Four strains, two isolates from soils in the Aral Sea region and two from Northern Italy soil, were compared by whole genome hybridisation with type strain Rm 1021. Four slides with three copies of each ORF were used for each comparison and the results were analysed as described in Methods. Genes were considered to be variable if a statistically significant difference (P < 0.001) in hybridisation intensity was detected between the type and the strain under comparison. The fraction of variable genes detected with comparative genomic hybridisation (CGH) on the microarray containing oligonucleotide probes for all currently predicted protein-coding genes of strain Rm1021, ranged from 4.6 to 6.5% (Table 2). In particular, strain BL225C showed the highest number of variable genes (401), while strain AK58 displayed the lowest number (287). The majority of variable genes (77–94%) showed decreased hybridisation intensity (Log2-ratio > 0) of the natural isolate versus the Rm 1021 strain, suggesting deletion or nucleotide divergence in the region covered by the oligonucleotide. The remaining fraction of variable genes, (6–23%, with a Log2-ratio < 0), showed an increased hybridisation signal of the natural strains compared to Rm 1021, suggestive of gene duplication (Table 2).

In order to corroborate the results of the microarray hybridisation analysis, we randomly selected 116 ORFs, with 66 of these being included in the variable ones (P < 0.001) and 52 found to show no significant difference compared to the type strain (table 3).

The 66 genes showing differential hybridisation (Table 3) were PCR amplified from genomic DNA of both strain Rm1021 and the natural isolate showing the difference. The 19 ORFs with a Log2-ratio < 0 selected (suggesting gene duplication) showed positive amplification. For 7 of these 19 ORFs, Southern hybridisations were carried out on restricted DNA of both tested and reference strains. All 7 ORFs showed more than one band in the DNA of wild strain compared to the single band of strain Rm 1021, confirming that the higher intensity of the microarray hybridisation of the wild strain was indeed due to a duplication of the ORF. Of the 47 ORFs with a Log2-ratio > 0 (indicating gene deletion or divergence), 39 gave no amplification in the wild strain, confirming the microarray result that suggested that the ORF was deleted in this strain. Eight ORFs, on the contrary, were amplified both in the wild strain and in strain Rm1021. These ORFs were sequenced and showed the occurrence of nucleotide variations in the DNA of the wild strain within the region covered by the 70-mer oligonucleotide microarray probe. In that latter case, the lower level of microarray hybridisation of the wild strains was attributed to the mismatches between the genome sequence of the natural isolate and the 70-mer probe sequence. These data confirmed the assumption made for the interpretation of the results.

We also amplified 52 ORFs randomly selected from those with a low level of probability to be variable (Table 3). All of them showed amplification from DNA of strain Rm 1021 and from DNA of the wild strain. Four of these were further analysed by Southern hybridisation showing no sign of copy number variation. Seven from the 31 ORFs with a Log2-ratio < 0 but with P > 0.001 were sequenced and no nucleotide polymorphism was observed.

The genes that were found to be variable in the comparison between the type strain Rm 1021 and the four natural isolates were not randomly distributed among the three replicons. A highly significant enrichment (probability of observing this proportion <0.0001) for pSymA was found in all the strains, both for duplicated and deleted/mutated genes (Figure 1). pSymB was also found to be enriched for duplicated ORFs, though not as significantly as pSymA, except in strain AK58 where pSymB was significantly enriched for duplicated ORFs (probability of observing this proportion <0.0001).

Within the replicons, the variable ORFs had a significant tendency to be spatially clustered (runs-test). In particular, in the pSymA plasmid, one region appeared to be duplicated in all natural strains. This region of at least 1000 bp includes two genes located upstream of the nodD2 gene. These genes encode the transcription factors SMa0748 and SMa0750 of the putative MucR/LysR-families. This duplication was confirmed by Southern-blot analysis on DNA extracted from strain AK83 (not shown). Among the putative deleted/mutated genes, several clusters were also identified (Figure 2)

Figure 3 reports the proportion of functional groups among the variable ORFs using the biological classification as defined by the S. meliloti consortium. The most frequently affected functional categories in all strains were, within the Elements of External Origin, Transposases (V.A) and, within Miscellaneous, Unknown Function ORFs (VI.D). These categories were found to be statistically significantly enriched with variable genes (see Methods).

