Both ribosomal data sets passed the chi-square test for homogeneity of base frequencies across taxa, whereas the rbcL data set failed the test (Additional file 1). After exclusion of third codon positions, the rbcL data set passed the test, indicating that the heterogeneity of base frequencies was restricted to third codon positions. This became also obvious by separate computation of mean values and standard deviations of base frequencies for first, second and third codon positions across all twenty-three taxa. The second codon position was most homogeneous concerning standard deviations (A: 27.8 ± 0.3%; C: 21.4 ± 0.4%; G: 19.9 ± 0.4%; T: 30.9 ± 0.3%), the first codon position had an intermediate position (A: 24.6 ± 1.0%; C: 17.8 ± 1.4, G: 38.9 ± 0.7, T: 18.7 ± 1.1), whereas the third codon position was most heterogeneous (A: 29.0 ± 2.6%; C: 16.1 ± 4.6%; G: 10.3 ± 3.8%; T: 44.6 ± 3.2%). Phylogenetic analyses with bootstrap resampling under all optimality criteria and a Bayesian analysis, however, showed that despite of an obvious bias, these positions contributed the most to the support of the clades in the rbcL data set. This was confirmed by separate phylogenetic analyses of first, second and third codon positions (Additional file 1). The highly conserved rbcL protein sequences consistently failed to recover three clades that were otherwise highly supported in all DNA sequence data sets (Cryptomonas marssonii, C. ovata and C. pyrenoidifera), because phylogenetic information was predominantly based on synonymous substitutions. Therefore, we used the rbcL gene tree with complete codon positions for a comparison with the nuclear and nucleomorph ribosomal DNA phylogenies (Figure 1A to 1C).

Almost all Cryptomonas clades were unequivocally recovered with significant or at least moderate support in nuclear, nucleomorph and plastid gene trees (Figure 1A to 1C). This refers to C. curvata (Ccu), C. marssonii (Cma), C. ovata (Cov), C. paramaecium (colorless strains; Cpa), C. pyrenoidifera (Cpy), and C. tetrapyrenoidosa (Cte; clades named according to Hoef-Emden and Melkonian 2003). C. borealis (Cbo) was significantly supported in nuclear and nucleomorph phylogenies but not in all phylogenetic analyses in the rbcL phylogeny (Figure 1C). Significant support for this clade, however, was found in the rbcL protein phylogeny apparently due to nonsynonymous substitutions (tree not shown; for support values, see Additional file 1). Clade NoPyr (for no pyrenoids 11), otherwise highly supported in nuclear and nucleomorph phylogenies, could not be resolved in the rbcL phylogeny (Figure 1C).

In all phylogenies, C. borealis, C. gyropyrenoidosa and C. lundii formed a "super-clade" together with the colorless C. paramaecium (termed clade LB for long-branch in 11; Figure 1A to 1C). Only in the nucleomorph SSU rDNA tree, however, convincing support for clade LB was found (Figure 1B). In the nucleomorph SSU rDNA and rbcL phylogenies, representing the two genomes of the complex plastid, evolutionary rates and topologies of the strains in clade LB resembled each other. In both phylogenies, evolutionary rates were extremely accelerated in clade LB, and the branching pattern was similar, except for the position of C. gyropyrenoidosa, which was a sister to C. lundii in the nucleomorph SSU rDNA (but without bootstrap support), but not in the rbcL phylogeny (where it was the first divergence). In the nuclear-encoded ribosomal DNA phylogeny, predominantly the strains of clade NoPyr displayed increased evolutionary rates, whereas evolutionary rates were less pronounced in C. borealis and C. paramaecium of clade LB (Figure 1A). In clade NoPyr, an acceleration of evolutionary rates was also present to a lesser extent in the nucleomorph SSU rDNA; in the rbcL phylogeny, however, branch lengths of this clade were inconspicuous (Figure 1B and 1C).

