To search for genes of putative chlamydial origin in eukaryotic algae, the genomes of two diatoms (Thalassiosira pseudonana, Phaeodactylum tricornutum), three green algae (Ostreococcus tauri, Ostreococcus lucimarinus, Chlamydomonas reinhardtii), and the red alga Cyanidioschyzon merolae were screened for the presence of protein-coding genes most similar to genes in the genome of Candidatus Protochlamydia amoebophila (see Methods). The initial screen identified 89 putative chlamydial proteins in these algae (Additional File 1). Maximum likelihood phylogenetic analyses were performed with each of the 89 proteins using a taxon sampling that included other eukaryotic algae and chlamydiae, embryophytes, cyanobacteria, and other bacteria (an average of 30 taxa for each protein); 37 proteins revealed a monophyletic lineage (with 70–100% bootstrap support (BS) for 33 proteins, and 57–68% BS for 4 proteins) consisting of chlamydiae and at least one of the three plastid-containing eukaryotic lineages represented by genomes (diatoms, green algae/embryophytes, and red algae; Table 1) to the exclusion of all other proteins. For two additional proteins (putative glycerol-3-phosphate acyltransferase (Fig. 1A), hypothetical protein pc0324 (no 32, Table 1)), the phylogenetic trees contained only chlamydiae and plastid-containing eukaryotes because proteins with significant similarities to the chlamydial proteins could not be retrieved from any other taxa. Among the remaining 50 phylogenetic trees, several trees also displayed a topological sister group relation between chlamydiae and plastid-containing eukaryotes, however, without bootstrap support (not shown).

The proteins of putative chlamydial origin in plastid-containing eukaryotes perform a broad spectrum of functions with isoprenoid, fatty acid and carbohydrate metabolism, and phosphate homeostasis featuring prominently (Additional File 1). All are encoded on the host's nuclear genome, and most (but not all, e.g. CMP-KDO synthetase) are predicted to be targeted to the plastid. Metabolic pathways involving putative chlamydial proteins reveal a chimerical origin (some genes of the plastidic isoprenoid and fatty acid biosynthesis pathways are of cyanobacterial origin).

The distribution of the 39 chlamydial proteins among four groups of photosynthetic eukaryotes (Viridiplantae (green algae and embryophytes), Rhodoplantae (red algae), Glaucoplantae (glaucophytes), and Bacillariophyta (diatoms)) is shown in a Venn diagram (Fig. 2). The largest number of chlamydial proteins (34) was found in the Viridiplantae, followed by the Rhodoplantae (24), Bacillariophyta (22) and Glaucoplantae (5). It should be noted that the low number of chlamydial proteins recovered from the Glaucoplantae presumably relates to that fact that no genome of this small, yet phylogenetically important, algal lineage has been sequenced to date and the chlamydial proteins had to be retrieved from the limited EST (cDNA) data available for two Glaucoplantae, Cyanophora paradoxa and Glaucocystis nostochinearum. The Viridiplantae contain the largest number of unique chlamydial proteins (9), followed by the Bacillariophyta (2), whereas no unique chlamydial proteins were recovered from the Rhodoplantae and Glaucoplantae. Viridiplantae and Rhodoplantae share 21, Viridiplantae and Bacillariophyta 17, and Rhodoplantae and Bacillariophyta 16 chlamydial proteins. Both Viridiplantae and Rhodoplantae share the same 5 chlamydial proteins with the Glaucoplantae (four of which are also present in the Bacillariophyta). Seven chlamydial proteins are unique to Viridiplantae and Rhodoplantae, 4 to Viridiplantae and Bacillariophyta, and 3 to Rhodoplantae and Bacillariophyta. Chlamydial proteins involved in the same biosynthetic pathway may be differentially distributed in the different algal lineages. In the plastidic isoprenoid pathway, the chlamydial genes ispD and ispE (nos 10 and 6 respectively, Table 1) are present in the Rhodoplantae and Viridiplantae, whereas the chlamydial ispG (gcpE; no 24, Table 1) is only present in the Viridiplantae and Bacillariophyta; the Rhodoplantae and Glaucoplantae contain the cyanobacterial homologue of ispG. Similarly, in the fatty acid synthesis pathway, the chlamydial gene fabB (no 16, Table 1) is present in both Rhodoplantae and Viridiplantae, whereas the chlamydial fabI (no 25, Table 1) is restricted to the Viridiplantae (and Bacillariophyta and Apicomplexa); the Rhodoplantae contain the cyanobacterial fabI. The complex distribution pattern of chlamydial proteins in the different lineages of photosynthetic eukaryotes called for an in depth analysis of the phylogeny of each chlamydial protein using a broader taxonomic sampling that included diverse chlamydiae, cyanobacteria and other groups of bacteria such as proteobacteria and Firmicutes (representative trees in Fig. 1A–F, summary in Table 1 and Additional File 2).

