For comparing and interpreting the results of phylogenomic analysis, it was necessary at first to examine the evolutionary relationships among α-proteobacteria in phylogenetic trees. Phylogenetic analyses for α-proteobacteria was carried out based on concatenated sequences for 12 highly conserved proteins (see Methods). The relationships among these species were examined by both traditional phylogenetic methods (viz. neighbour-joining (NJ), maximum parsimony (MP) and maximum-likelihood (ML)) and by the character compatibility approach 27. Figure 1 presents a neighbour-joining distance tree for α-proteobacteria, showing the bootstrap scores for various nodes using the NJ, MP and ML methods. In this tree, all of the main groups or orders within α-proteobacteria (viz. Caulobacterales, Rhizobiales, Rhodobacterales, Rhodospirillales and Sphingomonadales), except the Rickettsiales, formed well-resolved clades by different methods. For the Parvularculales, sequence information was available from a single species and it branched with the Caulobacterales. The two main families of the Rickettsiales i.e. Rickettsiaceae and Anaplasmataceae did not form a monophyletic clade, although they constituted the deepest branching lineages within α-proteobacteria. Except for the NJ method, the relative branching of different main groups within α-proteobacteria was not resolved by other methods. The branching orders of different groups as seen here is similar to that observed previously in the rRNA trees 1151628. Recently, after this work was completed, Williams et al. 29 have reported similar results based on phylogenetic analysis of a different large set of protein sequences.

The evolutionary relationships among α-proteobacteria were also examined using the character compatibility approach 27. This method removes all homoplasic and fast-evolving characters from the dataset 273031 and it has proven useful in obtaining correct topology in cases which have proven difficult to resolve by other means 313233. These analyses were carried out on a smaller set of 27 species containing all main groups of α-proteobacteria and two ε-proteobacteria. The concatenated sequence alignment for the 12 proteins contained 896 positions that were useful for these studies (i.e. those sites where only two amino acids were found with each present in at least two species). The compatibility analysis of these sites resulted in 12 largest cliques each containing 350 mutually compatible characters. The two main relationships observed in these cliques are shown in Fig. 2. The other cliques differed from those shown only in the relative branching positions of various Rhizobiaceae species, which varied from each other by a single character and are shown as unresolved in Fig. 2. In both these cliques, the species from all main orders within α-proteobacteria were clearly distinguished by multiple unique characters. Further, in contrast to the phylogenetic trees where Rickettsiaceae and Anaplasmataceae did not form a distinct clade (Fig. 1), their monophyletic grouping was strongly supported by 9 unique shared characters (Fig. 2). The Rickettsiales and Rhodospirillales formed the deepest branching lineages in these cliques and other groups within α-proteobacteria branched in the following order: (Sphingomonadales (Rhodobacterales (Caulobacterales (Rhizobiales)))). This branching order was supported by multiple unique characters at each node giving confidence in the results.

The two main cliques obtained in these analyses differed from each other in terms of the branching position of the species belonging to the order Rhodospirillales. In one clique (Fig. 2A), the Rhodospirillales branched with the Rickettsiales, whereas in the other this group of species was found to branch after the Rickettsiales and it formed outgroup of the other α-proteobacteria (Fig. 2B). However, only a single character supported the former relationship indicating that it was not reliable. The two families within the Rhodospirillales order (Rhodospirillaceae and Acetobacteraceae), although they branched close to each other, no unique character common to them was identified, indicating that they are highly divergent. The exact branching position of the Xanthobacter was also not resolved in these cliques. In some cliques, it appeared as an outgroup of the Bradyrhizobiaceae (as shown by the dotted lines in Fig. 2), whereas in others it was placed in the middle of the Rhizobiaceae and Bradyrhizobiaceae families (as seen for the Aurantiomonas).

