The data set of erythritol loci utilized in this work was constructed in a two-step process. First BLASTN was used to identify sequenced genomes containing homologs to the core erythritol catabolic genes R. leguminosarum and S. meliloti 18 . The use of BLASTN rather than BLASTP at this stage allowed us to refine the search to bacteria with sequenced genomes. Furthermore, limiting the search to genes with highly similar sequences by using BLASTN allowed us to limit our search to only genes that are likely involved in erythritol catabolism, since all of these genes encode proteins in highly ubiquitous families found throughout bacterial genomes. Initially BLASTN searches were performed using all the core erythritol genes shared between R. leguminosarum and S. meliloti (eryA, eryB, eryC and eryD). However, the search using eryA provided the most diverse data set that also showed a sharp drop in E-value and query coverage. Using either eryA from R. leguminosarum, or eryA from S. meliloti for the BLASTN search resulted in an identical data set. Genomes containing homologs to eryA were selected on the basis of E-values less than 1.00E-5. In cases where multiple strains of the same bacterial species were found to have highly homologous putative erythritol genes (>99% identity) only a single representative of the species was used to avoid redundancy. Additionally B. melitensis 16M and B. suis 1330 were chosen as representatives of the Brucella lineage despite a large number of Brucella species that were identified in our search due to the high degrees of similarity between their erythritol catabolic genes.

Second, the genetic region containing eryA in these organisms was identified and analyzed using the IMG Ortholog Neighborhood Viewer (http://img.jgi.doe.gov) 19 in order to construct the gene maps (loci). The amino acid sequence of EryA from S. meliloti was used as a query for the IMG Ortholog Neighborhood Viewer search.

To analyze the genetic content of organisms in our data set, the amino acid sequence encoded by each gene involved in erythritol catabolism in R. leguminosarum, or in erythritol, adonitol or L-arabitol catabolism in S. meliloti, was individually used in a BLASTP search of the 19 genomes in the data set. The sugar binding proteins of the S. meliloti and R. leguminosarum transporter were used as representatives of the entire ABC transporter. Identity cut-off values that were used to delineate potential homologs to erythritol proteins were unique to each query amino acid sequence. Cut-off values were as follows: MptA: 56%, EryD: 44%, EryA: 46%, RbtA: 50%, EryB: 65%, LalA: 49%, RbtB: 51%, RbtC: 40%, EryC: 68%, TpiB: 69%, EryR: 61%, EryG: 73%. These values were manually determined and generally correlated to a large drop in percentage identity within the BLASTP hits.

Homologs identified that were not within the primary eryA containing loci were used as a query within IMG-Ortholog neighborhood viewer to analyze the region surrounding them. Secondary loci containing homologs to some of these genes were identified in Mesorhizobium sp. and Sinorhizobium fredii. These loci are putative erythritol loci based on homology to known loci involved in erythritol catabolism in Sinorhizobium meliloti 15 16 , Rhizobium leguminosarum 20 and Brucella abortus 21 . Despite not having been experimentally verified we will refer to all loci in our data set as erythritol loci for the purpose of this manuscript.

Amino acid sequences of homologs to proteins previously shown to play a role in erythritol, adonitol or L-arabitol catabolism from each of the organisms in the data set were collected and used for phylogenetic analysis. The 16S rDNA and RpoD sequences were also extracted from the NCBI database for species examined in this study in order to obtain a potential species tree that could be compared with the various phylogenetic gene trees obtained from the individual genes located within the polyol (i.e. erythritol, arabitol, and adonitol) utilization loci. Amino acid sequences were aligned using Clustal-X 22 and PRALINE 23 the resulting alignments were refined manually with the GeneDoc program v2.5.010 24 .

Phylogenies were generated with maximum likelihood analysis (ML) as implemented in the Molecular Evolutionary Genetic Analysis package (MEGA5) 25 and with MrBayes 26 . MEGA5 was used to identify the most suitable substitution models for the aligned data sets. In order to evaluate support for the nodes observed in the ML phylogenetic trees bootstrap analysis 27 was conducted by analysing 1000 pseudo replicates.

