In all analyses, except analyses of the red/brown algal GroEL orthologues, an RtTEV model, that accommodated positional rate heterogeneity in some form was the model selected as best by ProtTest under the AIC criterion. In the case of the GroEL orthologues in red/brown algae, we noted that small differences in the AIC criterion separated a variety of different models. When an RtREV model was assumed, and positional rate heterogeneity was approximated by either a constant proportion of variable sites or a discrete gamma distribution of rate classes, we noticed that the optimal estimates for these parameters varied among evolutionary lineages (i.e different estimates were obtained for cyanobacterial GroEL1, cyanobacterial GroEL2, red algal/heterokont plastid GroEL1 like sequences, and highly diverged Cpn60-like sequences). When a uniform rate distribution was assumed, p_var values ranged from 0.3 variable sites (cyanobacterial GroEL1) to 0.9 variable sites (highly diverged Cpn60-like sequences). When p_var was set to 1, alpha shape parameter values ranged from 0.2 (for cyanobacterial GroEL1) to 0.8 (for highly diverged Cpn60-like sequences). We suggest that this degree of variation in parameter estimates for phylogenetic grouping of anciently diverged sequences is likely to reflect lineage specific differences in structural and functional constraints 101112. As such it is potentially problematic for phylogenetic reconstruction, since parallel increases in proportions of variable sites in different evolutionary lineages can sometimes induce a form of long branch attraction 101112.

Optimal PhyML (protein maximum likelihood) trees showed similar topological relationships over a wide range of p_var values specified to accommodate positional rate heterogeneity. The robustness of phylogenetic relationships to sampling error was also found to be relatively stable, when evaluated using non-parametric bootstrapping (100 replicates). Figure 1 shows the optimal unrooted phylogenetic tree built assuming an RtREV model with positional rate heterogeneity modelled with a discrete gamma distribution (α = 0.92; 4 discrete rate categories).

Although, it contains more GroEL homologues than was available to Wastl et al. 6, our phylogenetic reconstruction here is nevertheless consistent with observations and inferences made by Wastl et al. 6. Figure 1 shows that (a) the groEL1 like genes from the plastid genome of red algae/heterokonts and cryptophytes are most closely related to the cyanobacterial groEL1 genes. (b) Interestingly the cyanelle groEL1 like homologue is somewhat intermediate between cyanobacterial and red algal/chromist groEL1-like sequences, a finding that may reflect the cyanobacterial-like nature of this plastid. (c) The non-photosynthetic eubacterial groEL sequences are arguably more similar to the groEL2 sequences found in cyanobacteria. In any event, the groEL2 orthologues in cyanobacteria are genetically more diverse than the groEL1 sequences in the same taxa. These observations may suggest that groEL2 orthologues represent an ancestral from of GroEL. However, our inference that structural/functional constraints differ amongst GroEL homologues, means that it is not possible to exclude other interpretations. (d) Assuming the root of the tree joins the branch leading to the non-photosynthetic taxa, Figure 1 places the Gloeobacter "A" and "B" sequences as the ancestral forms of GroEL homologues in photosynthetic taxa. (e) In the nuclear genomes of a diatom (Thalassiosira pseudonana), an alveolate (Plasmodium falciparum), the higher plants and the nucleomorph genomes of a chlorarachniophyte (Bigelowiella natans), a cryptophyte (Guillarida theta) and a red algae (Cyanidoschyzon merolae) groEL2 type sequences are found. In the case of higher plants, two forms of groEL2 occur, and appear to represent forms of groEL duplicated from an ancestral groEL2-like sequence. Inferences concerning the origin of this duplication are potentially problematic. A relatively high bootstrap value in Figure 1 (77%) suggests that it may have occurred prior to the divergence of plastids. However, potential long branch attraction problems concerning the placement of α and β Cpn60-like sequences from eukaryotes make this conclusion tentative. Gene duplication within the green lineage and differences in the functional/structural constraints of green α and β Cpn60-like sequences might also explain the results observed Although Chlamydomonas reinhardtii Cpn60 α and β sequences (AAA98642 and AAA98643) are not included in our phylogenetic analysis shown in Figure 1 (because their inclusion significantly reduced the alignment length, and increased phylogenetic uncertainty). However, it is clear from other phylogenetic analyses (unpublished) that this green alga contains the two forms of groEL2 also present in higher plants.

