Sequencing of the genome of A. caulinodans strain ORS571 (hereafter designated A. caulinodans) revealed a single circular chromosome of 5,369,772 base pairs 15. Relevant genome features generated with the BLASTatlas tool 16 are presented in Figure 1 and can be viewed in detail as a web-based resource 17. The putative origin of replication was predicted based on the position of a GC skew shift (Figure 1) 18 and coincided with the occurrence of a gene cluster typically associated with origins of replication in circular chromosomes of α-proteobacteria (Figure 2A) 19. The specific distribution and orientation of the FtsK Orienting Polar Sequences (KOPS) motif 5'-GGGNAGGG-3', which is involved in loading the FtsK DNA translocase and directing it to the replication terminus in α-proteobacteria 20, confirmed the predicted location of the origin between AZC_4717 and AZC_0001 (Figure 2B).

Although the overall GC content of the A. caulinodans genome is 67% and the average GC incidence at the third position of the codon (GC3) is 85%, the chromosome has many islands of varying size with different GC (Figure 3A) and GC3 contents (Figure 3B). In accordance with the overall high GC content, the codon usage is shifted toward GC-rich codons (Figure 4A) and, consequently, GC-coded amino acids are overrepresented (Figure 4B).

Combined computer prediction and similarity searches (Methods) revealed 4717 protein-encoding genes with an average coding density of one gene in every 1123 bp (89%). With the BLASTP program (Methods), the amino acid sequences were compared with the sequences in the nonredundant protein database at NCBI. A putative function could be assigned to 3588 genes (76.1%), 954 genes (20.2%) were similar to hypothetical genes, and the remaining 175 (3.7%) had no significant similarity to any registered gene (Figure 1; Table 1; Additional file 1).

Three rRNA clusters are ordered as 5S-23S-16S (located between the protein-coding genes AZC_0613-AZC_0614, AZC_4195-AZC_4196, and AZC_4435-AZC_4436) and all have an insertion of a tRNA-Ile and a tRNA-Ala between the 16S and 23S genes. A total of 53 tRNA genes representing 44 tRNA species for all 20 amino acids were assigned by sequence similarity and computer prediction with the tRNAscan-SE program 21. Most of the tRNA genes are dispersed on the genome and are probably transcribed as single units. Thirty out of 57 ribosomal protein genes occur in a cluster (AZC_2529-AZC_2559), whereas the others are scattered over the genome (Additional file 1).

For phylogenetic analysis (Methods), the genomes of A. caulinodans and of 44 α-proteobacteria were compared (Additional file 2). The data set was assembled based on the available complete genome sequences (closure date August 15, 2007) and ecological or phylogenetic relatedness. The resulting maximum-likelihood tree (Figure 5A) showed a great concordance with α-proteobacterial trees based on complete 16S rRNA genes 22 or sets of protein families 23. Our analysis placed A. caulinodans closest to X. autotrophicus, Nitrobacter winogradskyi, Rhodopseudomonas palustris, and Bradyrhizobium japonicum, consistent with previous taxonomic studies 91013. With A. caulinodans as a reference genome, a graphical representation of the BLAST hits of the proteins encoded by the genomes of the 13 closest relatives was generated with the BLASTatlas tool (Figure 1) 1617.

For a broader view of the gene relationships, the occurrence and organization of the proteins encoded by these 45 genomes were evaluated (Methods). Each gene of a total data set of 146,315 was classified in one of four groups: orphans, genes without homologs in other bacteria of the data set; singletons, genes with one representative in the genome and homologs in other genomes; phage or integrase-related genes; and duplicated genes or paralogs with more than one paralog in the genome. The distribution of each of these categories differed in the surveyed genomes (Figure 5B). Paralog representation ranged from 5% for the Neorickettsia sennetsu strain Miyayama (genome size 0.86 Mb) to 44% for Rhizobium leguminosarum bv. vicae (strain 3841) (genome size 7.79 Mb), whereas A. caulinodans had 36% paralogs (genome size 5.37 Mb). The data confirmed the observation that the number of paralogs strongly correlates with the genome size in a linear regression 24.

