Rlv3841 has a genome of 7,751,309 base pairs, of which 65% is in a circular chromosome and the rest are in six circular plasmids (Table 1 and Figure 1). This is consistent with earlier electrophoretic and genetic data on this strain 28. All three rRNA operons, which are identical, and all 52 tRNA genes are chromosomal. This is in contrast to A. tumefaciens, S. meliloti and Brucella spp., in which some of these genes are on a second large replicon (> 1 Mb) termed a 'megaplasmid' or 'second chromosome' 3678. The chromosome of Rlv3841 (5.06 Mb) is much larger than those of A. tumefaciens (2.84 Mb), S. meliloti (3.65 Mb) and B. melitensis (2.12 Mb), and the total plasmid content (2.69 Mb) is also large.

All six plasmids of Rlv3841 have putative replication systems based on repABC genes, which is the commonest system in (and apparently confined to) α-proteobacteria. RepA and RepB are thought to be a partitioning system that is essential for plasmid stability, whereas RepC is needed for plasmid replication 2930. Rlv3841 has the largest number of mutually compatible repABC plasmids yet found in one strain of any bacterial species. Although clearly homologous, each of the RepA and RepB polypeptides is highly diverged from all of the others, presumably allowing coexistence of all six plasmids (amino acid identities range from 41% to 61% for RepA, and from 30% to 43% for RepB). Most RepC sequences are also diverged (55% to 68% identity) but the pRL9 and pRL12 RepCs are 97.6% identical, suggesting that a recent recombination has taken place and that divergence of RepC is not critical for plasmid compatibility. Plasmid pRL7 has an 'extra' repABC operon (genes pRL70092-4) and a third version lacking repB (pRL70038-9).

The chromosome and all plasmids except one (pRL7) are remarkably similar in their mix of functional classes (Figure 2). Core functions (to the left in Figure 2) are most abundant on the chromosome, but they are also strongly represented on the plasmids. The proportion of novel and uncharacterized genes ('no known homologues' or 'conserved hypothetical') is as high on the chromosome (31.5%) as it is on the plasmids (30%).

Most putatively essential genes (for example, those that encode core transcription machinery, ribosome biosynthesis, chaperones, and cell division) are chromosomal, but there are exceptions. The only copies of minCDE, which - although not absolutely essential for viability - are involved in septum formation and required for proper cell division 31, are on pRL11 (pRL110546-8). In A. tumefaciens, S. meliloti, and B. melitensis, minCDE are on the linear replicon, pSymB and chromosome II, respectively, raising the possibility that minCDE are important for segregation of large replicons other than the main chromosome. Other 'essential' genes on plasmids include major heat-shock chaperone genes cpn10/cpn60 (groES/groEL) on pRL12 (pRL120643/pRL120642), cpn60 on pRL9 (pRL90041), and ribosomal protein S21 on pRL10 (pRL100450). However, these genes have chromosomal paralogs, so the different copies may serve specialist functions 32 or be functionally redundant 33.

pRL7 is very different from the rest of the genome, with more than 80% of its genes being apparently foreign and/or of unknown function (Figure 2). In fact, 53 genes (28%) encode putative transposases or related proteins, and 31 (including some transposases) are pseudogenes. This plasmid appears to have accumulated multiple mobile elements, often overlapping each other. For example, gene pRL70047A (an intron maturase) is interrupted by pRL70047, which encodes a homolog of the putative transposase of the Sinorhizobium fredii repetitive sequence RFRS9 34 and pRL70047D (conserved hypothetical), and the latter is in turn interrupted by an IS element (pRL70047A, B, C).

The overall G+C content in Rlv3841 is 61% (Table 1), which is closer to that of S. meliloti (62%) than to that of A. tumefaciens (58%). Plasmids pRL10, pRL7, and pRL8 have G+C content under 60%, but the other plasmids resemble the chromosome (61%). However, these averages conceal much local variation, and a plot of GC3s (G+C content of synonymous third positions) reveals many chromosomal 'islands', in which most genes have below average GC3s (Figures 3 and 4). Several of the most distinct islands, for instance RL0790-RL0841 (54 kilobases [kb]), RL2105-RL2200 (105 kb), RL3627-3670 (52 kb) and RL3941-RL3956 (12 kb), precisely abut a tRNA gene; this suggests that they may be mobile elements that target tRNA genes, as has been described for the symbiosis island of M. loti 10 and many genomic and pathogenicity islands in other bacteria.