Table 1 lists some characteristics of various α-proteobacterial genomes that have been sequenced. The genomes vary in size from less than 1 Mb for Neorickettsia sennetsu to more than 9.0 Mb for Bradyrhizobium japonicum. To identify proteins that are distinguishing features of various higher taxonomic groups within α-proteobacteria, systematic Blastp searches were performed on each ORF in the genomes of B. japonicum USDA 110, Brucella suis 1330, Caulobacter crescentus CB15, Gluconobacter oxydans 621H, Mesorhizobium loti MAFF303099, Nitrobacter winogradskyi Nb-255, Novosphingobium aromaticivorans DSM 12444, Rhodobacter sphaeroides 2.4.1, Silicibacter sp. TM1040, Rhodospirillum rubrum ATCC 11170 and Wolbachia (Drosophila melanogaster) endosymbiont (see Methods section). The genomes chosen covered all main taxonomic groups within the sequenced α-proteobacteria. These analyses have identified large numbers of proteins that are uniquely found in particular groups of α-proteobacteria. A brief description of these results is given below.

We previously described 6 proteins (viz. CC1365, CC1725, CC1887, CC2102, CC3292 and CC3319) that appeared distinctive characteristics of α-proteobacteria 17. The α-proteobacterial specificity of these proteins in earlier work was only assessed by means of Blastp searches and it was not confirmed by PSI-Blast, as in the present work (see Methods). Further, since the earlier work, the number of sequenced α-proteobacteria and other genomes has more than doubled. Hence, it was important to confirm the α-proteobacteria specificity of these proteins. Our results reveal that four of these proteins viz. CC1365, CC2102, CC3292 and CC3319, are indeed specific for the α-proteobacteria as a whole, whereas for the remaining two proteins homologs showing significant similarities are also found in other bacteria. Of these four proteins, CC3292 is present in all sequenced α-proteobacteria including Candidatus Pelagibacter ubique, which is the smallest known free-living bacterium 2. The protein CC2102 is only missing in P. ubique, while the other two proteins are only missing in 1–2 rickettsiae species (CC1365) and P. ubique (CC3319). These proteins, which are uniquely present in virtually all α-proteobacteria, provide distinguishing markers for this Class of bacteria (Fig. 3).

In earlier work, 9 proteins were identified that were present in nearly all α-proteobacteria, except the Rickettsiales 17. These latter species are all intracellular bacteria that have lost many genes that are not required under these conditions 334. The Blastp and PSI-Blast reexamination of these proteins have confirmed that 7 of these proteins (CC0100, CC0520, CC1211, CC1886, CC2245, CC3010 and CC3470) exhibit the indicated specificity. Except for CC0520 and CC3470, the other five proteins are also found in P. ubique, providing evidence for its placement within α-proteobacteria 2. These results also provide evidence that P. ubique is not specifically related to the Rickettsiales. The phylogenomic distributions of these genes/proteins can be explained by either their evolution after the divergence of the Rickettsiales (Fig. 3), or by gene loss from this lineage.

The Rhizobiales species comprise more than 1/3^rd of the sequenced α-proteobacterial genomes (see Table 1). This order includes a wide assortment of species many of which interact with the eukaryotic organisms to produce diverse effects. This group includes various rhizobia (Rhizobium, Mesorhizobium, Sinorhizobium) and Bradyrhizobia species that induce root nodules in plants and live symbiotically within them to enable nitrogen fixation 4563536. Another Rhizobiaceae species, A. tumefaciens, induces Crown gall disease (tumors) in plants 37. Bartonella and Brucella species are intracellular pathogens responsible for a number of diseases in human and animals including trench fever and brucellosis 7383940. Other members of this order exhibit enormous versatility in terms of their metabolic capabilities and life styles 14142. Earlier studies identified six proteins that appeared distinctive characteristics of the Rhizobiales species 17. Reexamination of these proteins by the more stringent criteria used in the present work confirmed that three of these proteins (BQ00720, BQ07670 and BQ12030) are indeed specific for the Rhizobiales and they are present in virtually all sequenced species from this large order. In addition to the Rhizobiales, these proteins are also present in Stappia aggregata, which is presently grouped with the Rhodobacterales 12, but was originally known as a strain of Agrobacterium 43.