S. meliloti AK58 and AK83 (Table 1) are a part of alfalfa nodulating rhizobia collected by RIAM (St. Petersburg, Russia) and were trapped from soil samples collected in the Northern Aral Sea Region during May 2001 by M. falcata. Isolates BO21CC and BL225C, from Lodi, Italy, were trapped on M. sativa 34. Rhizobia were cultured at 30°C in liquid TY medium (Tryptone 5 g/l, Yeast extract 3 g/l, CaCl₂ 0.4 g/l). DNA was extracted with the FastDNA Kit (Bio 101, Inc.) according to the manufacturer's instructions. Extracted DNA was quantified by spectrophotometric reading (Biophotometer, Eppendorf).

PCR amplification reactions were performed with a Primus 96 Thermal Cycler (MWG-AG Biotech) in a 50 μl total volume with 30 ng of extracted DNA as template and contained 5 μl of 10× reaction buffer (Polytaq, Polymed, Italy), 1.5 mM MgCl₂, 0.2 mM of each dNTP, 1 U of Taq DNA polymerase (Polytaq, Polymed, Italy), 10 pmols of each primer. The cycling conditions were as follows: after incubation at 95°C for 2 min, samples were cycled for 35 cycles through the following temperature profile: denaturation at 94°C for 30 sec, annealing at 57°C for 30 sec, extension at 72°C for 2 min. Finally, the mixtures were incubated at 72°C for 5 min. Then, 5 μl of each amplification mixture were analysed by agarose gel (1.2% w/v) electrophoresis in TAE buffer containing 1 μg/ml (w/v) of ethidium bromide. Southern blot analysis was performed with 1 microgram of total DNA, digested overnight at 37°C with the restriction enzymes XhoI, EcoRI or PvuII, and electrophoresed for 3 h on a 0.7% agarose gel in TAE buffer with a DIG-labelled DNA marker II (Roche). DNA was blotted on a nylon membrane (Amersham). The cDNA probe preparation, the hybridisation and detection conditions were as described previously in Biondi et al. 40. Automated DNA sequencing was performed directly from the primers used for the amplification on the purified PCR products using the BigDye Terminator v.1.1 chemistry and an ABI310 sequencer (PE-Applied Biosystems) according to the manufacturer's recommendations.

Microarray slides were printed by the Center for Biotechnology, University of Bielefeld 42. Microarrays contained 6208 70 mer oligonucleotides directed against protein-coding ORFs of S. meliloti 1021, four 70 mer oligonucleotides directed against transgenes (gusA, lacZ, nptII, aacC1), two 70 mer stringency control oligonucleotides (80% identity), 12 alien 70 mer oligonucleotides and three alien DNA fragments (Stratagene) that can be used as spiking controls. Each microarray slide contained 6.229 triplicate spots in 48 grids of 20 rows and 21 columns. The 48 grids were arrayed in a 4 × 12 pattern of 4 metacolumns and 12 metarows. Alien oligonucleotides and 12 "housekeeping" genes were arrayed in 13 additional replicates. Oligonucleotides directed against the S. meliloti 1021 genome and the alien oligonucleotide controls were taken from the Sinorhizobium meliloti Array Ready Oligo Set Version 1.0 (Qiagen).