The presence of a relatively small number of genes from another domain of life in a genome is usually regarded as signalling horizontal gene transfer (HGT). If a large number of foreign genes that have a single origin are involved, an endosymbiotic gene transfer (EGT) is often the favoured hypothesis. EGT is a special case of HGT, in which the donor (endosymbiont) transfers its genome or part of it to the acceptor (host) and physically persists for an extended period of time within the host. If the donor disappears from its host, it may be difficult, if not impossible, to distinguish between HGT and EGT. As HGT involves two genomes, the direction of gene flow is not a priori known, e.g. the presence of chlamydial proteins in photosynthetic eukaryotes could either signal HGT from chlamydiae to eukaryotes or HGT from eukaryotes to chlamydiae. Phylogenetic analyses can often resolve this conundrum. However, in photosynthetic eukaryotes an additional level of complexity is added by the presence of nuclear-encoded cyanobacterial genes. Since the 39 chlamydial proteins are, with a few notable exceptions (three chlamydial genes in Dictyostelium that may have originated by another HGT; nos 7, 27,39; Table 1), confined to plastid-containing eukaryotes, it is also conceivable that the presence of proteins of putative chlamydial origin in these eukaryotes simply reflects their phylogenetic relationship to the respective cyanobacterial homologues rather than an HGT between chlamydiae and eukaryotes, the former perhaps obscured by loss or modification of the respective genes in extant free-living cyanobacteria. Again phylogenetic analyses can help to reach an informed decision. The presence of cyanobacterial genes in the host genome has another connotation; it may indicate that HGT perhaps did not take place between the chlamydial and eukaryotic genomes directly but between chlamydial and cyanobacterial genomes either before or after the cyanobacterial endosymbiosis (note that the latter scenario requires the simultaneous presence of cyanobacteria and chlamydiae within the host). In this case the chlamydial genes were transferred to the nuclear genome by EGT from the genome of the cyanobacterial endosymbiont.

For 9 of the 39 chlamydial proteins detected in plastid-containing eukaryotes, a cyanobacterial homologue could not be recovered (Fig. 1A and Table 1). Analyses of the remaining 30 chlamydial proteins revealed no specific phylogenetic relationships of the cyanobacterial and chlamydial proteins to the exclusion of other bacterial homologues (e.g. Fig. 1B–F). In fact, the cyanobacterial proteins sometimes formed strongly supported clades with proteins of other bacteria to the exclusion of proteins from chlamydiae (and their associated homologues in plastid-containing eukaryotes). Often, but not always, the cyanobacterial proteins were recovered as sisters to the respective proteins of the Firmicutes (e.g. polyribonucleotide nucleotidyltransferase (bootstrap support: 100/1.0 ML/PP; no 13, Table 1) and aspartate aminotransferase (100%/1.0 ML/PP; no 2, Table 1)). The cyanobacteria/Firmicutes sister group relationship had previously been inferred from multigene phylogenies (e.g. Battistuzzi et al. 18 using 32 proteins (7,597 positions) from 54 different bacterial strains)). Although this result does not rule out an ancient relationship between cyanobacteria and chlamydiae, it strongly suggests that the chlamydial proteins present in photosynthetic eukaryotes are not simply modified cyanobacterial proteins and thus are not of cyanobacterial but of chlamydial origin. That cyanobacterial homologues of chlamydial proteins can be readily identified as such when present in photosynthetic eukaryotes, is exemplified by three proteins with differential distribution in photosynthetic eukaryotes, the cyanobacterial homologue being associated with Glaucoplantae and/or Rhodoplantae (nos 24, 26, 30; Table 1, Fig. 1D), while the chlamydial homologue is present in Bacillariophyta and/or Viridiplantae.

