We compared all available CTG gene sets against UniProt using BlastP 31. CTG genes with top database hits to bacterial species were identified as putative horizontally transferred genes and the resultant Blast files were inspected manually. A D. hansenii gene (protein ydhR precursor) with a top database hit to a bacterial sequence was not considered for further analsyes as it has previously been described 22. After this process two genes from C. parapsilosis were considered for further analysis; one encodes a putative proline racemase, and the second encodes a member of the phenazine F superfamily. Related family members were identified by a second round of database searching against GenBank to ensure all available genomic data was utilized.

The C. parapsilosis gene (designated CPAG_02038) is most similar to a proline racemase homolog from Burkholderia cenocepacia AU 1054 protein (66% pairwise identity; Figure 1A). Amino acid racemases catalyze the interconversion of L- and D-amino acids by abstraction of the α-amino proton of the enzyme bound substrate 32. CPAG_02038 lies within a large contig and is also present in a previously published genome survey of C. parapsilosis 33, suggesting its presence does not the result from contamination. We could not locate any related genes in any other CTG genome (using BlastP or TBlastN). Family members are widely distributed throughout the prokaryotes however, and are also located within the Pezizomycotina.

We extracted 321 putative proline racemases from 207 organisms, including members of the α, β, γ, and δ-proteobacteria, Actinobacteria, Fungi, Protozoa and Metazoa. Numerous species were found to have several family members [see additional file 2]; all were included for complete comparative purposes. A maximum likelihood (ML) phylogeny was reconstructed from an alignment of all the PR proteins (Figure 2).

There are a large number of polytomies displayed in Figure 2. These probably result from duplication of PR genes followed by diversifying selection, leading to a high degree of sequence heterogeneity. For example, Agrobacterium tumefaciens str. C58 contains three PR homologs [see additional file 2], with an average amino acid pairwise percentage identity of ~31%. Burkholderia cenocepacia AU 1054 contains 2 proline racemase homologs [see additional file 2], which are only 28% identical. To help resolve the evolutionary history amongst PR homologs we reconstructed an additional ML phylogeny based on a reduced dataset (Figure 3). We also reconstructed a Bayesian phylogeny using the heterogeneous CAT site model. The CAT model can account for site-specific features of sequence evolution and has been found to be more robust than other methods against phylogenetic artifacts such as long branch attraction 34. The resultant Bayesian phylogeny is highly congruent with the ML phylogeny (not shown).

The putative C. parapsilosis PR homolog lies in a strongly supported (100% Bootstrap support (BP)) clade with Burkholderia species (Figures 2 &3 clade-A). Burkholderia are β-proteobacteria. However, no other β-proteobacteria, or indeed any other bacterial genus were found within clade-A (Figures 2 &3).

Although no PR homologs were identified in other CTG species, or indeed in any other of the Saccharomycotina, there are homologs in family members of the Pezizomycotina. A Pezizomycotina specific subclade is evident in our phylogeny containing Phaeosphaeria nodorum, Aspergillus niger and Gibberella zeae (Figures 2 &3 clade-B 100% BP). This subclade is found in a strongly supported clade with members of the Actinobacteria (Figure 2 100% BP), containing Brevibacterium linens and an unclassified marine actinobacterium and excluding Rubrobacter xylanophilus (Figure 2 87% BP). This suggests that these Pezizomycotina species obtained their PR gene from the Actinobacteridae subclass rather than the Rubrobacteridae subclass. This transfer event is another independent HGT event of a PR gene into fungi, and we hypothesize it occurred early in the Pezizomycotina lineage, as it is shared by three distantly related species. Its patchy phyletic distribution suggests it has been subsequently lost in other Pezizomycotina species.

There are also PR homologs in the Metazoans. These are found in a eukaryote clade that also contains a number of Pezizomycotina representatives (Figures 2 &3 clade-C 93% BP). Several scenarios can explain this phylogenetic positioning. Firstly, the PR gene may have been present in the last universal common ancestor of all eukaryotes but has been differentially lost in all lineages except those leading to modern day Metazoa and Pezizomycotina. Alternatively, an ancient gene transfer from bacteria to the last common ancestor (LCA) of Metazoa and Fungi could have occurred, with subsequent gene loss amongst different Metazoan and Fungal lineages. A third hypothesis is that two independent gene transfers have occurred into the Metazoan and Pezizomycotina lineages from unsampled bacterial donors. Finally, a transfer from unsampled bacteria into one of the eukaryote clades (either Metazoa or Pezizomycotina) may have occurred with subsequent transfer from one eukaryotic group to the other.