The MrBayes program (v3.1) was used for Bayesian analysis 26 28 and the parameters set for amino acid alignments were mixed models and for the 16S rDNA gamma distribution with 4 rate categories. The models used (setting mixed model) for generating the final 50% majority rule trees were estimated by the program itself. The Bayesian inference of phylogenies was initiated from a random starting tree and four chains were run simultaneously for 1 000 000 generations; trees were sampled every 100 generations. The first 25% of trees generated were discarded (“burn-in”) and the remaining trees were used to compute the posterior probability values.

Phylogenetic trees were constructed for RpoD, 16S rDNA and all the key genes associated with the EryA genes. Phylogenetic trees were plotted with the TreeView program 29 using MEGA5 and/or MrBayes tree outfiles. Final trees were annotated using Adobe Illustrator.

Based on homology to eryA from Sinorhizobium meliloti and Rhizobium leguminosarum we have compiled a data set of 19 different putative erythritol loci from 19 different proteobacteria (Table  1). Previous studies suggested that erythritol loci may be restricted to the alpha-proteobacteria 20 . While a majority of the erythritol loci we identified followed this scheme, surprisingly we identified putative erythritol catabolic loci in Verminephrobacter eiseniae (a beta-proteobacterium) and Escherichia fergusonii (a gamma-proteobacterium). Erythritol loci are not widely distributed through the alpha-proteobacteria. A majority of the loci we identified were within the order Rhizobiales. Outside of the Rhizobiales we also identified erythritol loci in Acidiphilium species and Roseobacter species. Within the Rhizobiales, erythritol loci were notably absent from a large number of bacterial species such as Rhizobium etli, Agrobacterium tumefaciens and Bradyrhizobium japonicum that are closely related to other species that we have identified that contain erythritol loci. We also note that erythritol loci appear to be plasmid localized only in S. fredii and R. leguminosarum. In all other cases the loci appear to be found on chromosomes.

The genetic content of each of the organisms ery loci were analyzed by conducting a BLASTP search to the 19 genomes in our data set of the amino acid sequence of each gene associated with erythritol catabolism in R. leguminosarum, or erythritol, adonitol or L-arabitol catabolism in S. meliloti. The results of the BLAST search are presented in Table  2, depicting the presence or absence of homologs to erythritol, adonitol or L-arabitol catabolic genes in each of the genomes that was investigated. Gene maps of erythritol loci were constructed based on the output of our IMG Ortholog Neighborhood Viewer searches and are depicted in Figure  1.

+ indicates presence of homolog in the genome, - indicates absence of homolog in the genome, ++ indicates presence of 2 homologs in genome.

Genes encoding homologs to the core erythritol proteins EryA, EryB and EryD were ubiquitous throughout our data set (Table  2). With respect to the remaining genes, the genetic content of the species can be grouped into three broad categories. (1) Species that contain genes encoding homologs associated with erythritol, adonitol and L-arabitol catabolism. This includes S. meliloti, S. medicae, S. fredii, M. loti, M. opportunism, M. ciceri, R. denitrificans and R. litoralis. These genomes contained homologs to genes that encode enzymes specifically involved erythritol catabolism such as EryC, and TpiB as well as specifically involved in adonitol and L-arabitol catabolism including LalA, and RbtBC. They also contain genes encoding an ABC transporter homologous to the S. meliloti erythritol, adonitol and L-arabitol transporter (MptABCDE) and do not encode homologs to the R. leguminosarum erythritol transporter (EryEFG). One notable exception is M. ciceri which encodes EryEFG homologs rather than MptABCDE (Table  2). (2) Species that contain all the genes associated with erythritol catabolism, but lack the genes associated with adonitol or L-arabitol catabolism. These species include R. leguminosarum bvs. viciae and trifolii, A. radiobacter, O. anthropi, B. suis, B. melitensis, and E. fergusonii. These loci encode EryABCDR-TpiB as well as homologs to the R. leguminosarum ABC transporter EryEFG, but lack genes encoding homologs to enzymes associated specifically with adonitol and L-arabitol catabolism or the S. meliloti transport protein MptABCDE. E. fergusonii contains the most minimal set of homologs to erythritol genes of all the genomes investigated, and did not encode EryR and TpiB. (3) Species that do not encode the specifically erythritol associated EryC, EryR, and TpiB, but encode the adonitol/L-arabitol catabolic complement LalA-RbtABC and homologs to the S. meliloti polyol transporter MptABCDE. These include Bradyrhizobium spp. BTAi1 and ORS278, A. multivorum, A. cryptum and V. eiseniae.