Dinucleotide relative abundance (DRA) is usually thought to be relatively homogeneous within a bacterial species but different between genomes, even of close relatives 35. The closest sequenced relative of Rlv3841 is A. tumefaciens C58, but the DRAs of these two genomes are consistently different, with no overlap (Figure 5). The C58 linear replicon and large parts of Rlv plasmids pRL12, pRL11, pRL10, and pRL9 have very similar compositions to their respective chromosomes, implying that they have been confined to a narrow range of hosts long enough to acquire the distinctive DRA characteristic of their host. In contrast, the DRAs of pAT, pTi, pRL7 and pRL8, as well as parts of pRL10 and pRL11, resemble each other but are very distinct from those of the corresponding chromosomes (Figure 5). Plasmids pRL7 and pRL8 are transferable by conjugation 717, and so they may be part of a pool of mobile replicons that have not equilibrated to the DRA of their current host genomes. Some regions of the chromosomes and larger plasmids also have this distinctive DRA, perhaps reflecting the recent insertion of 'islands' of mobile DNA. Inclusion of two more genomes, namely those of S. meliloti and M. loti, in a similar analysis does not change the overall picture (KHH and JPWY, unpublished data); the core DNA forms genome-specific clusters whereas the accessory DNA of all four genomes has similar DRA.

The 8-base motif GGGCAGGG is much more frequent in α-proteobacterial chromosomes than expected 36. Its orientation is biased to the leading replication strand and it is most frequent near the terminus of replication, although the reasons for this have not yet been elucidated. Its distribution on the Rlv3841 chromosome clearly illustrates this pattern (Figure 6). Of the 357 copies of the motif, 346 are oriented from origin to terminus (taking about 5,000,000 and about 2,592,000 as the presumed origin and terminus, respectively). The motif is more abundant near the terminus (approximately one every 7 kb) than near the origin (every 25 kb). A novel observation is that it also occurs on plasmids, with a similar frequency and strand bias (Figure 6). However, there is one anomaly; the motif pattern on pRL12 predicts an origin at about 400,000 rather than near repABC. This suggests either that replication initiation of pRL12 is not near repABC or that pRL12 has recently been rearranged but can survive and replicate with the 'wrong' motif distribution.

We identified 648 Rlv3841 genes, 97% of them chromosomal, that have orthologs in each of six other fully sequenced α-proteobacterial genomes (identified in Figure 7). Overall, a phylogeny based on all of these 648 proteins (Figure 7) is consistent with the species relationships inferred from 16S ribosomal RNA, in which the closest relative of R. leguminosarum is A. tumefaciens, followed by S. meliloti, and then M. loti. However, many individual proteins actually support different phylogenetic relationships. To study this phylogenetic discordance in more detail we focused on four genomes, namely Rlv3841, A. tumefaciens, S. meliloti, and M. loti, which simplifies the analysis because there are just three possible topologies for an unrooted phylogeny of four organisms. We identified 2056 quartops (quartets of orthologous proteins 37) in these four genomes (the 648 proteins above are, of course, a subset of these). The consensus topology that is implied by Figure 7 was indeed the best supported: 551 quartops supported A. tumefaciens as the closest relative of Rlv3841 (with > 99% posterior probability). However, 222 supported S. meliloti and 125 M. loti as the closest relative of Rlv3841 (Table 2). The remaining quartops have insufficient phylogenetic signal to support any topology with probability above 99%.

Overall, the quartops represent only 27% of the 7,263 protein-encoding genes of Rlv3841. Although 38% of chromosomal genes encode quartop proteins, only 10% of plasmid genes do so. Even among chromosomal quartops, only 66% (488/745) of those with strong phylogenetic signal support the consensus phylogeny. Replacement of the original ortholog by horizontal gene transfer may explain why so many genes, especially on plasmids, support nonconventional phylogenies. Such discordant phylogenies must have arisen from many individual events, not just a few transfers of large regions, because the genes are scattered across the genome (Figure 3).