In Figure 1D, the phylogenetic trees of Figure 1A to 1C were scaled to the same substitution rate and in Figure 2, the maximum likelihood distances of C. pyenoidifera strain M1077 to the other taxa were plotted in a chart diagram for a direct comparison of genetic divergences. Among the three data sets, the rbcL data displayed generally the highest substitution rates and genetic distances. In the nucleomorph SSU rDNA, the evolutionary rates of C. gyropyrenoidosa, C. borealis and C. paramaecium were in an intermediate position. Apparently, evolutionary rates in clade LB increased successively from host to nucleomorph to plastid genome.

In Table 1, the codon usages of the six amino acids with two-fold degenerate NNY codons (asparagine, histidine, aspartate, tyrosine, cysteine and phenylalanine) are listed in absolute counts computed from the 396 codons that were included in the phylogenetic analyses. Cysteine was exceptional in codon usage in that it mostly showed a preference for UGU over UGC, thus it will not be further discussed (Table 1). For the remaining five amino acids, NNC codons were always preferred over NNU codons in C. marssonii, C. pyrenoidifera, C. tetrapyrenoidosa and in clade NoPyr (Table 1). In all strains of clade LB, on the other hand, indications for a change of codon usage were found, although to different extent. In almost all strains of C. paramaecium (except for histidine in strain CCAP 977/1) and C. borealis (except for asparagine in strain SCCAP K-0063) codon preferences were inversed from NNC to NNU for these amino acids (Table 1). C. lundii and C. gyropyrenoidosa were in an intermediate position concerning codon usages. In C. lundii only in two amino acids, aspartate and tyrosine, codon usage was inversed to prefer GAU over GAC (aspartate) and UAU over UAC (tyrosine), whereas in C. gyropyrenoidosa only one amino acid, aspartate, was affected (Table 1). Inversed codon usages were also found in two clades that were not part of LB, C. curvata (histidine and aspartate in strain CCAC 0080, aspartate in strain CCAC 0006) and C. ovata (aspartate and tyrosine; Table 1). In the three phylogenies, C. ovata displayed slightly increased evolutionary rates in nucleomorph SSU rDNA and rbcL phylogenies, whereas C. curvata had slightly longer branches only in the nuclear ribosomal DNA phylogeny (Figure 1A to 1C).

Most of the Cryptomonas clades were recovered with high support values in phylogenies of the concatenated nuclear ribosomal DNA sequences (SSU rDNA, ITS2 and partial LSU rDNA), of the nucleomorph SSU rDNA and of the plastid-encoded rbcL gene, but obvious differences in evolutionary rates among the different clades and genomes were displayed.

In the "super-clade" LB, consisting of three photosynthetic (C. borealis, C. gyropyrenoidosa and C. lundii) and one heterotrophic Cryptomonas species (C. paramaecium), largely congruent branching patterns and extreme evolutionary rates in the nucleomorph and plastid gene phylogenies suggested coevolution under similar selective constraints, as if the two genomes of the complex plastid were a genetic unit in this clade. In the nuclear ribosomal DNA phylogeny, an increase in evolutionary rates was in part also present but less pronounced. Support for this clade was low in the nuclear ribosomal DNA phylogeny, although several parts of the nuclear ribosomal operon were concatenated to improve resolution (the nuclear SSU rDNA alone failed to recover clade LB, but increased evolutionary rates in C. paramaecium and C. borealis were more pronounced than in the concatenated data set; not shown).

In a different clade, that also consists of photosynthetic and colorless Cryptomonas taxa, clade NoPyr, the situation was reversed; coevolution with increased evolutionary rates seemed to have taken place in the nuclear and nucleomorph genes 12, whereas no acceleration of evolutionary rates could be observed in the rbcL gene phylogeny (this study). We did not obtain an rbcL PCR product, however, from the colorless strains of clade NoPyr [12, this study].