New blast searches on the genomes of B. japonicum, B. suis, M. loti and N. winogradskyi have identified large number of other proteins that are specific for different subgroups within the Rhizobiales order. Seven proteins listed in Table 2A are uniquely present in most of the sequenced species belonging to the Rhizobiaceae (Rhizobium, Sinorhizobium, Agrobacterium), Brucellaceae, Phyllobacteriaceae (Mesorhizobium) and Aurantimonadaceae families. The absence of many of these proteins in Bartonella species is probably due to gene loss 734. These groups form a well-defined clade with high bootstrap scores in the NJ, MP and ML trees (see Fig. 1) and the genes for them likely evolved in a common ancestor of these genera/families (Fig. 3). Another 9 proteins (Table 2B) are uniquely present in most of the above species except Aurantimonas, which forms outgroup of the Rhizobiaceae, Brucellaceae, Bartonellaceae and Phyllobacteriaceae families (Figs. 1 and 2) 29. Thus, the genes for these proteins have evolved after the branching of Aurantimonadaceae (Fig. 3). For six of the proteins in Tables 2A and 2B (marked with *), homologs with low E values are also found in S. aggregata, providing further evidence for its relatedness to the Rhizobiaceae family. Nine other proteins in Table 2C are found in most Rhizobiaceae species and M. loti, but they are missing in Bartonella and Brucella species. Although these proteins suggest a closer relationship between the Rhizobiaceae and Phyllobacteriaceae families, it is more likely that their genes have been lost from the Bartonella and Brucella species 34, which are intracellular bacteria. Additionally, some proteins were only found in either Mesorhizobium and Rhizobium, or Mesorhizobium and Sinorhizobium (Table 2D). These analyses have also identified 43 proteins that are uniquely present in all four sequenced Brucella species and many other proteins that are present in either three or two of the sequenced Brucella species (see Additional file 1).

The analyses of proteins in the genomes of B. japonicum and N. winogradskyi have identified 12 proteins that are uniquely present in either all (or most) of the sequenced Bradyrhizobiaceae species as well as X. autotrophicus (Table 3A). The species from these two families form a strongly supported clade in the phylogenetic tree (Fig. 1)29. Sixty-two additional proteins in Table 3B are uniquely present in various species belonging to the Bradyrhizobiaceae family (i.e. Bradyrhizobium, Nitrobacter, Rhodopseudomonas). Many other proteins (see Additional file 2) are only found in two of the three Bradyrhizobiaceae genera and their distributions can result from gene losses, lateral gene transfers (LGTs), or other mechanisms.

The order Rhodobacterales is a heterogeneous lineage of bacteria that exhibit much diversity in terms of their metabolism and cell division cycles 113. This group includes many photosynthetic bacteria that are capable of CO₂ as well as nitrogen fixation and also many chemoorganotrophs that can metabolize various sulfur-containing compounds 4445. A number of budding, stalk forming and prosthecate bacteria also belong to this group 46. In addition to many completely sequenced genomes (see Table 1), information for several other species belonging to this order (e.g. Sulfitibacter, Oceanicola, Loktanella, Jannaschia, Dinoroseobacter, Roseovarius and Sagittula) is available in the NCBI database. To identify proteins that are specific for the Rhodobacterales, phylogenomic analyses were carried out on various ORFs in the genomes of R. sphaeroides 2.4.1 and Silicibacter sp. TM1040.

These studies have identified 29 proteins that are present in all-available Rhodobacterales species (Table 4A), but these proteins as well as those listed in Table 4B and 4C are not found in H. neptunium, Oceanicaulis alexandrii, Maricaulis maris or Stappia aggregata. These latter species are presently grouped with the Rhodobacterales 12, however, the absence of various Rhodobacterales-specific proteins in them and phylogenetic studies indicate that the placement of these species within this order is incorrect and needs be revised. In phylogenetic trees based on concatenated sequences for many proteins, O. alexandrii and M. maris consistently branched with the Caulobacter rather than the well-defined clade of Rhodobacterales (Figs. 1 and 2) 29. The studies by Badger et al. 47 also provide strong evidence for the grouping of H. neptunium with the Caulobacterales.