Phylogenetic analyses of the 39 chlamydial proteins also provide insight about the direction of HGT between chlamydiae and plastid-containing eukaryotes. Although two chlamydial proteins occur only in chlamydiae and plastid-containing eukaryotes (see above), and another chlamydial protein (ATP/ADP translocase, no. 3, Table 1) is present only in intracellular bacterial parasites and plastid-containing eukaryotes, all other chlamydial proteins have homologues among several major groups of bacteria and are thus embedded in the bacterial radiation. Conversely, the chlamydial proteins identified here are confined to plastid-containing eukaryotes among the eukaryote radiation (with the exception of the three chlamydial proteins found in Dictyostelium, see above). The presence of chlamydial genes in eukaryotes with secondary plastids (Bacillariophyta and Apicomplexa (Table 1)), and chlorarachniophytes (results not shown)) suggests that these genes have spread from Plantae (comprising Glaucoplantae, Rhodoplantae, and Viridiplantae) to other eukaryote supergroups such as heterokonts, alveolates, and Rhizaria together with their secondary, eukaryotic endosymbionts. In conclusion, chlamydial proteins in plastid-containing eukaryotes are most likely derived from chlamydiae by HGT/EGT.

When and how often did HGT/EGT between chlamydiae and plastid-containing eukaryotes occur? For chlamydial proteins with homologues in all three lineages of the Plantae (5 chlamydial proteins; Table 1), a sister group relationship was recovered between chlamydiae and Plantae for each protein (e.g. Fig. 1B). This suggests that HGT from chlamydiae to plastid-containing eukaryotes predated the divergence of the Glaucoplantae, Rhodoplantae and Viridiplantae.

For 8 additional proteins, a sister group relationship between chlamydiae and Rhodoplantae+Viridiplantae was observed (with BS support ranging from 79–100%; e.g. Fig. 1C; nos 6, 7, 10, 11, 15–18, Table 1) strongly suggesting that HGT for these proteins predated the split between Rhodoplantae and Viridiplantae. It is anticipated that most, if not all of these 8 proteins will also contain homologues in the Glaucoplantae, which are, however, currently unknown reflecting the paucity of genome information in the Glaucoplantae (see above). For another 7 chlamydial proteins in which a monophyletic origin of chlamydiae, Rhodoplantae, and Viridiplantae was recovered (Table 1), either the branching order among chlamydiae, Rhodoplantae and Viridiplantae was unresolved (BS support ≤ 63%; nos 9, 13, 14, 19–21; Table 1) or the (single) protein occurred only in chlamydiae and plastid-containing eukaryotes (putative glycerol-3-phosphate acyltransferase; Fig. 1A). With the addition of more taxa, in particular from the Glaucoplantae, the branching order between chlamydiae and plastid-containing eukaryotes for these 7 chlamydial proteins may be resolved and a monophyletic origin of the chlamydial proteins in plastid-containing eukaryotes likely be revealed. Only one chlamydial protein (putative 7-dehydrocholesterol reductase: no 8, Table 1) deviates from this scheme (for this protein, long branch attraction between Rhodoplantae and chlorophytes, and the absence of any bacterial proteins other than Candidatus Protochlamydia amoebophila and Coxiella may have obscured phylogenetic relationships among taxa; Table 1).

For chlamydial proteins that display a sister group relationship between chlamydiae and Viridiplantae (13 proteins; Table 1), either the homologue from Rhodoplantae is missing (7 proteins; nos 23, 27–29, 31–33) or the Rhodoplantae homologue is specifically associated (often as a sister group) with cyanobacteria (e.g. Fig. 1D; 6 proteins; nos 22, 24–26, 30, 34; Table 1). The first situation may reflect selective loss of chlamydial genes in the rhodoplant lineage, the second offers several contrasting explanations (see below).

Chlamydial proteins that reveal a sister group relationship between chlamydiae and Rhodoplantae (3 proteins; Table 1) also display homologues in the Viridiplantae, whose positions, however, remained unresolved in the phylogenetic trees (Fig. 1E; nos 35–37; Table 1).

Finally, chlamydial proteins showing a sister group relationship between chlamydiae and Bacillariophyta to the exclusion of Plantae (putative folylpolyglutamate synthase (Fig. 1F) and transketolase; nos 38 and 39 respectively; Table 1) either contain the three lineages of Plantae in a monophylum together with cyanobacteria (transketolase) or two lineages of Plantae (Rhodoplantae and Viridiplantae) together with cyanobacteria, Firmicutes and other bacteria in an unresolved radiation (Fig. 1F). It is likely that at least the transketolase of the Bacillariophyta originated by an HGT involving a chlamydial donor different from the donor(s) that contributed the chlamydial genes in the Plantae (see also 19).