Cho et al. 21 demonstrated that extremely accelerated evolutionary rates were present in three mitochondrial genes (two protein-coding genes, cox1, atp1, and one RNA-coding gene, rrn16, the gene for the SSU rDNA in mitochondria) in the flowering plant genus Plantago, but not in plastid or nuclear genes of the same taxa. We chose two RNA-coding genes and a protein-coding gene as representatives for three of the four genomes in Cryptomonas. Despite their differing functions, the phylogenetic trees suggested that at least two (clade NoPyr), or even all three genomes (clade C. borealis and C. paramaecium) may have evolved in parallel under similar selective constraints or by interacting with each other.

Possible explanations for accelerated evolutionary rates of DNA sequences include relaxation or loss of functional constraints due to either changes in mode of nutrition, adaptations to new environmental conditions, genetic bottlenecks or obligate asexuality 22232425. Endosymbiotic, parasitic and organellar genomes are notorious for high A+T contents in their genomes likely caused by biased substitution rates under neutral mutational pressure 19202627. Minimum amounts of G and C are required to maintain the codon information for a functional protein or to preserve the secondary structure of an RNA. Depending on the strengths of the functional constraints, the resulting selection bias may differ from the genome composition bias to varying degrees (reviewed for protein-coding genes in 16). Thus, lineage-specific relaxed selective constraints may be identified by increases in A+T content.

This notion is supported by the observation that the nucleomorph SSU rRNA genes in clade LB accumulate mononucleotide repeats of A and T in highly variable regions 1112. In functional protein-coding genes, the triplet structure constrains mutation rates by selection. Synonymous substitutions do not replace amino acids, thus are more likely to occur than nonsynonymous substitutions. In previous studies, however, also synonymous substitutions were reported to be skewed towards specific codons in correlation with expression levels of the respective protein 152829. The codon biases were explained as a result of a competition between selection and genome compositional bias 17. In highly expressed genes, codons with abundant or perfectly matching tRNAs (major codons) are apparently preferred over codons are translated by rare or "wobbling" tRNAs (minor codons) 2930. In plastid genomes, usually only 30 to 31 tRNAs are available to translate all 61 codons (in the Guillardia theta plastome, 30 tRNAs were found) 1016. In two-fold degenerate NNY codons, the preferred major codon in highly expressed genes is usually NNC, thus codon bias due to selection can be comparably easily distinguished from codon bias due to mutation drift 1618. Among the six amino acids with two-fold degenerate codons (asparagine, aspartate, histidine, tyrosine, phenylalanine and cysteine), cysteine seems to be the sole exception [16, this study].

In the rbcL genes of most Cryptomonas clades, the major NNC codons for asparagine, aspartate, histidine, tyrosine, or phenylalanine were preferred over their NNU alternatives, however, codon preferences were reversed in several or all of these amino acids in clade LB [this study]. There was even a gradient of decreasing selective constraints and presumably also expression levels along the LB clade: In the early diverging C. gyropyrenoidosa and C. lundii, reversed codon usages were found in only one or two amino acids, whereas in the late diverging C. borealis and C. paramaecium in four or five NNY-coded amino acids, NNU codons were preferred. Morton and Levin used the codon adaptation index (CAI) to compare codon usage of two-fold degenerate NNY codons in psbA genes among dicot and monocot plants, and discussed putatively decreasing selective constraints from basally to terminally diverging lineages 18. However, no hemi- or holoparasitic angiosperm plants were included in their study.

In previous studies, convergent codon usage resulted in artificial tree topologies 31. Despite an obvious bias in codon usage, the rbcL phylogeny was largely confirmed by the nuclear and nucleomorph ribosomal DNA phylogenies. It is likely that the rbcL genes examined in this study had not yet diverged enough to cause artifacts. It may have been different, though, if rbcL had been used to infer phylogenetic trees across cryptophyte genera. For higher level phylogenies, it may, thus, be a better option to use protein sequences instead.

One of the explanations for accelerated evolutionary rates and relaxed functional constraints in plastid genomes is loss of photosynthesis, since this usually results in large-scale degradation and compaction of plastomes leading to loss of almost all photosynthetic genes, except perhaps for rbcL 223233.

