFLP analysis.

View larger version (18K): [in this window] [in a new window]   FIG. 2.   Dice/UPGMA cluster analysis of combined DdeI, HaeIII, and MspI restriction patterns of amplified 16S-23S rDNA IGSs of nodule isolates of endemic woody legumes of the Canary Islands and reference Bradyrhizobium strains (see Table 1). The individual RFLPs are shown as bands defined on the actual restriction fragments. Six genotypes were resolved. Not all groupings are consistent with those obtained by ARDRA.

Cluster analysis of combined 16S rDNA and rDNA IGS RFLPs. We reasoned that since the 16S rDNA gene and the IGS occupy contiguous DNA regions in the genome, being parts of the same operon, a cluster analysis of their combined restriction patterns should be feasible. The use of this strategy, as shown in Fig. 3, resolved the same six genotypes as did the IGS RFLP analysis and clustered them in accordance with the results obtained by ARDRA.

View larger version (13K): [in this window] [in a new window]   FIG. 3.   Dice/UPGMA dendrogram based on the combined CfoI, DdeI, MboI, and MspI 16S rDNA RFLPs and 16S-23S rDNA IGS RFLPs of nodule isolates of endemic woody legumes of the Canary Islands and reference Bradyrhizobium strains (see Table 1). The groupings revealed in this cluster analysis are consistent with both the ARDRA and the IGS PCR-RFLP analyses.

Analysis of combined BOX, ERIC, and REP PCR genomic fingerprints. An analysis of the individual and combined BOX, ERIC, and REP PCR patterns revealed a considerable genetic diversity. Almost every Canarian isolate yielded a unique and complex genomic fingerprint with each primer combination, as indicated by the low linkage values in Fig. 4.

View larger version (46K): [in this window] [in a new window]   FIG. 4.   Product-moment/UPGMA cluster analysis of linearly combined BOX, ERIC, and REP PCR genomic fingerprints of nodule isolates of endemic woody legumes of the Canary Islands and reference strains (see Table 1). The only significant cluster is denoted by I. See the text for explanations. On the scale, r values are expressed as percentages. R. etli, Rhizobium etli; R. trop., Rhizobium tropici; A. tum., Agrobacterium tumefaciens; P. fluor., Pseudomonas fluorescens; B. jap., Bradyrhizobium japonicum.

Cluster analysis of combined 16S rDNA RFLPs and rep-PCR genomic fingerprints. As noted above, the great diversity of rep-PCR patterns obtained resulted in low linkage values of the fingerprints, yielding only one significant cluster. Therefore, a cluster analysis was performed on the combination of four 16S rDNA restriction patterns and the three rep-PCR genomic fingerprints (Fig. 5). The resulting dendrogram, here referred to as BERCDMM (for BOX, ERIC, and REP PCR plus ARDRA with restriction enzymes CfoI, DdeI, MboI, and MspI) was found to reflect the overall diversity and phylogenetic positions of the environmental isolates analyzed in this study (see Table 3).

View larger version (25K): [in this window] [in a new window]   FIG. 5.   Product-moment/UPGMA cluster analysis of linearly combined BOX, ERIC, and REP PCR genomic fingerprints and CfoI, DdeI, MboI, and MspI restriction patterns of amplified 16S rDNA of nodule isolates of endemic woody legumes of the Canary Islands and reference strains (see Table 1). Clusters A, B, and C contain all the Canarian isolates and the two reference Bradyrhizobium strains. On the scale, r values are expressed as percentages. For species abbreviations, see the legend for Fig. 4.

Partial 16S rDNA sequence analysis. When subjected to PCR with primer pair Y1/Y2 or its derivatives, all Canarian isolates and reference strains Bradyrhizobium sp. strain (Centrosema) CIAT 3101 and B. japonicum USDA 110spc4 yielded a 264-bp product, after subtracting the terminal primer sequences, corresponding to positions 44 to 337 in the E. coli 16S rRNA sequence (4, 48). Three different sequences (S1 to S3) were found among the Canarian strains (Fig. 6 and Table 3). Cluster S1 contains isolates BC-C2, BC-P5, BES-1, and BTA-1 and corresponds to ARDRA cluster C (Fig. 1 and Table 3). Members of the S1 cluster lacked a 100% homolog in the GenBank sequence database and thus represent a new sequence type for the genus Bradyrhizobium. Strains BC-C1 and CIAT 3101 (ARDRA group B) have the same rDNA sequence (cluster S2), which is identical to that of phototrophic Bradyrhizobium sp. strain (Aeschynomene) BTAi1 (GenBank accession no. M55492) (48).

View larger version (15K): [in this window] [in a new window]   FIG. 6.   Alignment of sequence stretches containing variable positions within the 264-bp 16S rDNA segment amplified by primer pair Y1/Y2 from Canarian and reference bradyrhizobia (see Table 1). GenBank accession numbers of the complete sequences are listed in Materials and Methods.

