Two fundamental approaches were used to identify genes associated with marine adaptation in the marine actinomycete genus Salinispora. The function-based approach relied on BLAST analyses using key words derived from previously reported marine adaptation genes (MAGs). The comparative genomics approach was annotation independent and detected genes that were present in Salinispora species but absent or rare in other Actinobacteria. Thus, the first approach tested for common mechanisms of marine adaptation among marine bacteria while the later had the potential to detect new or unknown gene functions that may be relevant to marine adaptation. All genes detected using these two approaches were then tested for evidence of a recent common ancestry with bacteria associated with hyperosmotic environments.

The function-based approach yielded the largest number of candidate marine adaptation genes (MAGs), however the vast majority identified using both approaches did not pass the phylogenetic test and therefore did not advance to the final MAG pool (Table 1). Ultimately, 60 and 58 MAGs were identified in the S. tropica and S. arenicola genomes, respectively. Of the MAGs identified based on gene function, 13 are involved in electron transport, 12 encode transporters, and 28-30 (depending upon species) encode channels or pores. Based on comparative genomics, more genes related to marine adaptation appear to have been gained than lost from the two Salinispora spp. (Table 1).

An Actinobacterial species tree was constructed using 19 of 31 AMPHORA marker genes 26 (Additional file 1) derived from 186 Actinobacterial genome sequences (Figure 1). This phylogeny is largely congruent to that previously published 27 with the notable exception of the close relationship of Stackebrandtia nassauensis DSM 4478 (family Glycomycetaceae) to the Micromonosporaceae. This relationship is supported by all of the individual gene trees and has also been reported by others 28. The tree clearly shows that the marine Actinobacteria for which genome sequences are available are polyphyletic and not deeply rooted. It is also notable that the order Actinomycetales is paraphyletic with respect to the Bifidobacteriales and that the previously reported polyphyly of the families Frankineae and Streptosporangineae is maintained in this tree 29.

Genes associated with the sodium-dependent NADH dehydrogenase (Nqr), which have been reported in Gram-negative marine bacteria, were not detected in either Salinispora genome or in any available Gram-positive marine bacterial genomes. Thus, when it comes to respiratory electron transport, there appear to be fundamentally different mechanisms by which Gram-negative and Gram-positive bacteria have adapted to the marine environment. None-the-less, 35 candidate MAGs with annotation linked to NDH-1 were initially detected in both Salinispora genomes (Table 1). These genes comprise three partial and one complete NDH-1 operon (Additional file 2). The 14 genes in the complete NDH-1 operon (nuoA-N) as well as those in the first partial NDH-1 operon were not considered further because their phylogenies are in general agreement with the Actinobacterial species tree, and thus there was no evidence they had been acquired from marine bacteria.

In contrast, phylogenetic analyses of all 13 genes in the second and third partial NDH-1 operons revealed close relationships with marine bacteria and thus these genes remained in the final MAG pool (Table 1). The annotation of the seven genes in the second partial NDH-1 operon predict that they encode the membranous portion of the enzyme complex, which pumps sodium ions or protons to generate an ionic motive force 30. Among these seven genes, Stro769 and Sare711 are annotated as hypothetical proteins but likely encode NuoJ because top BLAST hits are annotated as such. The phylogenies of the corresponding seven Nuo protein sequences are similar and place them in a clade with nine other Actinobacteria (Figure 2A). The next three most closely related clades are comprised of nine Proteobacteria of which six are of marine origin. The Actinobacteria that possess these nuo genes are scattered throughout the species tree (Figure 3A), which could be interpreted as evidence for common ancestry within the Actinobacteria. To more formally infer the likelihood of gene loss (vertical inheritance) vs. gene gain (horizontal acquisition) in accounting for the distribution of these genes, the minimum number of loss events and maximum number of gain events was calculated. The resulting loss to gain ratio of 2.8 indicates that nearly three times as many loss events would be required to explain the observed distribution and thus provides support for the horizontal acquisition of this partial NDH-1 complex in Salinispora spp. (Figure 3A).