In the cryptophyte Guillardia theta, photosynthesis genes are spread across plastid (46 genes), nucleomorph (30 genes) and nucleus (α-subunits of phycoerythrin, and an unknown number of additional photosynthetic genes) 9103435. Thus, loss of photosynthesis is not an unlikely explanation for a parallel acceleration of evolutionary rates across three genomes in C. paramaecium. However, observations of elevated evolutionary rates in closely related photosynthetic taxa contradict this notion. Instead of being the cause for increased substitution rates, loss of photosynthesis may rather be a result of an accelerated evolution that had started already in the photosynthetic ancestors of the colorless lineages [12, this study]. Similar observations have been made in plastid gene phylogenies of hemi- and holoparasitic land plants 36.

Previous studies have shown that mutations in genes of DNA repair or DNA replication may result in overall increases of substitution rates in bacterial and eukaryotic genomes, e.g. 37. Many genes of the cryptophyte nucleomorph have been transferred to the host nucleus including DNA polymerases 91035. Thus, some potential mutator genes in cryptophytes can be expected to be nuclear-encoded. It is tempting to speculate that a spontaneous mutation in a nuclear-encoded plastid-targeted protein, for example in the proofreading subunit of a DNA polymerase III or in a DNA repair enzyme, could have accelerated successively mutation rates in nuclear, nucleomorph and plastid DNA in the photosynthetic ancestors of clade LB. Accelerated mutation rates may have resulted in loss of photosynthesis in C. paramaecium, which in turn perhaps resulted in less functional constraints on the rbcL protein and, thus, in further increase of evolutionary rates.

Another possible cause for accelerated evolutionary rates in clade LB was discussed previously 12. The genus Cryptomonas is dimorphic, a feature that usually correlates with sexual reproduction. Final proof for sexual reproduction is still missing, but, however, in clade LB only strains with campylomorph cells were found1112. It may, thus, as well be possible that loss of sexual reproduction caused an increase in mutation rates by loss of recombination affecting also nuclear-encoded plastid-targeted proteins. However, the observed increase in evolutionary rates from host to nucleomorph to plastid genome suggests that the evolutionary processes may have started in the plastid genome.

Photosynthetic and heterotrophic Cryptomonas strains were obtained from different algal culture collections (Table 2). Photosynthetic strains were maintained in modified WARIS-H freshwater culture medium 3839, and heterotrophic strains in biphasic soil/water medium with one-eighth of a pea for supply with organic substances. Strains were grown at 15°C under a 14/10 h light/dark regime (15–35 μmol photons m^-2 s^-1; photosynthetic strains) or in the dark (colorless strains).

Total genomic DNA was isolated from the cells with the DNeasy Plant Mini Kit according to the manufacturer's protocol (Qiagen, Hilden, Germany). PCR amplification of nuclear SSU rDNA, ITS2 and partial LSU rDNA, and of nucleomorph SSU rDNA with nucleus- or nucleomorph-specific primers followed previously described protocols 1140. For PCR amplification of cryptophyte rbcL genes, new primers were designed using an alignment of bangiophyte or florideophycean red algal and cryptophyte rbcL sequences (cryptophyte sequences: Chroomonas sp., acc. no. AY119781; Cryptomonas paramaecium, acc. no. AY119780; Guillardia theta, acc. no. AF041468; Pyrenomonas helgolandii, acc. no. AY199782). Similarly, new sequencing primers were constructed using the same alignment (sequences of PCR primers and sequencing primers for rbcL are listed in Additional file 2). For PCR amplification of rbcL DNA sequences, the same cycling protocol as for the ribosomal DNA sequences was used except for a decrease of the annealing temperature (predenaturation for 3 min. at 95°C; 30 cycles: 1 min. at 95°C, 2 min. at 45 or 50°C, 3 min. at 68°C). PCR products were purified with the Dynabead M-280 system according to the manufacturer's protocol (Dynal, Oslo, Norway). For bidirectional sequencing, two sets of primer pairs were used for each PCR product; the forward primers were labeled with IRDye-800 and the reverse primers with IRDye-700 (see Additional file 2). Double-stranded sequences were determined with a Li-Cor 4200L bidirectional sequencer (Li-Cor Biosciences, Bad Homburg, Germany).