A. niger, A. oryzae and G. zeae all contain multiple PR homologs [see additional file 2]. One A. niger, one G. zeae and the three A. oryzae PR homologs are nested in a strongly supported Pezizomycotina specific subclade (Figures 2 &3 clade-D 100% BP). This subclade if found within a larger predominately proteobacterial clade (Figure 2 74% BP). This infers that there was an independent gene transfer event of a bacterial PR homolog into an ancestral Pezizomycotina species.

The phylogenetic position of the C. parapsilosis PR homolog (Figures 2 &3) resemble that described for the adenosine deaminase (ADA) gene in the Dekkera bruxellensis genome 21. In that analysis, the authors suggest that D. bruxellensis and Burkholderia species received the ADA gene from a species not yet represented in the public sequence databases. Our PR phylogeny suggests a similar event may have occurred within clade-A, which contains only C. parapsilosis and Burkholderia species (Figures 2 &3). Burkholderia species are known to have a genomic repertoire that allows the transfer and receipt of exogenous DNA 35 and a number of studies have reported successful gene transfers into Burkholderia species 3637. It is possible therefore that there have been other successful gene transfers into this bacterial lineage.

The vast majority of amino acids found in living cells correspond to the L-stereoisomer 38. However, D-amino acids are long known to be found in the cell walls of Gram positive and negative bacteria, where they are essential components of peptidoglycan 39. Apart from low levels of D-amino acids derived from spontaneous racemization as a result of aging 40, it was assumed that only L-amino acid enantiomers were present in eukaryotes 41. However, recent studies have reported the presence of numerous D-amino acids in an array of organisms, including mammals 42. The first eukaryotic (proline) amino acid racemase has recently been described from the human pathogen Trypanosoma cruzi 43. A high degree of sequence similarity was observed between the T. cruzi and bacterial homologs 43. Our phylogeny infers that T. cruzi obtained its PR homolog through interdomain HGT from a member of the Firmicutes (subclass Clostridia), as it is grouped beside members of this group with a high degree of support (Figure 2 clade-E 96% BP). We performed database searches 44, against other Protozoan genomes including Trypanosoma brucei, Trypanosoma congolense and Trypanosoma annulata. We failed to locate a homolog in all species except for T. vivax.

Previous analysis has shown that T. cruzi and T. vivax are not each others closest phylogenetic neighbors, relative to the other species sampled 45. This suggests an ancestral Trypanosoma gained the PR gene and multiple losses in different Trypanosoma lineages has subsequently occurred.

The C. parapsilosis PR homolog lies close to an ortholog (CPAG_02041) of orf19.1135 from C. albicans (Figure 4). The gene order to the left of this ORF is conserved in all CTG species, the order to the right is conserved in most CTG species apart from C. parapsilosis and L. elongisporus. C. parapsilosis and L. elongisporus are closely related 29, and an examination of synteny suggests that the PR gene (together with a second ORF, cpar5437) were inserted between CPAG_2041 and CPAG_2037 (Figure 4). cpar5437 encodes a neutral amino acid (AA) transporter. The presence of an AA transporter beside the PR homolog is interesting. If the putative proline racemase has a role in amino acid metabolism, then the presence of the transporter may be the result of an adaptive translocation to enhance the activity of the PR gene. Unlike the PR ORF the AA transporter is fungal in origin. Most CTG species contain a single neutral AA transporter; however C. parapsilosis and D. hansenii have four.

We located tRNA genes for nearly all CTG species beside the large conserved syntenic block (Figure 4). It has been shown that tRNA genes are associated with genomic breakpoints 46. We hypothesize that a genomic rearrangement has occurred at this site in the LCA of C. parapsilosis and L. elongisporus. We cannot determine if the bacterial PR homolog was inserted into the LCA of L. elongisporus/C. parapsilosis and subsequently lost in L. elongisporus, or gained by C. parapsilosis after speciation.