The overall conclusion from the phylogenetic analyses is that the vast majority of the 39 chlamydial proteins detected in photosynthetic eukaryotes during this study (a minimum of 31, perhaps all, except one) presumably entered the Plantae by HGT from chlamydiae before the divergence of the Glaucoplantae, Rhodoplantae and Viridiplantae.

Another important question relates to the possible nature of the donor in the HGT/EGT of the chlamydial genes into plastid-containing eukaryotes. For 7 chlamydial proteins, Candidatus Protochlamydia amoebophila was the only member of the chlamydiae in the trees; proteins from other chlamydiae could not be retrieved from the data bases (Fig. 1F; Table 1). The phylogenetic tree for one of these proteins, asparaginyl-tRNA synthetase (no 1; Table 1), also included all three lineages of the Plantae, and in this case, P. amoebophila formed a weakly supported sister group to the Plantae + Bacillariophyta (58% BS support for Plantae + Bacillariophyta; Table 1). For the other 6 proteins, P. amoebophila was either in an unresolved position among the photosynthetic eukaryotes (3 proteins; nos 8, 9, 20; Table 1), was the only bacterial protein in a tree, that, in addition to P. amoebophila, comprised only one lineage of Plantae (Viridiplantae; hypothetical protein pc0324; no 32; Table 1) or was a sister group to the Bacillariophyta (e.g. Fig. 1F; 2 proteins; nos 36, 38; Table 1).

Four of the 5 trees that contained all three lineages of Plantae, displayed, in addition to P. amoebophila, other chlamydiae (Fig. 1B; nos 2–5; Table 1). In one of the 4 trees (aspartate aminotransferase; no 2; Table 1), the chlamydiae were paraphyletic (with P. amoebophila as a sister group to the Plantae (BS 77% for Plantae), in two others (nos 4 and 5; Table 1) they were monophyletic, with P. amoebophila as their first divergence, and the chlamydiae in sister group position to the Plantae. In one tree (ATP/ADP translocase; no 3; Table 1) relationships among the chlamydiae were unresolved. In the 4 trees, the chlamydiae (except P. amoebophila) exhibited relatively long branches, and in the case of the aspartate aminotransferase, this may have caused artificial attraction of these chlamydiae to a very long branch comprising all other bacteria (except cyanobacteria) rendering the chlamydiae paraphyletic (not shown).

Extending the comparison to the other 28 phylogenetic trees that contained chlamydial proteins from both P. amoebophila and other chlamydial taxa, again two topologies emerged: a monophyletic chlamydiae (12 proteins; Table 1) in which P. amoebophila always represented the first divergence within the chlamydiae (e.g. Additional File 3D), or a paraphyletic (sometimes polyphyletic) divergence of the chlamydiae (16 proteins; Table 1) with P. amoebophila either in a sister group position (with or without BS support) to the chlamydial proteins of the plastid-containing eukaryotes (e.g. Additional File 3E; 9 proteins; nos 6, 7, 18, 22–24, 27, 31, 35), or positioned (albeit without resolution) within the Plantae (3 proteins; nos 12, 15, 16, Table 1). For 4 chlamydial proteins a sister group topology between the chlamydiae (excluding P. amoebophila) and either a single clade of plastid-containing eukaryotes (twice with Viridiplantae (nos 21, 28; Table 1), and once with Rhodoplantae (no. 19; Table 1); in nos 19 and 21 without bootstrap support; Table 1) or with two clades of plastid-containing eukaryotes (Viridiplantae and Bacillariophyta; no 26; Table 1) was observed. In all 10 phylogenetic trees in which the chlamydiae were paraphyletic and P. amoebophila was sister to the plastid-containing eukaryotes (see above), the chlamydiae (excluding P. amoebophila) exhibited much longer branches than the branches of both P. amoebophila and the plastid-containing eukaryotes (which may have drawn them closer to the long branches of other bacteria or conversely resulted in short branch attraction of P. amoebophila and the plastid-containing eukaryotes), again likely rendering the chlamydiae paraphyletic (see above). Support for the monophyly of the chlamydiae is strongest (≥ 95% BS; 6 proteins; nos 13, 17, 30, 33, 37, 39) in cases in which the branch lengths of P. amoebophila and the other 4–8 chlamydiae are of comparable lengths (not shown) or in which branch lengths of the plastid-containing eukaryotes and the bacteria (other than chlamydiae) are of similar lengths (RNA delta(2)-isopentenylpyrophosphate transferase; Additional File 3B; no 4; Table 1). When support for the monophyly of the chlamydiae was strong (≥ 95% BS), so was support for the sister group relationship between chlamydiae and plastid-containing eukaryotes (BS ≥ 96%; 5 proteins; nos 4, 13, 17, 33, 37; Table 1).