The six genes in the third partial NDH-1 complex have annotation related to nuo genes however upon closer analysis these genes appear to encode the sodium proton antiporter Mrp. The ambiguous annotation is not surprising as mrp genes are known to have sequence similarity to nuo genes 11. Both Salinispora strains have mrpA-G, which are required for a functional antiporter 31, however mrpA and B are fused indicating that this is a group two Mrp operon 11. MrpG was incorrectly predicted by auto-annotation but subsequently resolved based on homology with B. halodurans. MrpCEF and G are each too short to produce a robust phylogeny, however, the blast matches for these genes, and the fused mrpAB gene, were similar to the longer MrpD sequence and therefore it is inferred they share the same evolutionary history. The phylogeny of MrpD (Figure 2B) places the two Salinispora spp. in a primary clade that includes five Corynebacteria spp. and the Gram-negative marine bacterium A. marinum. This clade then shares a common ancestor with a large and diverse group of bacteria that contains at least four phyla including many marine and alkaliphilic species. The ratio of gene loss to gain events for each gene in the mrp operon is 2.3 (Figure 3B), thus supporting gene gain as the most parsimonious explanation for the occurrence of this gene in the two Salinispora spp.

S. tropica and S. arenicola contain 18 and 19 candidate sodium transporter genes respectively (Table 1), three of which were confirmed as MAGs following phylogenetic analysis (Additional file 2). One of these MAGs constitutes a Na⁺/bile acid symporter (Stro2582 and Sare2779). The orthologs in the two Salinispora genomes group phylogenetically with 15 Actinobacteria including two other marine Actinobacteria (Figure 4A). The next clade contains Acinetobacter spp. followed by a single Actinobacterium and a large clade of Pseudomonas spp., many of which are human pathogens, and one Myxococcus sp. Subsequent clades include five Gram-negative marine bacteria. The apparent acquisition of this symporter may provide a mechanism to exploit a natural sodium gradient to import bile salts, which can be converted to compatible solutes such as glycine or taurine 32. Interestingly, genes for the biosynthesis of the compatible solute glycine betaine were not found in either Salinispora genome while genes for the uptake of this compound displayed no evidence of acquisition from marine bacteria and thus were not identified as MAGs (data not shown). The second sodium transport gene is a Na⁺/Ca⁺² exchanger (Stro449 and Sare538) with phylogenetic links to three different Actinobacteria and then Nitrococcus mobilis, a member of the Gamma-proteobacteria derived from surface waters of the equatorial pacific (Figure 4B) followed by a group comprised entirely of marine proteobacteria (see treeBASE link provided in the Methods). The third gene is a Na⁺/Ca⁺² antiporter (Stro4216 and Sare4649) that is largely related to genes observed in marine Alpha-proteobacteria (data not shown). The gene loss to gain ratios of 1.7 and 3.8 for the Na⁺/bile acid symporter and Na⁺/Ca⁺² exchanger, respectively, supports the hypothesis that these genes were acquired. A gene loss to gain ratio was not calculated for the Na⁺/Ca⁺² antiporter because it was only observed in distantly related Actinobacteria and thus was assumed acquired. These calcium transporters may be related to the calcium requirement reported for Salinispora spp. 33.

TransportDB was used to identify 225 ABC transporters in each Salinispora genome (Additional file 2). After phylogenetic analysis of each protein, it was shown that the phosphate transporter Pst and branched chain amino acid transporter Liv have phylogenetic links to both marine and human associated bacteria (Figure 4C and 4D) and therefore advanced to the final MAG pool (Table 1). The four genes encoding the Pst transporter (Stro286-Stro289) display the same phylogenetic relationships and are closely related to homologs in marine cyanobacteria (Figure 4C). This transporter may be more efficient at scavenging phosphate from seawater than the form observed in soil Actinobacteria. The gene loss to gain ratio of 3.9 for each pst gene (Figure 3F) provides additional support for the acquisition of these genes in Salinispora spp. The five Liv proteins (Stro1801-1805) maintain the same phylogeny and reveal a close relationship to homologs from the marine Actinobacterium Janibacter sp. and then four bacteria from the Phylum Deinococcus-Thermus (Figure 4D). The next clade contains marine and pathogenic Proteobacteria (data not shown). The gene loss to gain ratio of 3.3 for each gene in this operon supports gene acquisition (Figure 3E).