We also investigated the gene order around the Pezizomycotina PR homologs [see additional file 3]. Gene synteny around the PR homologs found in clade-D (Figures 2 &3) is not conserved (not shown). Interestingly however, both A. niger and G. zeae in clade-D (Figures 2 &3) have genes containing a FAD dependent oxidoreductase domain in close proximity to their PR homologs (not shown). According to Pfam 47, FAD dependent oxidases include D-amino acid oxidases, that catalyze the oxidation of neutral and basic D-amino acids into their corresponding keto acids. The presence of these oxidases may be another example of an adaptive translocation to enhance the activity of the PR gene in these Pezizomycotina species.

A. oryzae has three PR homologs (Figures 2 &3 clade-C). All of these have orthologs in its close relative A. flavus (Figure 3 clade-C), and synteny around these is conserved [see additional file 3 clade-D]. The remaining two species in clade-C are A. niger and G. zeae. There is no evidence of conserved gene order within these species, or with A. oryzae or A. flavus. Gene order around the A. flavus and A. terreus PR homologs found in the Metazoan/Pezizomycotina clade (Figures 2 &3) is also conserved [see additional file 3], as is the order between A. fumigatus and N. fishceri [see additional file 3]. We could not locate amino acid transporters or FAD dependent oxidases beside any of the PR homologs found in clades B or C (Figure 2).

It has been shown that recently acquired genes often display an atypical codon preference when compared to other genes in the genome 4849. However, the transferred PR homologs have a codon usage consistent with the rest of their genomes [see additional file 4]. We undertook an analysis of variation in synonymous codon usage on all PR genes shown in Figure 2. Homologs from related species cluster together [see additional file 5]. For example, the Actinobactria, the Firmicutes and the Burkholderia species all inhabit unique areas in two dimensional correspondence analysis space [see additional file 5].

The majority of fungal and Metazoan PRs are clustered together [see additional file 5]. The C. parapsilosis PR homolog has a codon usage distinct from the other Pezizomycotina fungal PR homologs [see additional file 5], which is unsurprising as C. parapsilosis belongs to the Saccharomycotina subphylum. The C. parapsilosis homolog is also separate from the Burkholderia (β-proteobacteria) genes with which it forms a closely related phylogenetic group (Figures 2 &3). This suggests that the gene may have originated from a genome with no other close relatives among the species analyzed here.

The PR active site from Trypanosoma cruzi, Clostridium sticklandii, Agrobacterium tumefaciens, Brucella melitensis and Pseudomonas aeruginosa all contain cysteine at amino acid position 330 4350. This amino acid is essential for enzymatic function, because substitution with serine abolishes activity 41. However, PR homologs from human, mouse, Rhizobium and Brucella contain a threonine instead of a cysteine at position 330 41. We observed that cysteine is found in the equivalent position in many of the bacterial proteins. The Pezizomycotina PR genes found in clade-B and clade-D contain a cysteine at the active site (Figure 3). The PR homologs found in the Metazoan/Pezizomycotina clade (clade-B) have a threonine at position 330. Similarly, the C. parapsilosis PR homolog, together with its relatives from Burkholderia all contain a threonine (Figure 3). However, Burkholderia species have multiple PR homologs [see additional file 2] with a cysteine as the active site (not shown). It is not clear what effect the substitution has on enzyme activity. It has been suggested that homologs containing threonine at the active site are not true PRs 41, but may instead belong to a superfamily. We cannot detect any difference in the ability of C. parapsilosis, the other CTG species or any of the Pezizomycotina species to utilize D-proline as growth media (data not shown). We therefore cannot confidently infer the function of the PR homologs in the fungi analyzed here.

The C. parapsilosis gene (designated CPAG_03462) is most similar to a Photorhabdus luminescens phenazine F (PhzF) protein with 61% pairwise identity (Figure 1B). Phenazines are biologically active compounds, all of which have a characteristic tricyclic ring system and have been shown to confer a selective growth advantage to organisms which secrete them, as they possess broad-spectrum antibiotic activity towards bacteria, fungi and higher eukaryotes 51. In Pseudomonas, the best studied phenazine producer, PhzF is part of an operon required for the conversion of chorismic acid to phenazine-1-carboxylate (PCA) 52. PhzF homologs were identified in most of the CTG species tested as well as several other fungal species. However, we could not identify a PhzF homolog in the L. elongisporus genome, even when multiple TBlastN and BlastN searches were used.

PhzF homologs were extracted from GenBank for subsequent phylogenetic analysis. In total 181 representative protein coding sequences distributed amongst 154 organisms were used. These taxa were distributed amongst α, β, γ and δ-proteobacteria, Actinobacteria, Fungi, Firmicutes a well as other bacterial groups.