Analysis of host range. Nodulation experiments were carried out to examine the host range of the strains. The observed lack of nodulation on two soybean varieties, the modern commercial cultivar McCall and the primitive cultivar Peking, indicates that the Canarian isolates have a different host range than typical B. japonicum strains (Table 4), despite their close phylogenetic relatedness as revealed by ARDRA and partial 16S rDNA sequencing (Fig. 1 and 6). Conversely, B. japonicum USDA 110spc4 did not nodulate tagasaste. Interestingly, Bradyrhizobium sp. strain (Centrosema) CIAT 3101, a Colombian isolate, induces fully effective nodules on A. pendula, C. proliferus, and V. unguiculata. None of the Canarian isolates nodulated the last species effectively, but all of them formed N₂-fixing nodules on M. atropurpureum and A. pendula (Table 4). Experiments with L. corniculatus showed that the nodulation patterns of the Canarian isolates are not identical; BC-P6 was found to be the only strain capable of inducing effective nodulation on L. corniculatus.

	DISCUSSION

Top Abstract Introduction Materials & Methods Results Discussion References

Woody legumes from the tropics and subtropics are nodulated by a diverse set of rhizobia, including both fast- and slow-growing strains (6, 13, 50). Dommergues et al. (6) have proposed that N₂-fixing trees and shrubs should be divided into three broad groups, based on their nodulation patterns with fast-, fast- and slow-, and slow-growing rhizobia. With one possible exception (31), all strains isolated so far from endemic shrub or tree legumes from the Canary Islands are slow-growing strains, phenotypically resembling Bradyrhizobium spp. (21, 31, 41a). The available data on the nodulation patterns of these legumes are scarce but indicate that nodulation with bradyrhizobia seems to predominate in the Canaries. Gault et al. (11) have reported that in southeastern Australia tagasaste plants are nodulated effectively both by fast- and slow-growing rhizobia compatible with several Lotus species and propose a symbiotic relationship between the genera Chamaecytisus and Lotus. Indeed, all Canarian strains were able to nodulate L. corniculatus in our experiments, but only BC-P6 induced N₂-fixing nodules. Taken together, these data suggest that tagasaste belongs to group two, as defined by Dommergues et al. (6).

León-Barrios et al. (21) have reported that strains BES-1, BGA-1, BRE-1, and BTA-1 nodulate two taxa in the C. proliferus taxonomic complex as well as Teline canariensis, Teline stenopetala, and M. atropurpureum, but not G. max cv. Amsoy, Lupinus spp., or Lotus campylocladus. The last is an endemic Lotus species from the pine wood belts of the Canaries, frequently found associated with C. proliferus stands. Our nodulation experiments with the advanced commercial soybean cultivar McCall or the primitive cultivar Peking confirm the findings of León-Barrios et al. (21) and suggest that these isolates are not typical B. japonicum strains with regard to host range. Conversely, the well-known soybean strain B. japonicum USDA 110spc4 did not nodulate tagasaste in our experiments. In a later study, Santamaría et al. (31) reported that strains BGA-1 and BTA-1 nodulate Cajanus cajan. The results reported here extend the host range database for the Canarian isolates by providing data on the nodulation of another four legume species (Table 4).

The taxonomy of the genus Bradyrhizobium needs to be further developed, especially with regard to isolates from tree or shrub legumes, since it still relies predominantly on the soybean nodulation phenotype. Currently three Bradyrhizobium species are validly described: B. japonicum (15), B. elkanii (18), and B. liaoningense (47), all nodulating soybeans. The taxonomic positions of most of the tree-associated bradyrhizobia not nodulating soybeans remain uncertain, and isolates of this type are generically referred to as Bradyrhizobium sp. strains (host genus).

For the Canarian strains isolated so far from woody legumes only scarce phenotypic data are available (21, 31), and no cluster analysis based on phenotypic traits has been carried out. Several reports have stressed the lack of correlation between phenotype- and genotype-based Bradyrhizobium classification methods (32, 40). For this reason, we chose genotypic typing methods to survey bradyrhizobial biodiversity.

Our 16S rDNA PCR-RFLP and sequencing analysis grouped the Canarian isolates within the B. japonicum/Rhodopseudomonas palustris rDNA cluster, lending strong support to the classification of each of these strains as a Bradyrhizobium sp. This in turn is consistent with their slow growth, alkalization of culture media and NAD-dependent 6-phosphogluconate dehydrogenase activity (15, 21, 31, 41a).

ARDRA with four tetrameric restriction enzymes revealed the existence of three well-defined groups of bradyrhizobia within our sample set (Table 3), indicating the existence of intrageneric diversity within the small collection of strains analyzed. A recent computer-based study (25) evaluated the efficacy of selected tetrameric restriction enzymes for rDNA RFLP phylogenetic analysis of natural isolates, showing that a minimum of three had to be used to detect more than 99% of the operational taxonomic units within a model data set of over 100 proximally and distally related full-length rDNA sequences. Our choice of enzymes was based on this work and the results of Laguerre et al. (19).