Of the 35 and 33 channel and pore genes identified as candidate MAGs based on functional annotation in S. tropica and S. arenicola, respectively, 30 and 28 passed the phylogenetic test (Table 1). All of these were previously identified as polymorphic membrane proteins (Pmps) and showed a strong phylogenetic relationship with homologs in marine bacteria 25. These genes are in high copy number (≥ 28) in both Salinispora genomes relative to the closely related genus Micromonospora, in which only two copies are observed. A structural alignment of the predicted Pmp proteins indicates that each forms a beta-barrel structure, which likely forms a pore in the membrane, and contains a signal sequence common to all Pmps supporting that these proteins target the cell membrane.

A representative dataset comprised of 36 Actinobacterial genomes was used to identify 105 genes that are unique to both Salinispora spp. based on the RSD test of orthology (Table 1). Phylogenetic analyses revealed that seven of these genes shared a close relationship with homologs in marine bacteria and therefore advanced to the final MAG pool. However all seven of these genes were included among the MAGs previously identified based on gene function and thus comparative genomics revealed no new MAGS based on gene gain.

To assess gene loss based on comparative genomics, the Micromonospora sp. L5 genome was used as the reference sequence for the pair-wise RSD test of orthology in 27 representative Actinobacterial genomes, including both Salinispora spp. Four of 430 genes with predicted orthologs in at least 24 of the 27 genomes are absent in both Salinispora sequences (Additional file 2). These four genes are 1) a large conductance mechanosensitive channel (mscL) 2) an ABC transporter phosphate-binding protein (pstS), 3) a HAD-superfamily hydrolase, and 4) a peptidoglycan synthetase (ftsI). Homologs of mscL play a role in osmotic adaptation in halotolerant bacteria 34 and provide a mechanism to survive osmotic down shock 1235. Thus, the loss of this gene may play a key role in the inability of Salinispora strains to survive when transferred to low osmotic strength media. The gene loss to gain ratio for mscL is 0.04 and thus highly supports gene loss in both Salinispora spp. (Figure 3G). Based on the RSD analysis, pstS was also identified as being lost in both Salinispora spp. However, all four genes in the pst operon are present in both Salinispora genomes and were already identified as MAGs based on functional annotation and evidence they were acquired from marine cyanobacteria (Figure 4C). Thus, it appears that the pst genes observed in both Salinispora spp. were too divergent to be detected as orthologs based on a comparison with the Micromonospora L5 genome. In support of this, a synteny plot in the region of the pst operon suggests that a homologous recombination event has resulted in the replacement of the entire Salinispora operon with a cyanobacterial version (Figure 5). The HAD-superfamily hydrolase and peptidoglycan synthetase (ftsI) were not considered further as MAGs based on their functional annotation.

The genomes of S. tropica strain CNB-440 (accession # CP000667) and S. arenicola strain CNS-205 (accession # CP000850) were downloaded from the U.S. Department of Energy Joint Genome Institute website (http://genome.ornl.gov/microbial/stro/03jan07 and http://genome.ornl.gov/microbial/sare/18jul07). Strains CNB-440 and CNS-205 were cultured from sediments collected at a depth of 20 m from the Bahamas and Palau, respectively. Artemis was used to visualize gene arrangement and annotation in each genome 39. A Fasta file of predicted protein sequences from the two genomes served as a database for BLAST searches 40. Candidate marine adaptation genes (MAGs) were identified based on 1) gene function (annotation-derived) and 2) comparative genomics. The resulting pool of candidate MAGs was then analyzed using phylogenetic approaches and those with evidence of a shared ancestry with bacteria associated with hyper-osmotic environments were kept in the final MAG pool. Thus, this study is largely focused on the identification of marine adaptation genes that were acquired from other marine bacteria.

Keyword and BLAST searches were performed on the two Salinispora genomes using proteins previously linked to marine adaptation in studies of marine bacteria. The key words searched were associated with electron transport (complex I), sodium transporters, ABC transporters, and pores (Additional file 3). To improve the annotation of transporters prior to the key word searches, the two Salinispora genomes were submitted to transportDB 41, which annotates transporters according to the transport classification system 42. The BLAST searches were performed using complex I genes or mscL (Additional file 3). All sequences identified using these methods were subject to phylogenetic analysis as described below.