We aligned all sequences and reconstructed a PhzF ML phylogeny (Figure 5). The C. parapsilosis PhzF homolog is found in a clade with members of the β-proteobacteria (Burkholderia multiovorans, Burkholderia cepacia, Burkholderia ambifaria), α-proteobacteria (Roseovarius) and the γ-proteobacteria (Azotobacter vinelandii, Acinetobacter baumannii, Shewanella baltica and Photorhabdus luminescens) (81% BP). In contrast, all other PhzF homologs from CTG species are in a completely separate clade (Figure 5). These form a sister group (63% BP) to PhzF homologs from other Saccharomycotina species (C. glabrata, Saccharomyces cerevisiae, Kluyveromyces lactis and Vanderwaltozyma polyspora). All three clades are grouped together in a larger clade with high support (75% BP).

The sister group relationship between the PhzF homologs from the Ascomycota and the proteobacteria clade is intriguing (Figure 5), as it suggests that an ancestral Saccharomycotina species gained the PhzF homolog from a proteobacteria. The bacterial PhzF gene has subsequently been retained after multiple speciation events, but lost in C. parapsilosis. We hypothesize that C. parapsilosis has recently reacquired a bacterial PhzF homolog from a proteobacterial source, as it is grouped (81% BP) within a proteobacterial subclade. To test this hypothesis we reconstructed constrained trees that placed C. parapsilosis together with the remaining Ascomycota species [see additional file 6 C-H]. The AU test of phylogenetic tree selection 53, showed that the original unconstrained tree (groups C. parapsilosis with proteobacteria) receives the optimal likelihood tree score, and the differences in likelihood scores when compared to the constrained trees [see additional file 6], are significant (P < 0.05). This is also supported by spectral analysis [see additional file 7].

Our phylogeny shows that the Schizosaccharomyces pombe PhzF homolog is found in a clade containing all CTG PhzF homologs (Figure 5 99% BP). Furthermore it is grouped beside D. hansenii (66% BP). S. pombe is not a member of the Saccharomycotina, it belongs to the Taphrinomycotina subphylum. The genome sequences of Schizosaccharomyces japonicus and Schizosaccharomyces octosporus have recently been completed 54. We could not locate a PhzF homolog in S. japonicus but did locate a homolog in S. octosporus using a TBlastN search strategy. Phylogenetic analysis has shown that S. pombe and S. octosporus are more closely related to one another than to S. japonicus 55. Therefore we hypothesize that the LCA ancestor of S. pombe and S. octosporus gained the PhzF gene from an ancestral D. hansenii-like species after speciation from S. japonicus. We reconstructed a constrained tree that placed S. pombe outside the Saccharomycotina clade [see additional file 6B]. The approximately unbiased test of phylogenetic tree selection (AU test) 53, showed that the phylogenetic inferences of the unconstrained tree are significantly better (P < 0.05) than the constrained tree [see additional file 6]. This infers that S. pombe has obtained a PhzF homolog from a member of the CTG clade.

A small basidiomycete clade is evident amongst prokaryote species (Figure 5). Both Ustilago maydis and Malassezia globosa belong to the Ustilaginomycotina subphylum. Therefore our phylogeny infers that an ancestral Ustilaginomycotina species gained a PhzF gene from an unknown bacterial source, and both species have retained this after speciation.

A correspondence analysis of synonymous codon usage for all PhzF homologs was also performed and is shown in additional information [see additional file 8]. The S. pombe PhzF homolog has a codon usage pattern very similar to the D. hansenii protein.

Analysis of the genes adjacent to the PhzF homolog in C. parapsilosis shows that there is a high conservation of gene synteny and supports our hypothesis that PhzF was recently acquired in this species (Figure 6). Homologs in the other CTG species are located in completely different regions of the genome relative to C. parapsilosis (not shown). For example, the C. albicans PhzF homolog is located between orf19.5619 and orf19.5621, whereas the C. parapsilosis homolog is found between orf19.6689 & orf19.6687 relative to C. albicans SC5314 (Figure 6). However, the L. elongisporus genome contains no PhzF homolog, either at a position equivalent to the C. parapsilosis copy or elsewhere in the genome.