Six additional proteins in Table 4B are present in most of the Rhodobacterales species, but they are missing in R. sphaeroides and P. denitrificans, which form a distinct clade that appears as the outgroup of other Rhodobacterales species (Fig. 1)29. Thus, the genes for these proteins have likely evolved after the branching of these two genera. Thirteen additional proteins, which are only found in Silicibacter and Roseobacter genera (Table 4C) support a close relationship among them, as seen in phylogenetic trees (Fig. 1).

Of the proteins that are specific for Rhodobacterales, two of them (YP_613058 and YP_613059 in Table 4A, corresponding to proteins ABK27256 and ABK27255 in R. capsulatus genome) were previously identified as part of a complex referred to as gene transfer agent 48, based on similarity to certain virus-like elements. Another protein in the same category (viz. ABK27253) is specific for the Rhodobacter genus. The significance of these results is presently unclear.

The order Caulobacterales is comprised of a single family with only four genera 1249. These chemoorganotrophic bacteria are commonly found in marine aerobic environments and they are distinguished by their ability to form stalked cells and unusual cell division cycle 4950. The complete genome of only C. crescentus, which is the best-studied organism from this group, is presently available 51. However, as discussed above, a number of other species which are presently classified as Rhodobacterales viz. O. alexandrii, M. maris, H. neptunium and also Parvularcula bermudensis, consistently branch with C. crescentus in different phylogenetic trees (Figs. 1 and 2) 2947. Thus, phylogenomic analysis of the ORFs from C. crescentus genome was of much interest.

These analyses have identified 2 proteins (CC0486 and CC2480), which are uniquely present in this species as well as O. alexandrii, M. maris, H. neptunium and P. bermudensis (Table 5) One additional protein, CC2764 is present in all of these species except H. neptunium. The remaining eight proteins in Table 5 are only found in C. crescentus, O. alexandrii and M. maris, indicating that the latter two species are more closely related to C. crescentus in comparison to either H. neptunium or P. bermudensis. These results are strongly supported by the branching patterns of these species in phylogenetic trees (Figs. 1 and 2) 2947. Previously, Badger et al. 46 have reported identification of 62 proteins that were only found in C. crescentus and H. neptunium. However, the blast threshold used in this study to infer the absence of these proteins in other species was very high i.e. 1e-10. By the criteria used in the present work (see Methods), none of these 62 proteins was found to be unique to these two species.

The species belonging to the order Sphingomonadales are present in both terrestrial and aquatic environments 52. A distinguishing characteristic of many species from this group is the presence of glycosphingolipids in their cell envelope rather than lipopolysaccharides 5253. Several species from this group (e.g. N. aromaticivorans) can degrade a wide variety of aromatic hydrocarbons 52, whereas others such as Zymomonas mobilis, can highly efficiently ferment sugar to ethanol 53, making them of much interest and importance from biotechnological standpoints. This group also includes phototrophic organisms (e.g. Erythrobacter litoralis), which contain bacteriochlorophyll a and can derive significant fraction of their metabolic energy via anaerobic photosynthesis 54. The complete genomes of 5 species from this order are now available (see Table 1). In addition, large numbers of sequences for Sphingomonas sp. SKA58, are also available in the NCBI database. Blast searches on the ORFs in the genome of N. aromaticivorans have identified 16 proteins (Table 6A) that are uniquely present in all 6 of the Sphingomonadales species for which information is available. Thirteen additional proteins (Table 6B) are present in all other Sphingomonadales species, except Z. mobilis, which is the deepest branching species within this group (Fig. 1). The genes for these proteins likely evolved after the branching of Z. mobilis. Many other proteins are present in only 3 or 4 of these species (viz. N. aromaticivorans, E. litoralis, S. alaskensis, S. wittichiii and Sphingomonas sp. SKA58) (see Additional file 3) and their phylogenomic distribution can result from a variety of mechanisms including shared ancestry, gene loss and LGTs among these species.