From these data it is concluded (1) that the chlamydiae are monophyletic, (2) that the apparent paraphyly (polyphyly) of chlamydiae seen in some trees is the result of a long branch (or short branch) attraction artifact because of largely differing evolutionary rates between the genes of P. amoebophila and the later diverging intracellular chlamydial parasites of animal hosts, (3) that the later diverging chlamydiae lost several genes that P. amoebophila (and the plastid-containing eukaryotes) retained, (4) that the donor of the chlamydial genes found today in the plastid-containing eukaryotes was related to the common ancestor of P. amoebophila and other chlamydiae, and thus (5) that the HGT/EGT of chlamydial genes into plastid-containing eukaryotes occurred more than 700 MYA, the estimated time of divergence of P.amoebophila from other chlamydiae 14. The apparent ancient origin of the chlamydial genes in plastid-containing eukaryotes is independently supported by the recovered monophyly of five chlamydial proteins found to date in all three lineages of Plantae (with 58 – 100% BS support for the monophyly of the Plantae; Table 1). The divergence time between Rhodoplantae and Viridiplantae has been variously estimated, based on molecular clock analyses, to be at least 1,100 MYA 202122.

Because chlamydial genes apparently entered the nuclear genome of plastid-containing eukaryotes before the divergence of the three major lineages of Plantae (see above), their fate can be traced through algal evolution and subsequent secondary endosymbioses in much the same way as that of plastidial proteins of cyanobacterial origin. Whereas the monophyly of the Plantae is rarely questioned today (but see 23 and reviews by 2425), the sequence of evolutionary divergence of the three principal types of plastids that correspond to the three lineages of Plantae, namely cyanelles (Glaucoplantae), rhodoplasts (Rhodoplantae) and chloroplasts (Viridiplantae) still remains unresolved. All possible topologies of the three lineages of plastids (or the corresponding Plantae) have been recovered in phylogenetic analyses, i.e. a sister group of Rhodoplantae and Viridiplantae (R+V; 262728), a sister group of Glaucoplantae and Rhodoplantae (G+R; 2930), or a sister group of Glaucoplantae and Viridiplantae (G+V; 26313233. Often plastid and host trees yielded conflicting results and statistical support was low, taxon sampling was insufficient, especially in analyses of host genes (10–16 taxa of Plantae; 33) or apparently long-branch attraction artefacts prevailed.

Chlamydial proteins do not (yet) shed light on the sequence of divergence of the three lineages of Plantae, likely because too few such proteins are currently known from the Glaucoplantae. For the 5 chlamydial proteins present in all three lineages of the Plantae, none of the three topologies obtained for the Plantae was significantly supported by bootstrap values (Table 1). Two proteins supported a G+R sister group (nos 3 and 5, Table 1), whereas two other proteins supported a R+V sister group (nos 2 and 4, Table 1). The fifth chlamydial protein (Asparaginyl-tRNA synthetase, no. 1, Table 1) supported a G+V sister group, however the Viridiplantae were paraphyletic in this analysis (Table 1). For two additional proteins, chlamydial homologues were present only in the Viridiplantae, whereas the Glaucoplantae and Rhodoplantae contained the cyanobacterial homologues (nos 24 and 30, Table 1). The distribution of the latter two proteins among the three lineages of Plantae is most parsimoniously explained by either a G+R or R+V topology. To further evaluate the different topologies, phylogenetic analyses with concatenated data sets using P. amoebophila and two other chlamydiae as outgroup were performed. In a 4-protein analysis (aspartate aminotransferase, ATP/ADP translocase, tRNA delta(2)-isopentenylpyrophosphate transferase and diphosphate-fructose-6-phosphate 1-phosphotransferase; nos 2–5, Table 1) using 576 aa positions, a moderate (71%) BS support for a sister group relationship between Rhodoplantae and Viridiplantae (R+V) was obtained (data not shown). When more protein sequences of chlamydial origin will become available from the Glaucoplantae, chlamydial proteins may provide a unique opportunity to address the pattern of evolutionary divergence in the three lineages of Plantae.

