Several eukaryotes have been previously analyzed for their selenoprotein content (selenoproteomes). These studies identified 24-25 selenoproteins in mammals and 0-4 selenoproteins in other organisms. It is generally thought that many eukaryotic selenoproteins evolved in vertebrates, but evolutionary paths have not been examined for the majority of these proteins. In this work, we analyzed the selenoproteomes of several additional model eukaryotes, whose genomes have been completed. These included marine algae (Ostreococcus tauri and O. lucimarinus), a diatom (Thalassiosira pseudonana), a soil amoeba (Dictyostelium discoideum), an insect (Drosophila pseudoobscura), and a red alga (Cyanidioschyzon merolae).

Selenoproteins are found in all three domains of life, which share several protein and RNA components involved in Sec biosynthesis and insertion, suggesting an origin of the Sec machinery that predates the last universal common ancestor. Thus, Sec decoding is an ancient trait that has been maintained for hundreds of million of years without widespread expansion or loss.

We compiled newly and previously characterized selenoproteomes and analyzed the occurrence of particular selenoproteins against taxonomic distribution of species based on the tree of life 33. The number of selenoproteins varied from zero (in plants, yeast and some protists) to 29 (in Ostreococcus) (Figure 6a). Significant differences in the composition of selenoproteomes could be seen even among related organisms. For example, among viridiplantae, all higher plants lacked selenoproteins, whereas the green algae Chlamydomonas and Ostreococcus had 12 and 26-29 selenoproteins, respectively (Figure 6b). Three selenoproteins were found in Mesostigma viride, a Streptophyte and a common ancestor of land plants 34.

Tracing individual selenoproteins, we found that some selenoprotein families were present in many organisms and others in only a few species, yet each identified family had a unique pattern of occurrence (Figure 6a). None of the selenoproteins matched the overall Sec trait (compared to the occurrence of Sec machinery). SelK was among the most widespread selenoproteins. This protein of unknown function is present in nearly all eukaryotes that utilize Sec (but is replaced with a Cys-containing homolog in nematodes and several other organisms). An additional widespread selenoprotein was SelW, which also occurs in most (but not all) selenoprotein-containing eukaryotes. Several other selenoproteins, such as glutathione peroxidase and thioredoxin reductase, also had a wide distribution.

Since mammalian selenoproteomes were large and included essentially all known eukaryotic selenoproteins, they were initially thought to represent the entire eukaryotic selenoproteome. Subsequent identification of selenoproteins with highly restricted occurrence added further complexity, but did not challenge the overall idea of recent evolution of the majority of eukaryotic selenoproteins. However, our analysis of selenoproteomes of six eukaryotic model organisms and their comparison with the previously characterized selenoproteomes revealed that 20 of the 25 human selenoproteins have Sec-containing homologs in many unicellular organisms. Similarly, taking into account protein families, at least 11 of the 16 mammalian selenoprotein families could be traced back to single-cell eukaryotes. SelU, which is not a selenoprotein in mammals, is present in some animals and protozoa and may be viewed as an additional ancient selenoprotein family. Overall, these data suggest that the origin of many selenoproteins not only precedes animal evolution, but can be dated back to the ancestral eukaryotes. Thus, many of these original selenoproteins were preserved during evolution and remain in vertebrates (including mammals), green algae and a variety of protists, whereas many other organisms manifested massive selenoprotein losses.

It should be noted that Cys/Sec replacement is not always unidirectional and that prior evolutionary analyses suggest that both a Sec loss and gain is possible 35. However, the probability of independent parallel Sec gain, as well as consecutive homoplastic Sec-to-Cys and Cys-to-Sec substitutions in a single protein position, is extremely rare, and no selenoprotein families are known that evolved more than once. Two factors are required for a Cys-to-Sec change to take place. First, the presence of Sec insertion machinery, such as Sec tRNA, SECIS-binding protein SBP2, Sec-specific elongation factor and Sec synthase. This requirement is met (for example, all components of the machinery are present) if at least one other selenoprotein is present in the same organism. Second, a SECIS element should evolve in the 3'-untranslated region. While only a single nucleotide change is sufficient to change the codon from Cys to Sec (that is, UGA instead of UGC or UGU), evolution of new SECIS elements is difficult. On the other hand, once Sec is replaced with Cys, the presence of the SECIS element provides no competitive advantage and this structure is quickly lost. Unless the reverse Cys-to-Sec mutation takes place before disruption of the SECIS element, the probability of restoring Sec is extremely low. Unless strong pressure exists to preserve Sec, its functional replacement with Cys may be expected. Combined, these factors allow us to assume that the character-state Sec follows Dollo's behavior.

In addition to the many ancient eukaryotic selenoproteins, several selenoproteins have a more narrow distribution. For example, SelP, SelN, MsrB and SelI appear to be specific to animals, whereas MSP, peroxiredoxin and thioredoxin-like protein could be detected only in unicellular eukaryotes. These observations suggest an emerging picture of selenoprotein evolution wherein core selenoprotein families evolved first, followed by the origin of additional selenoproteins in more narrow groups of organisms. The new selenoproteins further increased the size of the selenoproteomes and remain prevalent in organisms with large selenoproteomes. In our current analysis, several Ostreococcus and Thalassiosira selenoproteins fit this pattern, in addition to the rare selenoproteins previously discovered (for example, SelU, SelJ and Fep15). However, it could not be excluded that new selenoproteins might also occasionally evolve in organisms with small selenoproteomes (for example, red algae).