We have also identified a 4 aa insert in a highly conserved region of Gyrase B that is mainly specific for the species from this order (Fig. 4). This indel is present in all available Sphingomonadales species, but it is not found in most other alpha proteobacteria or other bacteria (results not shown). Besides Sphingomonadales, a similar size indel is also present in three other species (viz. C. leidyia, R. blasticus, and Pesudorhodobacter incheonensis). Because other Rhodobacterales or Caulobacter species lack this insert, the presence of this indel in these three species could be either due to LGTs or possibly due to taxonomic misclassification of these species.

The order Rhodospirillales is comprised of diverse species including some photosynthetic and magnetotactic bacteria, some acidophiles as well as other bacteria commonly associated with flowers, fruits and fermented beverages that are involved in the partial oxidation of carbohydrates and alcohols 55. The order is made up of two main families, Rhodospirillaceae and Acetobacteraceae 1255. The complete genomes of four species, two from each family, Rhodospirillaceae (Magnetospirillum magnticum, R. rubrum) and Acetobacteraceae (Acidiphilium cryptum and G. oxydans), are available (Table 1) 5657. Phylogenomic analyses of various ORFs in the genomes of G. oxydans and R. rubrum have led to identification of one proteins, GOX0963, which is uniquely found in all of these species (Table 7A). Three other proteins in this Table are present in at least 3 of the 4 species from this order. This table also lists 14 proteins each that are distinctive characteristics of either the Acetobacteraceae (Table 7B) or the Rhodospirillaceae (Table 7C) families, providing molecular markers for these families. We have also identified a 25 aa insert in a conserved region of the RNA polymerase beta subunit (RpoB) that is unique to various sequenced Rhodospirillales species, but not found in any other bacteria (Fig. 5).

The Rickettsiales species are intracellular pathogens responsible for a number of diseases in humans and other animals 334. This order is comprised of three families, Rickettsiaceae, Anaplasmataceae and Holosporaceae. All of the sequenced genomes are from the first two families. Phylogenomic analysis of various ORFs from the genome of Wolbachia (D. melanogaster) endosymbiont has identified 3 proteins viz. WD0161, WD0715 and WD0771 (Table 8A) that are specific for the entire Rickettsiales order. Five other proteins in Table 8B (viz. WD0083, WD0157, WD0821, WD0827 and WD0863) are present in all sequenced Ehrlichia, Anaplasma, Wolbachia and Neorickettsia species (belonging to the Anaplasmataceae family), but they are not found in any of the Rickettsiaceae or other bacteria. Ten additional proteins (Table 8C) are uniquely present in the Ehrlichia, Anaplasma and Wolbachia species, but are absent in Neorickettsia. In view of the deep branching of Neorickettsia in comparison to these other genera (Fig. 1), the genes for these proteins have likely evolved after the branching of Neorickettsia. In earlier work, a number of proteins that appeared specific for the Rickettsiaceae family were also identified 17. Additional Blastp and PSI-Blast searches on these proteins confirm that three of these proteins viz. RP030, RP187 are RP192, are indeed specific for the Rickettsiaceae family. Of these, RP030 is also found in Orientia tsutsugamushi.