The complex distribution pattern of chlamydial proteins in the plastid-containing eukaryotes in combination with the detailed phylogenetic analyses (see above), suggest that chlamydial proteins were differentially lost from individual lineages of Plantae. Of the 13 chlamydial proteins recovered from Viridiplantae only, in seven cases chlamydial homologues were absent from Rhodoplantae suggesting that these were lost in the Rhodoplantae after their divergence from the Viridiplantae. For two of these proteins (nos 23 and 28, Table 1), the chlamydial homologues were also absent and thus presumably lost in the chlorophyte sublineage of the Viridiplantae, perhaps indicating that these proteins perform specific functions confined to the streptophyte sublineage of the Viridiplantae. It should be noted that taxon sampling in the Rhodoplantae was mostly limited to Cyanidioschyzon merolae and Galdieria sulphuraria, two members of a specialized group of thermophilic red algae with extremely small genomes (i.e. the Cyanidiophyta). It is thus possible, although not perhaps likely, that the absence of these chlamydial proteins may be confined to this sublineage of the Rhodoplantae and that more "typical" Rhodoplantae will be shown to contain them. For the other 6 chlamydial proteins confined to the Viridiplantae, the Rhodoplantae contain a cyanobacterial homologue (nos 22, 24–26, 30, 34, Table 1). This can be interpreted in different ways. It may be argued that this is indicative of a separate origin of chlamydial proteins in the Viridiplantae (but see results and discussion above). Alternatively, it indicates that both the cyanobacterial and the chlamydial homologues were present in the ancestor of the Plantae and persisted until after the divergence of Rhodoplantae and Viridiplantae, when they were differentially lost (the cyanobacterial homologue in the Viridiplantae and the chlamydial homologue in the Rhodoplantae (and likely also in the Glaucoplantae; nos 24 and 30, Table 1)). In case of the phosphate transporter (no. 22, Table 1), it is likely that the cyanobacterial homologue was lost only from the streptophyte sublineage of the Viridiplantae as the chlorophyte sublineage contains a cyanobacterial homologue (not shown). It is important to stress that in no case cyanobacterial and chlamydial homologues of the same protein have been found in a single taxon of Plantae (Table 1) suggesting that their functions (within the plastid) are redundant and their occurrence thus mutually exclusive. Although several different scenarios may explain the simultaneous presence of cyanobacterial and chlamydial homologues of a larger number of proteins throughout the early history of the Plantae, the most intriquing is certainly the simultaneous presence of two different intracellular bacteria throughout the early history of the Plantae, i.e. cyanobacteria and chlamydiae (see Discussion).

For 22 chlamydial proteins putative homologues were identified in the two sequenced genomes of the Bacillariophyta (Fig. 2). Two of these chlamydial proteins apparently occur exclusively in the Bacillariophyta and not in the Plantae (nos 38 and 39, Table 1) which instead contain cyanobacterial homologues of these proteins (e.g. Additional File 3F). Since it is well established that the Bacillariophyta (and the heterokonts in general) obtained their plastids through a secondary endosymbiosis involving a rhodoplant symbiont (reviews by 343536), either the rhodoplant symbiont of the heterokonts still contained both homologues (which were later differentially lost in the Rhodoplantae and the heterokonts) or, at least in the case of transketolase (no. 39, Table 1) more likely, the chlamydial protein was independently acquired by HGT in the Bacillariophyta. The latter scenario is favored for transketolase because Dictyostelium is a sister to the two Bacillariophyta, and all three lineages of Plantae lack the chlamydial homologue.