There is a strong relationship between phylogenetic distribution and the nucleotide composition of genes. Genes in the quartops have GC3s (mean ± standard error) of 77.9 ± 0.2%, irrespective of the phylogeny that they support, but the GC3s of nonquartop genes is only 72.6 ± 0.1%.

For a broader view of gene relationships, we recorded the presence or absence of a close homolog of each Rlv3841 gene in each of the three related genomes (Table 3). There are 2,253 genes that occur in all three, 2,272 that are absent from all, and 2,740 that occur in some but not all of the other genomes (identified in Additional data file 1). The largest category in this last class comprises 546 genes that are shared by all three rhizobia but are missing from A. tumefaciens, which is surprising in light of the core phylogeny. Furthermore, 264 of these genes have close homologs in Bradyrhizobium japonicum, which shares the phenotype of root nodule symbiosis with the other rhizobia but is much more distantly related according to its core genes (Figure 7). This set of 264 genes includes, of course, the known symbiosis-related genes (discussed below under Nitrogen fixation), but we hypothesized that many of the others might have unrecognized roles in symbiosis. However, after excluding the known symbiosis genes, the representation of the Riley functional classes was not significantly different among these genes from that in the genome as a whole, and so there is no obvious evidence that they are enriched in genes that encode a particular kind of function. The Riley classes provide only a broad outline of course, especially for genomes with many genes of unknown function. However, if a significant difference existed it could readily be detected, as illustrated in the case of pRL7 (Figure 3). As more genome sequences become available there will be scope for more comprehensive analyses, which might include other measures such as the distribution of protein domain classes 38. There is no evidence to suggest that any of these genes shared by rhizobia is directly regulated by NodD, because none of them has a nod box regulatory sequence. Apart from the four known nod boxes that regulate nodA, nodF, nodM, and nodO, the only putative nod boxes that we found in the genome were upstream of RL4088 and pRL120452, both of which encode putative transmembrane proteins of unknown function, but neither of which are in the rhizobium-specific gene set.

To illustrate the differences between core and accessory genomes, we examined one group of genes that includes both core and accessory members, namely those that specify RNA polymerase σ factors. Rhizobia have many genes for RNA polymerase σ factors, and Rlv3841 is predicted to have 11 on the chromosome, one on pRL10, and two each on pRL11 and pRL12 (Table 4). In addition to the 'housekeeping' RpoD, there are two RpoH (heat shock), one RpoN (which is involved, among other things, in assimilation of certain N sources), plus other σ factors of the ECF (extracytoplasmic factor) subclass 39, only some of whose targets are known.

The chromosomal rpoD, rpoN, and rpoH genes exemplify the core genome, their products being highly conserved in close relatives (Table 5). Only one other σ factor gene (RL3703, rpoZ), which is of unknown function 40, had this pattern. In contrast, other Rlv3841 σ factor genes only occur in some of its close relatives or in none at all. One such 'Rlv-only' gene is rpoI (pRL120319), which encodes an ECF σ factor for promoters of the adjacent vbs genes, which are involved in siderophore synthesis and which are also missing from the other related genomes 41. Thus both rpoI and its vbs 'targets' are part of the Rlv accessory genome. The GC3s of the σ factor genes generally concurs with their proposed core or accessory status. Thus, rpoD, rpoN, rpoZ, and the two rpoH all have GC3s above 77%. In contrast, pRL120319 has only 64% GC3s and pRL120580 is even lower (59%). One striking exception is pRL110418, whose product (of unknown function) is absent from close relatives of Rlv but resembles σ factors in B. japonicum and in actinomycetes. It has higher GC3s (79%) than is typical for the Rlv accessory genome, although lower than that of the related genes in Bradyrhizobium (84%) or the actinomycetes (84-94%). It is possible that this is a genuinely 'foreign' gene with a composition that still reflects its origin, rather than a long-term component of the accessory genome.