Pair-wise comparisons were performed between S. tropica CNB-440 and 37 Actinobacterial genomes (including S. arenicola CNS-205) to identify orthologs that are present in both Salinispora genomes but absent in other Actinobacteria. The genomes selected for comparison include a broad phylogenetic range of Actinobacteria, three Micromonospora spp., and all marine Actinobacteria for which genomes sequences were available as of March 31, 2011. Orthologs were identified using the program Reciprocal Smallest Distance 43 based on e-values < 1e-5, no more than 50% sequence divergence over the entire alignment of the sequence, and the remainder of the parameters set at default. Orthologs were eliminated if they were < 350 amino acids in length or part of a mobile genetic element or secondary metabolite gene cluster as previously defined 25. Orthologs that passed these criteria were then evaluated phylogenetically to determine if they had a shared evolutionary history with bacteria derived from hyper-osmotic environments.

The RSD test was also used to identify genes that were lost in the two Salinispora genomes relative to other Actinobacteria. In this case, the Micromonospora sp. L5 genome served as the reference for the pair-wise prediction of orthologs in 27 representative Actinobacterial genomes, including both Salinispora genomes. Sequences present in > 24 Actinobacterial genomes based on the above RSD criteria for orthology, but not in the two Salinispora genomes, were considered as candidates for gene loss. Functional annotation was then used to determine if gene loss could represent a marine adaptation.

All Salinispora protein sequences identified as candidate MAGs based on functional class and comparative genomics were subject to phylogenetic analysis to test for a shared evolutionary history with bacteria derived from hyper-osmotic environments. If a candidate MAG was part of an operon, the entire operon was tested. Maximum likelihood phylogenies were constructed for each candidate MAG using the online program MABL 44 with default settings (phylogeny.fr/version2_cgi/simple_phylogeny.cgi). The top 100 BLASTP hits were downloaded from the NCBI protein database and those with an e-value < 1e-5 and length greater than 50% of the alignment were included in the tree. Genes that claded with orthologs from hyper-osmotic environments and ≤ 25 Actinobacterial species were kept in the final MAG pool. In cases where the nearest clade was not entirely comprised of strains from hyper-osmotic environments but a majority of strains in all other major clades were, the gene was included in the final MAG pool. Exceptions included trees that contained two or more Micromonospora sequences, as this was viewed as evidence of vertical inheritance. The files used to create the trees shown in Figures 2 and 4 are available at http://purl.org/phylo/treebase/phylows/study/TB2:S12306.

All finished and several draft Actinobacterial genomes were downloaded from the NCBI FTP site on March 31, 2011. For Actinobacterial species with several draft genomes, at least two strains were included. In addition, any unnamed Actinobacterial species that contained a MAG were also included. The program AMPHORA 26 was then used to retrieve, align, and trim phylogenetic markers from each genome. Any marker that was not found in all species was excluded. If more than one version of a marker was found in a genome, the longest version that most closely fit the expected species phylogeny was selected. If the two versions were the same size and fit the expected phylogeny, one was selected randomly. Draft genomes were removed from the dataset if any marker gene was ≤ 25% of the size of all other sequences. However if the draft genome contained a MAG then none of the sequence data was removed. Finally, all aligned genes were concatenated and trimmed with Gblocks. The resulting alignment was input to PhyML for the construction of an Actinobacterial species tree.

The species tree was used to calculate whether horizontal gene transfer or vertical inheritance is the most parsimonious explanation for the observed evolutionary history of each MAG. This was done by first documenting the distribution of each MAG in the species tree. The minimum number of gene loss events was then calculated by identifying the deepest branches in the tree within which all strains lacked the MAG. These branches or points were then summed. The calculation started at the last common ancestor of all strains that possessed the gene. The maximum number of gene gain events was calculated assuming each MAG was acquired independently and summing the terminal branch tips representing each lineage in which the gene was observed. The ratio of the minimum number of gene loss events to the maximum number of gene gain events was then calculated and values above one considered to support the hypothesis that the gene was acquired (ie, a higher number of gene loss events would be required to account for the observed distribution and thus gene gain is the more likely explanation) while values below one were used to support gene loss.