For the remaining 20 chlamydial proteins, the Bacillariophyta formed a sister group with either the Rhodoplantae (10 proteins; RB clade, Table 1) or the Viridiplantae (10 proteins, VB clade; Table 1). Both topologies received similar support values, however, BS values were sometimes low. The four chlamydial proteins that were detected in all lineages of the Plantae, revealed exclusively an RB clade with variable BS support (94, 66, 80 and 55% BS values; Table 1). Of the 16 chlamydial proteins that occurred in both Rhodoplantae and Viridiplantae, 9 were also present in the Bacillariophyta (Fig. 2). Of those 9 proteins, one protein (no. 10; Table 1) revealed a strongly supported RB clade (94% BS), four proteins a VB clade (nos 6–8, 13, supported by BS values of 68, 54, 100, and 86%, respectively; Table 1), and the remaining 4 proteins exhibited either one or the other topology, but without BS support (nos 9, 11, 12, 17; Table 1). Of the 13 chlamydial proteins that were found only in the Viridiplantae, four were also detected in the Bacillariophyta (Fig. 2), for two of which the Bacillariophyta formed a clade with the Viridiplantae (no. 24; 92% BS; Table 1) or a sister group with the streptophyte sublineage of the Viridiplantae (no. 26; 100% BS; Table 1), for one protein, the VB topology was unsupported (no. 27; Table 1) and for the remaining protein, the Bacillariophyta were sister to the Apicomplexa (no. 25; 64% BS; Table 1). All three chlamydial proteins that were detected only in the Rhodoplantae (Fig. 2), also displayed homologues in the Bacillariophyta: for two proteins, a well-supported RB clade was recovered (nos 35 and 37; 97 and 96% BS support, respectively; Table 1), for the third protein, the Rhodoplantae and Bacillariophyta formed a clade that also included P. amoebophila with 100% BS support (no. 36; Table 1). It should be noted that in two of the chlamydial proteins revealing a well-supported RB clade (nos 3 and 35), the RB clade contained, in addition to the two Cyanidiophyta, also members of the Rhodophyta (the second major lineage of the Rhodoplantae according to 37).

To study the phylogeny of the chlamydial proteins from the Bacillariophyta in more detail, unrooted maximum likelihood phylogenetic analyses of concatenated data sets were performed (Fig. 3). In the first analysis (Fig. 3A), 7 chlamydial proteins that individually had revealed an RB topology (with or without BS support; nos 1–4, 10, 12, 17; Table 1) were concatenated and subjected to phylogenetic analysis using 12 taxa (2675 amino acid positions; Fig. 3A). The taxa were selected to maximize taxon congruence with the second data set of proteins displaying a VB topology (Fig. 3B), i.e. two Bacillariophyta, the chlorophyte and streptophyte sublineages of the Viridiplantae (5 sequences), the Rhodoplantae (2 sequences, except for no. 10 for which only one rhodoplant sequence was recovered) and, if possible, 3 sequences of chlamydiae were included. The RB clade was recovered with 95% BS values (ML) and 1.0 posterior probability in the Bayesian analysis (Fig 3A). In a second analysis (Fig. 3B), 5 chlamydial proteins that had individually revealed a VB topology (nos 6, 7, 9, 11, 13; Table 1; it should be noted that only protein no. 13 received BS support for a VB clade) were concatenated and subjected to phylogenetic analyses using the same 12 taxa as in the concatenated analysis of the proteins with an RB topology (1736 amino acid positions; Fig 3B). In this analysis, the VB topology was strongly supported (>95% BS values in ML and 1.0 posterior probability in the Bayesian analysis; Fig. 3B). Although in the individual analyses of the 5 chlamydial proteins, the position of the Bacillariophyta with respect to the sublineages of the Viridiplantae was not resolved and BS values for 4 of the 5 proteins were below 50% (Table 1), the concatenated data set clearly indicated that the 5 chlamydial proteins of the Bacillariophyta were sister to the respective proteins of the Viridiplantae, i.e. they diverged before the split of the Viridiplantae into its chlorophyte and streptophyte sublineages (Fig. 3B).