Altogether, these analyses demonstrate that currently X. autotrophicus is the closest sequenced relative of A. caulinodans. However, a comparison of the genomes with the ARTEMIS comparison tool 25 revealed a very low degree of synteny (Additional file 3). Although short sequence stretches are conserved, extensive rearrangements have taken place. The occurrence of four prophages and numerous transposases in the A. caulinodans genome suggests a high genome plasticity. In A. caulinodans, 1412 proteins have no counterpart in X. autotrophicus of which 544 (38%) are catalogued as unknown or hypothetical (Additional file 4). In the remaining group of functionally classified proteins, 46% have GC and GC3 contents different from the genome averages, suggesting recent acquisition.

The putative protein-encoding genes were ordered into 17 classes 26 (Table 1) and the metabolic potential of A. caulinodans was analyzed with the PathoLogic tool of the BioCyC/MetaCYC suite 27.

These analyses revealed the presence of many regulatory genes (8%) and several RNA polymerase σ factors, among which two household σ⁷⁰ factors (AZC_3643 and AZC_4253), two σ⁵⁴ factors (AZC_2924 and AZC_3925; see below), and five σ factors of the extracytoplasmic subclass (AZC_0389, AZC_1202, AZC_2427, AZC_2453, and AZC_3238), implying responsiveness to many environmental triggers. As A. caulinodans is a motile bacterium, a large gene cluster is present (AZC_0615-AZC_0666) for the formation of a type-III flagellum. A significant number of chemotaxis genes predicts the capacity to respond to a wide array of molecules (Additional file 5). While no complete quorum sensing system could be detected, the presence of no less than five LuxR-type response regulators suggests that A. caulinodans has the potential to listen in on acyl-homoserine lactone-mediated communication in its surroundings.

A variety of encoded proteins might offer protection against toxic compounds in the environment (Additional file 6). Examples are two cytochrome P450 monooxygenases and pathways to degrade or modify plant-derived molecules, such as protocatechuate, and xenobiotics, such as cyanate, 1,4-dichlorobenzene, octane, and gallate. Several multidrug efflux pumps, antibiotic-modifying enzymes, and heavy metal translocation systems probably confer resistance to deleterious compounds. The production of the siderophores enterobactin and aerobactin might guarantee iron acquisition from the surroundings.

The surface of bacteria is important for recognition, attachment, and colonization during the interaction with a host. Exopolysaccharides and lipopolysaccharides are involved in nodulation as protective compounds against defense molecules generated by the plant and as communication signals 282930. Other functions could relate to surface structures, important for interaction with the host (Additional file 7), e.g. putative adhesion proteins, antigens, and 29 genes that code for proteins with GGDEF/EAL domains. The latter typically play a role in the transition from a motile planktonic form to a sessile biofilm by controlling the formation and degradation of the secondary messenger cyclic di-GMP 31. Hormones also play an important role in plant-microbe interactions. Both a structural (AZC_0267) and a regulatory gene (AZC_0266) mediating degradation of the ethylene precursor 1-aminocyclopropane-1-carboxylate, are present in the genome.

Over 15% of the genes are dedicated to "transport and binding", of which more than 50% belong to the ATP-binding-cassette (ABC) transporter class. With 118 complete systems (consisting of a solute-binding protein, a permease, and an ABC component for the uptake systems, or an ABC component and a permease for the export systems), and numerous orphan subunits scattered over the genome, the transporter complement of A. caulinodans equals that of many other soil bacteria. These high-affinity transport systems are dedicated to the uptake of peptides, amino acids, sugars, polyamines, siderophores, nitrate/sulfonate/bicarbonate, or C4-dicarboxylate and many unknown substrates (Additional file 8). Accordingly, catabolic pathways are predicted for compounds, such as amino acids (including citrulline and ornithine), glucuronate, galactonate, galactarate, gluconate, quinate, L-idonate, creatinine, and 4-hydroxymandelate. Sugars, such as glucose, fructose, sucrose, ribose, xylose, xylulose, and lactose are not metabolized by A. caulinodans; instead, dicarboxylic acids are used as primary carbon source 10, as reflected by the presence of multiple C4-dicarboxylic acid transport systems. The occurrence of 16 putative alcohol dehydrogenase genes suggests that ethanol could be a major carbon source under flooded conditions. A. caulinodans is also capable of oxidizing hydrogen, an obligatory by-product of the nitrogenase, and the required hup, hyp, and hoxA genes are located in a large gene cluster (AZC_0594-AZC_0613) 32. Encoded energy metabolism pathways include glycolysis, Entner-Doudoroff, and TCA cycle. The absence of a gene encoding phosphofructokinase indicates the lack of a functional Emden-Meyerhof pathway.