The genetic context of eryA in each of the genomes in our data set supported that each of these organisms contained an erythritol locus. A physical map of the loci in each of these organisms is depicted in Figure  1. Of note, a number of putative erythritol loci were identified in organisms with incomplete genome sequences at the time of analysis, and thus are not discussed here, including: Octadecabacter antarcticus, Pelagibaca bermudensis Enterobacter hormaechei, Fulvimarina pelagi, Aurantimonas sp. SI85-9A1, Roseibium sp. TrichSKD4, Burkholderia thailandensis and Stappia aggregata.

The putative erythritol loci of bacteria in our data set ranged in genetic complexity with the loci from S. meliloti and S. medicae containing 17 different genes, to the simplest being the locus of E. fergusonii, which contained only two divergently transcribed operons that are homologous to the eryEFG and eryABCD loci of R. leguminosarum. A number of species contained loci that were identical in content and arrangement to the R. leguminosarum erythritol locus including members of the Brucella, Ochrobacterum, and Agrobacterium. The only species that contains a locus identical in content and arrangement to S. meliloti is the closely related Sinorhizobium medicae. The locus of Sinorhizobium fredii NGR234, contains all but one of the genes (fucA1) found in the other Sinorhizobium loci (Figure  2).

The loci of Mesorhizobium species were varied, however all three Mesorhizobium sp. contained an independent locus with homologs to lalA and rbtBC elsewhere in the genome (Figure  1). Interestingly, while Mesorhizobium loti and Mesorhizobium opportunism both contain transporters homologous to mptABCDE, Mesorhizobium ciceri bv. biserrulae contains a transporter homologous to eryEFG. This operon also contains the same hypothetical gene that is found at the beginning of the R. leguminosarum eryEFG transcript. The transporters however, are arranged in a manner similar to that seen in S. meliloti and the gene encoding the regulator eryD, is found ahead of the transporter genes, whereas in R. leguminosarum and Brucella, eryD is found following eryC (Figure  1). We also note that whereas M. loti and M. opportunism both contain a putative fructose 1,6 bis phosphate aldolase gene between the eryR-tpiB-rpiB operon and eryC, a homolog to this is also gene is found adjacent to the rpiB in Brucella.

Bradyrhizobium sp. BTAi1, and ORS278, A. cryptum and V. eiseniae all have similar genetic arrangement to that of S. meliloti, except that they do not contain a homolog to eryC, or an associated eryR-tpiB-rpiB operon. These loci also differ primarily in their arrangement of lalA-rbtBC (Figure  1).