Comparative sequencing of a short stretch of the 16S rRNA gene of each of the Bradyrhizobium strains listed in Table 1 provided an independent confirmation of the ARDRA-based classification. For this purpose we chose primer pair Y1/Y2 (48), since it amplifies a hypervariable region of the 16S rDNA, thus allowing phylogenetic comparisons of close relatives. Furthermore, this region has been analyzed in several other studies for the genotypic characterization of Rhizobium and Bradyrhizobium isolates (13, 32, 40, 48). As expected, all three types of sequences obtained (S1 to S3) were most similar to those of strains in the B. japonicum/R. palustris branch of the alpha subclass of the Proteobacteria (32, 40, 48). Interestingly, S1 was found to be a new sequence representing a line of descent not previously described within the Bradyrhizobium/R. palustris rDNA complex.

To analyze the diversity and phylogenetic relationships of the Canarian isolates at finer taxonomic resolution, 16S-23S rDNA IGS PCR-RFLPs and rep-PCR genomic fingerprints with BOX, ERIC, and REP primers were generated from all environmental and reference Bradyrhizobium strains. IGS RFLP with three enzymes resolved six genotypes. The clustering of these RFLPs was not entirely consistent with groupings revealed by ARDRA. These inconsistencies were overcome by combining ARDRA and IGS RFLP pattern analyses, which yielded a dendrogram consistent with the separately obtained groupings. The resolution obtained by this approach was significantly higher than that achieved by partial sequencing of the 16S rDNA region.

An even finer taxonomic resolution was found in the highly complex BOX, ERIC, and REP PCR genomic fingerprints. Only strains BTA-1 and BES-1 were found to have similar rep-PCR patterns, forming cluster I in Fig. 4 (see also Table 3). It has been shown previously (16) that rep-PCR genomic fingerprints have a greater discrimination power than serotyping, allowing the determination of the phylogenetic relationships of phenotypically nearly identical B. japonicum strains in seroclusters 123, 127, and 129. To maximize strain discrimination by rep-PCR genomic fingerprinting, Judd et al. (16) analyzed combined ERIC-plus-REP fingerprints manually. This is the first study on rhizobial isolates that fully exploits the taxonomic resolution of rep-PCR by combining BOX, ERIC, and REP PCR genomic fingerprints in computer-assisted pattern analysis, maximizing strain discrimination and the phylogenetic coherency of the obtained clusters (28). The great diversity (low correlation) of rep-PCR banding patterns obtained was associated with very low linkage levels. Thus, we explored the value of combining the three rep-PCR genomic fingerprints with the four 16S rDNA restriction patterns. Cluster analysis using the product-moment correlation coefficients (12) of the resulting combined gel yielded a dendrogram (Fig. 5) integrating the phylogenetic information of ARDRA and rep-PCR genomic fingerprinting. This approach allows for the illustration of phylogenetic relationships between isolates with a dynamic range from the genus to the strain level (Fig. 5). Presently, this is difficult to achieve with any other single technique available.

The great genotypic diversity revealed by our genotypic analysis is in good agreement with the great phenotypic diversity reported by Santamaría et al. based on serotyping and electrophoretic profiling of lipopolysaccharides (31). These authors detected 21 profiles within the 27 Bradyrhizobium isolates studied and reported that of the strains included in the present work only BC-P5 (BTA-8) and BC-P6 (BTA-9) have the same lipopolysaccharide profiles. This result is not consistent with our genotypic data (Table 3), which show that these two strains belong to different 16S rDNA homology groups and have very different IGS and rep-PCR fingerprints. Conversely, strains we have found to have identical rDNA genotypes do not display similar lipopolysaccharide profiles. Moreover, Santamaría et al. (31) have reported eight serogroups on the basis of indirect enzyme-linked immunosorbent assay tests; the serogroups were determined by serological cross-reaction to polyclonal antisera raised against strains BGA-1 and BTA-1. Almost no correlation could be found between our genotypic groupings and the serogroups defined by Santamaría et al. (31), further illustrating the lack of correlation between phenotype- and genotype-based methods in grouping Bradyrhizobium strains.

Together, our ARDRA, partial 16S rDNA sequencing, IGS RFLP, and rep-PCR data indicate a large genotypic diversity from intrageneric to strain levels, and provide the first analysis of the phylogenetic position and genotypic diversity of Bradyrhizobium strains nodulating endemic woody legumes from the Canarian archipelago. The type of combinational analysis of fingerprinting data presented here also reveals the power and suitability of computer-assisted pattern analysis for rapid, accurate, and thorough surveys of bacterial diversity.