Genomic DNA was labelled with FluoroLink Cy3- or Cy5-dCTP (Amersham Biosciences, Milano, Italy) by using the method described by Pollack et al. 46 and the components of the BioPrime DNA labeling system (Invitrogen, Milano, Italy). Two micrograms of each restriction enzyme (TaqI and MspI) digested genomic DNA was labelled by using 20 μl of the 2.5X Random Primer, 40 U of the Klenow fragment, and 3 μl of the Cy5-dCTP or Cy3-dCTP (1 mM stocks) at 37°C for 2 h. Unincorporated fluorescent nucleotides were removed by using Microcon 30 filter columns (Millipore, Milano, Italy). The appropriate Cy5 and Cy3 labelled probes were combined and mixed with 30 μl Cot-1 DNA (1 mg/ml), 20 μl Yeast t-RNA (5 mg/ml), 450 μl TE to concentrate the samples until about 40 μl using Microcon 30 filter columns (Millipore, Milano, Italy). To each combined sample 8.5 μl of 20 × SSC and 0.74 μl of 10% SDS were added. The sample was denatured to 100°C for 1.5 min, and then incubated for 37°C for 30 min. The hybridisation probe was added to the microarray under a coverslip, and hybridisation was performed at 65°C for 16 h. Slides were washed at 60°C with 2 × SSC for 5 min and then at 60°C with 0.2 × SSC containing 0.1% SDS for 5 min and finally at room temperature with 0.2 × SSC for 2 min. The last step was conducted twice. The slides were immediately dried and scanned for fluorescence intensity by using a GenePix 4000B microarray scanner (Axon Instruments, Union City, CA), and the results were recorded in 16-bit multi-image TIFF files. Competitive hybridisation was done twice for one strain. In the first experiment, the Rm1021 reference DNA and the sample DNA from natural strain were labelled with Cy3 and Cy5, respectively. In the second hybridisation, the dyes for labelling were swapped.

For each sample a total of four slides were hybridised (after dye swapping of the two different restriction enzyme DNA preparations); considering that one slide carries three replicas of each ORF, any sample was hybridized twelve times at each ORF.

Raw data from Genepix was imported into R (1.9) 47 and analysed using the LIMMA library (Linear Models for Microarray Data version 1.7, 48). Spots showing hybridisation intensity two standard deviations above background intensity and that were not flagged as bad were used in normalisation and model fitting. For unknown reasons, strain BO21CC showed a lower number of analysable ORFs (see Table 2) as the quality of slides was apparently comparable. A within-array loess normalisation of intensities was applied. A gene was considered to have a statistically significant differences in hybridisation (moderated t-statistics using empirical Bayes shrinkage of the standard errors) when 2 of the 3 spots on the array representing that gene had a p-value lower than p < 0.001. This stringent cut-off allows preventing false positive. This analysis was designed such that positive log2 fold change occurred when hybridisation was higher in the Rm1021 strain. Such a result is indicative of sequence divergence/gene loss in other strain compared to Rm1021. Negative Log2 fold change occurred when more hybridisation was detected in the other strain competitively hybridised with strain Rm1021. Such a fold change pattern is indicative of gene duplication in the other strain tested compared to Rm1021.

We estimated if deleted and duplicated genes in each strain were found significantly more frequently on a given replicon. We calculated the proportion of genes associated with chromosome, megaplasmid pSymA and megaplasmid pSymB in the whole genome and then the same proportion in the significantly divergent and duplicated gene lists (p < 0.001). The hypergeometric distribution was used to calculate the probability of observing this proportion of variable genes for each replicon in comparison to their total number of genes. A Bonferroni correction was applied to adjust the cut-off probability at which a replicon is considered significantly enriched for variable genes. We multiplied the p-value by the number of tests performed and considered a replicon to be significantly enriched if this adjusted probability was below 5%.

Genes were binned as 1 or 0 respectively if they were differentially hybridising or not. They were then ordered according to their position along the replicons and the distribution of 1 and 0 was analysed using a runs-test 49. This analysis tests the null hypothesis that successes in a series of binomial trials are randomly distributed. The alternative hypotheses of this test are that successes are spatially clustered or they are more evenly spaced than by chance. Genes identified as being duplicated or diverged were analysed separately.

Genes found to have a significant difference in hybridisation at a level of p < 0.001, hereafter referred to as variable genes, were used in a functional enrichment analysis. Each gene has been attributed a biological classification by the "S. meliloti strain Rm 1021 genome project" consortium 45. We calculated the proportion of genes associated with each biological process in the whole genome and then in the variable gene list for each strain. The hypergeometric distribution was used to calculate the probability of observing this proportion of variable genes for each biological process in a particular strain compared to the representation in the whole genome. A Bonferroni correction for multiple testing was applied to adjust the cut-off probability at which a gene list is considered significantly enriched for a given biological classification. We multiplied the probability of observing the proportion of variable gene in a category by the number of tests performed (dependant on number of functional categories represented) and considered a gene list to be significantly enriched if this adjusted probability was below 0.05.