We propose that the LCA of L. elongisporus and C. parapsilosis lost the PhzF gene present in the other CTG species, and a second (new) copy was subsequently gained by C. parapsilosis after speciation. We have partial sequence data (unpublished) from Candida orthopsilosis, a species so closely related to C. parapsilosis that it was once designated C. parapsilosis group II 56. We located a C. orthopsilosis PR homolog that is 83% identical (at the amino acid level) to the C. parapsilosis copy. This implies that the common ancestor of C. parapsilosis and C. orthopsilosis acquired the bacterial PhzF homolog after speciation from L. elongisporus.

Mechanisms of gene transfer into fungi are poorly understood. To date no DNA uptake mechanism has been identified in CTG species. Interkingdom conjugation between bacteria and yeast has been observed however 575859. Similarly, Saccharomyces cerevisiae has been shown to be transformant competent under certain conditions 60. CTG species are known interact with bacteria in vivo 61, and it is therefore possible that interkingdom conjugation and transformation may facilitate DNA transfer in C. parapsilosis. These mechanisms may also be applicable to the Pezizomycotina species examined in this analysis.

The complete C. albicans (SC5314) genome (Assembly 19) was obtained from the Candida genome database 65. The Broad institute have sequenced and annotated five CTG species (C. albicans (WO-1), C. tropicalis, L. elongisporus, P. guilliermondii, and Cl. lusitaniae). These genomes were obtained directly from the Broad Institute 66. Gene sets for the C. dubliniensis were downloaded from GeneDB 44.

The incomplete C. parapsilosis geneome was downloaded from the Sanger Institute 67. Gene annotations were performed using two separate approaches. The first involved a reciprocal best BLAST 31 search with a cutoff E- value of 10^-7 of Candida albicans SC5314 protein coding genes against the unannotated C. parapsilosis genome. Top BLAST hits longer than 300 nucleotides were retained as putative open reading frames. The second approach involved a pipeline of analysis that combined several different gene prediction programs, including ab initio programs SNAP 68, Genezilla 69, and AUGUSTUS 69, with gene models from Exonerate 70 and Genewise 71 based on alignments of proteins and Expressed Sequence Tags. Putative gene sets from both approaches were imported into Artemis 72 and cross corroborated manually. The resultant gene sets contained 5,823 protein-coding genes. The C. parapsilosis genome was also annotated by the Broad Institute, and where possible we have used the gene names they assigned.

The UniProt database (v11.1) was downloaded 73. Database searches against GenBank refer to release 164.0.

Taking one CTG species at a time, we located gene families of interest by comparing individual protein coding genes against the UniProt database (v11.1) using the BlastP algorithm 31 with a cutoff expectation (E) value of 10^-20. To use all available sequence data, CTG proteins with a top database hit to a bacterial protein in UniProt were extracted for a second round of database searching against GenBank (E value of 10^-20). Proteins which also had a top database hit to a bacterial protein in GenBank were considered as possible incidences of horizontal gene transfer. All putative homologs were extracted from GenBank and searched against the relevant CTG genome to ensure a reciprocal best Blast hit. For completeness, CTG proteins not yet deposited in GenBank were added to gene families of interest where appropriate.

Accession numbers for all sequences used in this analysis can be found in additional material [see additional file 2].

Gene families were aligned using MUSCLE (v3.6) 74 using the default settings. Obvious alignment ambiguities were corrected manually.

Phylogenetic relationships were inferred using maximum likelihood methods. Appropriate protein models of substitution were selected for each gene family using ModelGenerator 75. One hundred bootstrap replicates were then carried out with the appropriate protein model using the software program PHYML 76 and summarized using the majority-rule consensus method.

We performed the approximately unbiased test of phylogenetic tree selection 53, to assess whether differences in topology between constrained and unconstrained gene trees are no greater than expected by chance.

To determine if the putative HGT genes had a different codon usage pattern to the host genome an analysis of variation in synonymous codon usage was undertaken using the GCUA software 77. Individual correspondence analyses of raw codon counts for the Candida parapsilosis, Ustilago maydis, Malassezia globosa, Aspergillus flavus, Aspergillus niger, Gibberella zeae, Aspergillus oryzae, Phaeosphaeria nodorum, and Schizosaccharomyces pombe genomes were performed, with the first four principal axes being used to evaluate synonymous codon usage patterns. Similar analyses were also carried out on members of the proline racemase and phenazine F gene families displayed in Figures 2 and 4. We used spectrum 78 to perform a spectral analysis on a subset of the phenazine data.