Table 2 lists the genomic position of A. caulinodans genes related to free-living and symbiotic nitrogen fixation. These genes code for known functions, such as formation of the nitrogenase, assembly and stabilization of the complex, synthesis of the MoFe cofactor and the FeS clusters, electron transport, ammonium assimilation, and regulation of gene expression by nitrogen and oxygen, but also for proteins whose exact role await experimental confirmation. Several nif genes occur in more than one copy and are scattered over the genome as solitary loci or clusters of varying size with GC and GC3 contents matching the averages of the genome (Additional file 1). The NifH phylogeny was congruent with the phylogenetic relationships based on 16S rRNA 33 or on core protein families [Figure 5A]. The same holds true for the other genes listed in Table 2 (data not shown).

The transcriptional activator NifA (AZC_1049) acts together with a σ⁵⁴ factor RpoN (AZC_3925) to control the nif/fix gene expression 34. Nitrogen regulation of nifA expression is under control of the NtrBC (AZC_3086-AZC_3087) and NtrYX (AZC_3083-AZC_3084) two-component systems 3536 that respond to the intracellular and extracellular nitrogen status, respectively. The expression of these two loci depends on a hypothetical σ⁵⁴ factor RpoF 34, which presumably corresponds to AZC_2924. Oxygen control of nifA expression is mediated by FixLJ (AZC_4654 and AZC_4655) 37, and the transcription factor FixK (AZC_4653) 38. The nifA gene is further controlled at the transcriptional level by a LysR-type regulator 39 and at the translational level by the nrfA gene product (AZC_3080) 40. FixK also activates transcription of the cytNOQP operon (AZC_4523-AZC_4526), encoding the high-affinity terminal oxidase cytochrome cbb3 that is induced under microaerobiosis 4142. Mutants in cytNOQP still fix nitrogen under free-living conditions, suggesting the occurrence of another terminal oxidase 4143. The survey of the genome excluded the presence of a second cytochrome cbb3 complex, but revealed two cytochrome bd complexes (AZC_1353-AZC_1354 and AZC_3759-AZC_3760).

A region of 87.6 kb, delimited by a Gly-tRNA (position 4346061) and an integrase (AZC_3882) and flanked by direct repeats (Figure 6), is characterized by an overall lower GC (Figure 3C) and GC3 contents (Figure 3D) than the genome averages, and a different preferential codon usage (Figure 4A). No less than 18 putative transposases and three integrases are present, suggesting a complex history of horizontal gene transfer events. The region contains the three nod loci that are involved in the synthesis and secretion of the lipochitooligosaccharide Nod factors (NFs) 44, but also genes related to chemotaxis, amino acid uptake, and a putative type-IV secretion system (Additional file 1).

The three nod loci are not adjacent and have a GC content lower than that of the surrounding sequences (Figure 3C). The shifts in GC content correspond to the location of repeated elements that are flanked by insertion sequences or tRNAs (Figures 1 and 6). The constitutively expressed nodD gene (AZC_3792) 4546 codes for a LysR-type regulator that activates transcription of the two other flavonoid-inducible nod loci. The inducible operon nodABCSUIJZnoeCHOP (AZC_3818-AZC_3807) 474849 encodes most of the enzymatic machinery for NF backbone synthesis, decoration, and secretion. The biochemical role of these proteins has been extensively described, except for the last four open reading frames noeCHOP that are involved in NF arabinosylation and are still under study. Based on similarity with proteins involved in arabinosylation of the cell wall in Mycobacterium tuberculosis, noeC (AZC_3810), noeH (AZC_3809), and noeO (AZC_3808) might encode the synthesis of a D-arabinose precursor 505152. The third locus encodes the inducible nolK gene responsible for GDP-fucose synthesis for NF decoration (AZC_3850) 5354.