Our phylogenetic reconstruction suggests a complex pattern of genome transfers, losses and duplications in the evolution of groEL sequences from photosynthetic taxa. In the earliest cyanobacterial-like prokaryotes, the ancestral groEL sequence appears to have duplicated to form an "A" and a "B" type sequence that is still present today in the genome of Gloeobacter 1415. It appears that the "A" (groEL2-like) and "B"(more groEL1-like) forms have been inherited by most cyanobacteria, and also the endosymbiont(s) involved in primary plastid endosymbiosis. During the process of endosymbiosis, the groEL2-like orthologue has been transferred from the endosymbiont genome to the nuclear genome in heterokonts, red algae, green algae and high plants, and cryptophytes. We predict that this also will be true for haptophytes and glaucocystophytes. Less certain may be prediction of the pattern of evolution in peridinin-containing dinoflagellates, which in other respects appear highly distinctive in their evolution 1617. An interesting observation is that Cyanophora paradoxa, the molecular prototype of the glaucophytes, shows not only archaic features by encoding a groES in its plastid genome, but also by containing a groEL1 sequence, somewhat intermediate in sequence identity between cyanobacterial and plastid encoded groEL1 genes. In the case of the chlorophyte lineage, it appears that it has been the groEL1 form that has been lost, and in Chlamydomonas and higher plants, this has been replaced by a duplicated form of a groEL2-like sequence. This situation may also be true for the endosymbiont of the chlorarachniophyte B. natans (generally assumed to have been a green alga 16), since phylogenetic reconstructions, including those with partial Chlamydomonas GroEL sequences (not shown), provide some support for Bigelowiella and cpn60α containing taxa being most closely related (e.g. Figure 1). Several investigations on the evolution of the Plasmodium apicoplast indicate a red algal origin for this organelle 1819. If so, then based on the generalisations stated above, one would expect that the groEl/cpn60 distribution should be similar to the situation found in heterokonts. Additional duplications, of groEL1 genes (in the presence of groEL2) have also occurred in some species of filamentous cyanobacteria and this observation is discussed in the following section.

In all eukaryotes and prokaryotes that carry out oxygenic photosynthesis, and for which genome data is complete, two different forms of groEL/cpn60 are known to exist. Thus, at least two copies of the gene or protein seem to be necessary for a complete chaperon function. Further, the importance of having two divergent forms of GroEL for protein folding is suggested from observations on GroEL sequences in Synechocystis sp. PCC 6803, where it has been observed that GroEL1 and GroEL2 respond differently to heat shock and light conditions 2021. An interesting speculation is that the number and genetic diversity of GroEL homologues reflects complexity of morphotype in cyanobacteria. This is suggested from comparison within filamentous cyanobacteria. Those strains that harbour three GroEL genes exhibit complex developmental stages (akinetes and hormogonia) whereas filamentous strains with a simpler morphotype, such as Nostoc sp. PCC 7120, harbour only two GroEL genes. Further study is needed to evaluate whether particular groEL homologues are specifically expressed in different developmental stages. Interestingly, P. falciparum encodes one cpn60 gene, but no groEL1 in the apicoplast genome. If phylogenetic inferences of a close relationship with red algae 1819 are correct, then this observation may reflect relaxed constraints for protein folding for proteins of the apicoplast, and this speculation is also worth further investigation.