The Blastp searches were carried out on each ORF in the genomes of B. japonicum USDA 110, B. suis 1330, C. crescentus CB15, G. oxydans 621H, M. loti MAFF303099, N. winogradskyi Nb-255, N. aromaticivorans DSM 12444, R. sphaeroides 2.4.1, Silicibacter sp. TM1040, R. rubrum ATCC 11170 and Wolbachia (D. melanogaster) endosymbiont to identify proteins that are uniquely present in α-proteobacteria species at different phylogenetic depths. The blast searches were performed against all organisms (i.e. non-redundant (nr) database) using the default parameters, without the low complexity filter 62. The proteins that were of interest were those where either all significant hits were from the indicated groups (or orders) of α-proteobacteria, or which involved a large increase in E values from the last hit belonging to a particular group to the first hit from any other group and the E values for the latter hits were > 1e ^-4, indicating weak similarity that could occur by chance. However, higher E values were often considered significant for smaller proteins as the magnitude of the E value depends upon the length of the query sequence 62. All promising proteins were further analyzed using the position-specific iterated (PSI)-Blast program 62. In this study, we have also retained a few proteins where 1 or 2 isolated species from other groups had acceptable E values, as they provide possible cases of LGTs. For all of the proteins that are specific for α-proteobacteria, their protein ID's, accession numbers and any information regarding cellular functions (such as COG number or presence of any conserved domain) were tabulated and are presented. In describing various proteins in the text, "bll, bsl, blr", "BQ", "BR or BRA" "CC," "GOX" "ml", "Nwi", "Saro", "RSP", "TM1040", "Rru" and "WD" indicate the identification numbers of the proteins in the genomes of B. japonicum, Bartonella quintana, B. suis, C. crescentus, G. oxydans, M. loti, N. winogradskyi Nb-255, N. aromaticivorans, R. sphaeroides 2.4.1, Silicibacter sp. TM1040, R. rubrum and Wolbachia (Dros.) endosymbiont, respectively.

The amino acid sequences for 12 conserved proteins (viz. RNA polymerase β and β' subunits, alanyl-tRNA synthetase, phenyalanyl-tRNA synthetase, arginyl-tRNA synthetase, protein synthesis elongation factors EF-Tu and EF-G, RecA, Gyrase A, Gyrase B, Hsp60 and Hsp70) for different species were downloaded from the NCBI database and aligned using the CLUSTAL × program 63. In addition to the sequences for 50 α-proteobacteria species, sequences for two deep-branching species viz. Helicobacter pylori and Campylobacter jejuni 58, were also included for rooting purposes. The sequence alignments for all 12 proteins were concatenated into a single large file and poorly aligned regions were removed using the Gblocks 0.91b program 64. The final sequence alignment that was used for phylogenetic analyses contained 7652 aligned positions. A neighbour-joining tree based on this alignment was constructed based on Kimura's two parameter model distances using the TREECON program 65. Maximum-likelihood and MP trees were computed using the WAG+F model plus a gamma distribution with four categories using the TREE-PUZZLE 66 and Mega 3.1 program 67, respectively. All trees were bootstrapped 100 times 68, unless otherwise indicated.

The character compatibility analysis was performed on a concatenated sequences for the above 12 proteins for 25 α-proteobacteria species representing all its main orders plus two outgroup species (i.e. H. pylori and C. jejuni) 32. Using the program "DUALSITE" 32, all sites in the sequence alignments where only two amino acid states were found, with each state present in at least two species, were selected. All columns where any gap was present in any of the species were omitted. The useful two state sites were converted into a binary file of "0, 1" characters using the DUALSITE program and this file was used for compatibility analysis 32. The compatibility analysis was carried out using the CLIQUE program from the PHYLIP (ver. 3.5c) program package 68 to identify the largest clique(s) of compatible characters. The cliques were drawn and the numbers of characters that distinguished different nodes were indicated.

Multiple sequence alignments for various proteins constructed in this work were visually inspected to search for any indels in a conserved region that was unique to particular subgroups or orders of α-proteobacteria. The group-specificity of any indel was evaluated by carrying out blast searches on a short segment of the sequence (between 80–120 aa) containing the indel and flanking conserved regions against the non-redundant database. The sequence information for various α-proteobacteria was compiled into signature files that are presented. Sequence information for all other groups of bacteria, which lack these inserts, is not shown.