The symbiosis region also contains two conjugation-related gene clusters with GC and GC3 contents comparable to the genome averages. The cluster AZC_3844-AZC_3826 – flanked by two transposases – consists of repA and genes encoding conjugal transfer, partition, and plasmid stabilization proteins (Additional file 1). In the cluster AZC_3856–3877, flanked by a transposase and an integrase, genes are found that are homologous to the trbBCDEJLFGI genes, a type-IV secretion system involved in conjugative transfer of the tumor-inducing plasmid in Agrobacterium tumefaciens 55 (Figure 6).

The genome annotation indicates the presence of a few additional nodulation-related genes outside of the symbiosis region (Additional file 1). Two response regulators (AZC_1361 and AZC_2281) homologous to nodW genes of Bradyrhizobium japonicum and part of a two-component signal transduction system might be involved in the response to host-exuded flavonoids 56. A nodT-related gene (AZC_3288) 57 might act as the outer-membrane component in NF secretion together with the inner-membrane NodIJ proteins. None of these four potential nodulation genes has a different GC or GC3 content, in contrast to the nod genes of the symbiosis region.

The nucleotide sequence of the entire genome of A. caulinodans ORS571 was determined by the whole-genome shotgun strategy method. For shotgun cloning, DNA fragments of 2 to 3 kb were cloned into the HincII site of pUC118. For gap closing, the pCC1Fos vector (Epicentre, Madison, WI, USA) was used, and approximately 35-kb clones were prepared. The accumulated sequence files were assembled with the Phrap program 69. A total of 71,424 random sequence files corresponding to approximately 7.7 genome equivalents were assembled to generate draft sequences. Finishing was carried out by visual editing of the sequences, followed by gap closing, and additional sequencing to obtain sequence data with a Phred score of 20 or higher 7071. The integrity of the final genome sequence was assessed by comparing the insert length of each fosmid clone with the computed distance between the end sequences of the clones. The end sequence data facilitated gap closure as well as accurate reconstruction of the entire genome. The final gaps in the sequences were filled by the primer walking method. A lower threshold of acceptability for the generation of consensus sequences was set at a Phred score of 20 for each base. The nucleotide sequence is available in the DDBJ/EMBL/GenBank databases under the accession number AP009384.

Coding regions were assigned by a combination of computer prediction and similarity search. Briefly, the protein-coding regions were predicted with the Glimmer 2.02 program 72 and all regions equal to or longer than 90 bp were translated into amino acid sequences that were subjected to similarity searches against the nonredundant protein database at NCBI with the BLASTP program 73. In parallel, the entire genomic sequence was compared with those in the nonredundant protein database with the BLASTX program 73 to identify genes that had escaped prediction and/or were smaller than 90 bp, especially in the predicted intergenic regions. For predicted genes without sequence similarity to known genes, only those equal to or longer than 150 bp were considered as candidates. Functions were assigned to the predicted genes based on sequence similarity of their deduced products to that of genes of known function. For genes that encode proteins of 100 amino acid residues or more, an E-value of 10^-20 was considered significant, whereas a higher E-value was significant for genes encoding smaller proteins (E-value treshold of 10^-10). Genes for structural RNAs were assigned by similarity search against the in-house structural RNA database that had been generated based on the GenBank data. tRNA-encoding regions were predicted by the tRNAscan-SE 1.21 program 21 in combination with the similarity search.

MetaCyc analysis 2774 detected 229 metabolic pathways, containing 1037 reaction steps. To assess the presence or absence of a metabolic pathway and to decrease the likelihood of being misled by the many enzymes that are shared among multiple pathways, the analysis was emphasized on the presence of enzymes that are unique to a pathway.

The Maximum-likelihood tree was based on 108 core proteins of 45 α-proteobacteria 23 whose sequence data and annotation files were available and downloaded from the NCBI Microbial Genome Resource database 75. The set of core genes was determined by an all-against-all BLAST at protein level. Best reciprocal hits were selected, taking into account a cut-off value defined as 20% similarity and an overlap of at least 150 amino acids. Only proteins present in all 45 genomes as single copy were considered as "core proteins" and used to construct the phylogenetic tree. The total alignment contained 32,327 amino acids. The tree was constructed with the Phyml program 76 and a WAG substitution model 77 and 100 bootstrap replicates were run. Unless indicated otherwise, bootstraps are 100 (Figure 5A).