Cyanobacterial groEL genes were retrieved from Genbank and the cpnDB chaperonin sequence database 22. Two groEL genes were found to be present in Synechocystis sp. PCC 6803. One of these genes, termed groEL1 (slr2076), is arranged in an operon together with groES, whereas the other, groEL2 (sll0416), is not adjacent to a small subunit gene 23. In comparing homologues from other cyanobacteria to those of Synechocystis sp. PCC 6803, we have adopted the terminology of "groEL1" and "groEL 2". All unicellular forms of cyanobacteria retrieved from database searches were found to contain two GroEL genes. In contrast, filamentous strains showed variation in the number of groEL genes. Nostoc punctiforme and Anabaena variabilis ATCC29413 harbour three groEL genes, whilst Nostoc sp. PCC 7120 with two groEL genes, was similar to that of unicellular forms. Genbank also provided us with several entries for cpn60 from several land plants, Chlamydomonas reinhardii and the nucleomorph of the cryptophyte Guillardia theta. In a recently finished genome project on the nucleomorph genome of the chlorarachniophyte Bigelowiella natans (Gilson & McFadden, unpublished), a further nucleomorph-encoded cpn60 gene has been annotated. Two new genome projects on Cyanidioschyzon merolae 24 and Thalassiosira pseudonana 25 were also the source for additional nuclear-located cpn60 genes. In PlasmoDB 26, the genome data base for Plasmodium falciparum, two different cpn60 genes have been annotated in the nuclear genome 27, one encodes a mitochondrial, and the other an apicoplast targeted copy 28. Additional, BLAST searches of NCBI were made against available plastid genomes using cyanobacterial groEL genes. Significant hits were obtained with the plastid genomes from red (Gracilaria verrucosa, Cyanidium caldarium, Porphyra purpurea, Cyanidioschyzon merolae) and secondary red plastids (Odonetella sinensis, Guillardia theta, Pyrenomonas salina, Thalassiosira pseudonana). These sequences, together with the nuclear, nucleomorph, plastid, and cyanobacterial sequences were aligned using the progressive alignment procedure implemented in CLUSTALX 29, edited to remove any ambiguously aligned regions, and conserved blocks of aligned residues (containing 495 amino acids) were used for phylogenetic analyses. The aligned data matrix and accession details are available from the authors on request.

The online version of ProtTest v1.2.6 3031, implementing the Akaike Information criterion (AIC) was used to select the most appropriate amino acid substitution models for tree building analyses. The cyanobacteria GroEL1 and GroEL2 datasets used for estimating parameters had identical taxon sampling (10 taxa: Synechocystis PC 6803, Crocosphaera watsonii WH 8501, Trichodesmium erythraeum IMS101, Synechococcus elongatus 7942, Synechococcus vulcanus, Synechococcus sp. wh8102, Prochlorococcus marinus str. MIT9313, Prochlorococcus marinus str. CCMP1375, Prochlorococcus pastoris str. CCMP1986). Estimates were also made for bacteria (9 taxa: Thermus thermophilus, Rhodospirillum rubrum, Neisseria meningitidis, Geobacter metallireducens, Pseudomonas aeruginosa, Mycobacterium avium, Bacillus subtilis, Aquifex aeolicus and Chlorobium tepidum; and for red/brown algal chloroplast located orthologues (8 taxa: Guillardia theta, Pyrenomonas salina, Cyanidium caldarium, Porphyra purpurea, Gracilaria tenuistipitata var. liui, Odontella sinensis, Thalassiosira pseudonana and Cyanidioschyzon merolae) and also for highly diverged Cpn60-like sequences (5 taxa: Thalassiosira pseudonana, Cyanidioschyzon merolae, Plasmodium falciparum, Bigelowiella natans and Guillardia theta).

Trees were reconstructed from amino acid sequences using the windows version of PhyML 32. Trees were built assuming an RTRev model and the optimal tree shown in Figure 1 displayed using SplitsTree4.0 33. The robustness of phylogenetic reconstructions to variations in assumptions of positional rate heterogeneity was investigated by assuming (a) different proportions of variable sites (p_var range = 0.3–1.0) and (b) a discrete gamma distribution of rate classes and a range of alpha shape parameter values (0.2–1). Non-parametric bootstrap trees were analyzed to assess the significance of sampling variability.