From these data the following tentative conclusions are drawn: (1) The Bacillariophyta obtained chlamydial genes from two sources of photosynthetic eukaryotes, the Rhodoplantae and the Viridiplantae, (2) it is very likely that the chlamydial proteins forming RB clades were obtained during the secondary endosymbiosis of a rhodoplant symbiont into a heterokont (or chromalveolate) host, (3) there is preliminary evidence that the endosymbiosis of the rhodoplant occurred before the Rhodoplantae diverged into Cyanidiophyta and Rhodophyta (although this needs to be further studied using a larger taxon sampling in the Rhodoplantae), (4) chlamydial proteins of the Bacillariophyta forming VB clades were obtained by HGT or EGT from a viridiplant alga before the Viridiplantae diverged into Streptophyta and Chlorophyta, (5) since the number of chlamydial proteins forming VB clades in the Bacillariophyta found so far is roughly equivalent to the number of chlamydial proteins forming RB clades, it is suspected that EGT rather than HGT was the source of the viridiplant genes, the endosymbiont (or its plastid) being no longer present in extant heterokonts (or chromalveolates), (6) it cannot currently be decided which of the two secondary endosymbioses preceded the other. If the above conclusions are valid, the two secondary endosymbioses may have taken place during the same time period. Two chlamydial proteins involved in essential plastidic pathways (fatty acid biosynthesis, FABI; isoprenoid biosynthesis, ISPG; nos 24 and 25, Table 1) occur in the Bacillariophyta and Viridiplantae, but not in the Rhodoplantae (both Cyanidiophyta and Rhodophyta; tree not shown), which harbor the cyanobacterial homologues (Table 1). For ISPG, two Glaucoplantae also contain the cyanobacterial homologue. If the rhodoplant symbiont of the Bacillariophyta also contained the cyanobacterial homologues (the most parsimonious scenario), the latter must have been replaced in the Bacillariophyta by the chlamydial homologue from the viridiplant symbiont, requiring the simultaneous presence of both eukaryotic symbionts in the host.

We screened for algal proteins of possible chlamydial origin in the following ways: 1) The JGI databases for Chlamydomonas reinhardtii, Ostreococcus lucimarinus, Ostreococcus tauri, Phaeodactylum tricornutum and Thalassiosira pseudonana were searched for Blast hits with Protochlamydia amoebophilia using the advanced search function and an e-value cut off of exp -20. 2) The proteome of Cyanidioschyzon merolae was downloaded from http://merolae.biol.s.u-tokyo.ac.jp/ and blasted (NCBI BLASTP) locally against the proteom of Protochlamydia amoebophila using an e-value cut-off of exp-20. Proteins and contigs showing similarity over the entire length were selected for further analysis. To exclude false positive we blasted (BLASTP) each algal protein against the NCB non redundant protein database and the databases for Chlamydomonas reinhardtii, Ostreococcus lucimarinus, Ostreococcus tauri, Phaeodactylum tricornutum and Thalassiosira pseudonana at JGI and the Cyanidionschyzon merolae database at http://merolae.biol.s.u-tokyo.ac.jp/. For proteins showing an association with proteins of Chlamydiae in the distance tree generated by the NCBI BLAST server, we assembled a dataset containing all algal proteins and proteins from Arabidopsis, Oryza, at least 5 Chlamydiae (incl. Protochlamydia amoebophilia), at least 5 cyanobacteria and members of the Proteobacteria and Firmicutes and 5 – 10 bacterial strains obtained as top bacterial BLASTP hits using the protein from Protochlamydia as a query.

Alignment of single-gene data sets were generated using CLUSTAL W and manually refined with SeaView 79. Non-alignable regions were excluded prior to phylogenetic analyses. The evolutionary model fitting best the protein data was determined with ProtTest 1.2.6 with deactivated "+F" option 8081. Maximum likelihood trees were done with phyml 2.4.4 set to the optimal evolutionary model and including 200 bootstrap replicates (in most cases WAG+I+Γ; 8283). All trees that were chosen to be depicted, have been subjected in addition to Bayesian inference with MrBayes 3.1.2 84. For each data set, two runs with four chains and 3 million generations have been computed. Likelihood parameters were set to 4 gamma categories and proportion of invarable sites. Computation was done across all available amino acid substitution matrices (command "prset aamodel = mixed"). Every 100^th generation was sampled. Convergence of the runs was checked according to the output of the "sump" command. The output was also used to determine the burn-in phase.

Two concatenated data sets with a reduced taxon sampling have been assembled from single-gene data. In the first of these data sets, single genes that showed a closer relationship of Bacillariophyta to red algae have been combined, whereas the second data set comprised single genes showing a relationship of Bacillariophyta to Viridiplantae. All concatenated data sets have been subjected to the same analysis procedure as the single genes concerning maximum likelihood analysis (see above). Since in the concatenated data sets some sequences for single taxa were missing, no partitions were defined for Bayesian analyses (see legend of Fig. 3, additional file 5 for details).