The rbcL nucleotide and protein sequences were prealigned with clustalw and refined by eye using the multiple alignment sequence editor SeaView 41. The ribosomal DNA sequences were manually aligned according to secondary structure; non-alignable regions were excluded prior to the phylogenetic analyses.

Since the taxon sampling was congruent for plastid-, nucleomorph- and nucleus-encoded sequences, all unrooted data sets comprised 23 taxa (accession nos. are listed in Table 2). The unrooted rbcL nucleotide data set consisted of 1188 positions and was translated to perform phylogenetic analyses of protein sequences or modified for phylogenetic analyses of single codon positions (396 positions each data set; see Additional file 1 for additional information about the data sets). A rooted data set of rbcL nucleotide sequences consisted of 46 taxa and 990 positions, including 14 rhodophyte rbcL sequences as outgroup taxa (Additional file 3). The nuclear ribosomal DNA sequences were concatenated for phylogenetic analyses resulting in a data set with a total length of 2623 nucleotides (complete nuclear ITS2 and partial nuclear LSU rDNA comprising approx. 800 nt of the 5' terminus: 1083 positions; nuclear SSU rDNA: 1540 positions). The nucleomorph SSU rDNA data set comprised 1496 positions.

All nucleotide data sets were subjected to distance, maximum likelihood, maximum parsimony and Bayesian analyses. To determine the evolutionary model fitting best the data according to the Akaike Information Criterion (AIC), Modeltest 3.6 was used 42). Distance, maximum likelihood and maximum parsimony analyses were performed with the program PAUP* 4.0b10 43. Distance analyses were run under minimum evolution and set to the maximum likelihood parameters proposed by Modeltest. Data sets with heterogeneous base frequencies were also analyzed using the LogDet transformation. For both types of analyses, trees were inferred with the neighbor-joining algorithm. Maximum likelihood analyses were done using the proposed evolutionary model settings of Modeltest with three random addition replicates and heuristic tree search algorithm with tree bisection and reconnection (TBR). Unweighted maximum parsimony analyses were performed using 10 random addition replicates also in combination with the heuristic tree search algorithm. For all analyses under the distance or maximum parsimony criterion, 1000 bootstrap replicates were calculated; for maximum likelihood, 500 bootstrap replicates were computed. Bayesian analyses were performed using MrBayes 3.0B4 44. For the nucleotide data sets, likelihood settings were set to GTR, gamma-distributed among-site rate variation and covarion (includes proportion of invariable sites). Samples were drawn every 100^th generation for at least 3.5 million generations with one cold and three heated chains. Burn-in was determined for the individual data set according to the sump plot.

The protein data set was also subjected to distance, maximum likelihood, maximum parsimony and Bayesian analyses. The evolutionary model fitting best the data was determined with ProtTest 1.2.6 according to the AIC 4546 and used for maximum likelihood analysis with Phyml 2.4.4 47. Distance analysis was performed using protdist from the Phylip 3.62 package (set to JTT+Γ with global rearrangements; progam suite by Joe Felsenstein 48). The shape parameter α for the gamma distribution in protdist was calculated using Tree-Puzzle 5.2 49. Maximum parsimony analysis was done using PAUP* 4.0b10 (10 random addition sequence replicates). For Bayesian analyses with MrBayes 3.0b4, prior expectations were set to AAmodel=mixed (priors for all amino acid substitution matrices considered equal) and likelihood settings to gamma-distributed among-site rate variation and proportion of invariable sites. Samples were drawn every 100^th generation for 5 million generations using one cold and three heated chains. Burn-in was determined according to the sump plot.

The countcodon program from the web site of the Codon Usage Database 50 was used to determine absolute counts of all codons of the twenty-three rbcL sequences (396 codons). The DNA sequences were translated using the eubacterial/plastid codon table (code table 11).