The DNA sequences of 16S rDNA (data not shown) as well as the amino acid sequences of RpoD were extracted from GenBank to analyze the phylogenetic relationships of the organisms examined in this study, using the most phylogenetically distant organism Verminephrobacter eiseniae as an out-group. The results of the 16S rDNA and RpoD sequence analyses were in concordance with each other and are consistent with phylogenies that have been previously generated 42 . Initial comparison of the operon structures with the generated phylogenies suggested that the operon structure(s) did not correlate with the species phylogeny. Since the structure of some operons did not correspond well with the species phylogenies we wished to determine if operon structure did correlate with any of the erythritol genes found at the S. meliloti loci. Since homologs to EryA, EryB and EryD were ubiquitous through the data set, it was decided to construct phylogenies based on Maximum Likelihood and Bayesian analysis using the EryA, EryB and EryD data sets. The topology of the phylogenetic tree using EryA is presented in Figure  2. A tree including branch lengths is included as Additional file 1: Figure S1. V. eiseniae was also the most distant member with respect to the EryA phylogeny and again used as an outgroup. The phylogenetic trees of EryB and EryD are not shown but were generally consistent with the EryA phylogeny. The species tree, based on RpoD, was included as a mirror tree with the EryA tree to demonstrate possible horizontal gene transfer events (Figure  2).

The data show that there is a high degree of correlation between the loci configuration and the EryA phylogenetic tree (Figure  1, 2). We note the similarity of the loci of A. radiobacter and R. leguminosarum to Brucella species and O. anthropi but not to the more closely related Sinorhizobium species. This suggests that a horizontal gene transfer may have occurred between these organisms. This is in agreement with what has been previously reported 20 . It also seems likely that a horizontal gene transfer event may have occurred between the Brucella and E. fergusonii. This may explain the unique occurrence of the loci’s presence in a member of the gamma-proteobacteria. Finally, our mirror tree suggests that a horizontal gene transfer of the more complex erythritol locus may have occurred between M. loti and an ancestral species the Sinorhizobium species (Figure  2).

Comparison of the phylogenetic trees of EryA, EryB and EryD to the arrangement and content of the loci led us to more thoroughly investigate the phylogenies of a number of proteins that stood out as unique within the data set. These phylogenies have led us to postulate modes of evolution that may have occurred in these loci.

BLASTP analysis showed a clear distinction between the type of transporter encoded by each of the loci and the remaining genetic content. In general, loci that contained adonitol/L-arabitol type genes contained a transporter homologous to the S. meliloti MptABCDE (Table  2, Figure  1). Loci that contained only erythritol genes contained a transporter homologous to the EryEFG of R. leguminosarum. One exception to this correlation was M. ciceri bv. biserrulae which contained a homologous transporter to EryEFG rather than MptABCDE. This is interesting because M. ciceri groups with the other Mesorhizobia in the EryABD trees. In order to analyze the evolution of these transporters more clearly, phylogenetic trees were constructed of homologs to EryG and homologs to MptA (Figure  3). In general the phylogenies are in agreement with the EryABD phylogenies, with the exception of M. ciceri which falls on a basal branch of the EryG phylogeny. The disparities between the EryG and EryABD phylogenies of M ciceri strongly suggest that parts of its erythritol locus have a different origin. This may have been the result of horizontal gene transfer of a second R. leguminosarum type erythritol locus, followed by recombination between the two.

In two organisms, apparent duplications of genes were present. In M. loti one homolog of lalA was present in the erythritol locus, while a second copy was present elsewhere in the genome adjacent to homologues of rbtB and rbtC, consistent with its location in the other two Mesorhizobium genomes. In S. fredii homologs to the apparent small operon that contains eryR-tpiB-rpiB were found both, as expected, in the erythritol locus, but also elsewhere on the chromosome in the same arrangement. To analyze the evolutionary history of these duplications phylogenetic trees were constructed for the LalA and TpiB homologs (Figure  4 and 5). The two copies of the lalA gene in M. loti are most likely an example of paralogs, as they still group within the same clade among other lalA homologs (Figure  4). The tpiB genes (Figure  5) in S. fredii are possible examples of xenologs 43 as the phylogenetic tree shows that the two versions of the tpiB gene in S. fredii are only distantly related, with one homolog grouping within the expected clade that includes S. medicae and S. meliloti and the second homolog (not part of the main locus) showing monophyly with those found in a clade containing R. leguminosarum sp., B. suis, etc. (Figure  5).