We further identified and examined several groups of organisms characterized by massive selenoprotein loss. Location of these organisms on the eukaryotic tree of life suggests independent events of selenoprotein loss (Figure 6a). Five examples of selenoprotein loss are discussed below.

The mosaic occurrence of eukaryotic selenoproteins and their consistent loss in different phyla suggest that the decreased selenoproteome size is the result of a selective force. What could be the factors responsible for or associated with selenoprotein loss? Comparative analysis of organisms with large and small selenoproteomes shows that many of the selenoprotein-rich organisms live in aquatic environments. In contrast, almost all organisms that lack or have a small number of selenoproteins are terrestrial (Figure 6). Considering independent, large-scale selenoprotein loss in these organisms, a common denominator appears to be the non-aquatic habitat. It should be noted, however, that the differences between aquatic and terrestrial selenoproteomes are ultimately influenced by specific environmental factors that differ with habitat. Therefore, the aquatic/terrestrial association should not be viewed as the basis for selenoprotein loss/gain, but rather a convenient illustration of differences between these organisms. Once environmental factors are identified, this association may be modified to reflect these factors rather than habitat.

To further examine selenoprotein content of aquatic and terrestrial organisms, we analyzed organisms that are well represented by ESTs. We excluded large animals (vertebrates) from this analysis because their intra-organismal environment would be less affected by environmental conditions due to availability of their outside protective cover and complex morphology. With this limitation, aquatic eukaryotes had more selenoprotein genes than terrestrial organisms (Figure 7).

Whether C. merolae fits this association is not clear. This organism lives in highly acidic sulfate-rich hot springs (pH 1.5, 45°C). It is possible that this extreme environment is responsible for the reduced use of Sec in red algae. The pKa of Sec is approximately 5.5. Whereas this residue would be ionized in most organisms under physiological conditions, at low pH, protonation of Sec may minimize its catalytic advantages. Abundance of sulfate in hot springs might also be of importance, as selenium and sulfur have similar chemistries.

One possible explanation for the occurrence of large selenoproteomes in aquatic organisms is bioavailability of selenium in oceans. Dissolved organic selenides can account for approximately 80% of the dissolved selenium in ocean water 37 and represent an important source of selenium for phytoplankton. Following the food chain, this could explain a large number of selenoproteins in algae and fish. Likewise, a considerable number of selenoproteins in mammals could reflect the consequence of food sources, body size and relatively recent (in evolutionary terms) emergence of these organisms from marine environments. An additional factor may be constancy in the environmental conditions and nutrients in the aquatic environments. For aquatic organisms, environmental changes are slower and involve gradients of temperature, pH, pressure, oxygen and chemical environment. In contrast, in terrestrial environments, the changes are more frequent and they happen more suddenly. As a result, terrestrial organisms often face feast and starvation situations. An attractive factor to explain the differences between aquatic and terrestrial selenoproteomes may be oxygen content. Higher content of oxygen in air than in aquatic environments may make highly reactive selenoproteins more susceptible to oxidation in terrestrial organisms and select against the use of these proteins.

Whether mammals and other vertebrates fit the hypothesis on the preferential use of selenium in aquatic environments is not clear. We note, however, that fish have larger selenoproteomes than those living in terrestrial environments, including mammals, reptiles and birds. Further genomic analyses of these organisms could clarify evolutionary changes in utilization of selenium. In future studies, it would also be important to determine which of the factors discussed above influence the preferential use of Sec in aquatic organisms or are responsible for the loss of selenoproteins in terrestrial organisms.

All genome, EST and predicted protein sequences were downloaded from NCBI 38, except for the genomes of T. pseudonana, O. tauri, and O. lucimarinus, which were obtained from Joint Genome Institute 39. SECISearch 9 was used for identification of SECIS elements. FASTA package 40 and BLAST were used for similarity searches. MFOLD version 3.2 41 was used for prediction of RNA secondary structures.

Query sequences included a full set of human selenoproteins 9 as well as the following selenoproteins absent in mammals: Chlamydomonas MsrA 12, Gallus gallus SelU 42, Danio rerio SelJ and Fep15 1516, and Emiliania huxleyi protein disulfide isomerase 43. A stand-alone version of TBLASTN program was utilized for detection of nucleotide sequences corresponding to known selenoprotein families. A candidate Sec residue should correspond to a Sec residue in a known selenoprotein family or a Cys residue in orthologous proteins in order to be considered further. Downstream regions of predicted selenoprotein sequences were analyzed for the presence of candidate SECIS elements using SECISearch and for SECIS-like structures using MFOLD 41. All detected SECIS candidates were further examined for compliance with the current SECIS consensus model.

Nucleotide sequences were scanned using SECISearch (Additional data file 1). In addition, the default and loose patterns of SECISearch were modified as described elsewhere 12 to accommodate organism-specific selenoprotein searches. These modifications allowed increased sensitivity of SECISearch and supported identification of unusual SECIS structures. The overall strategy of the searches was similar to that previously described 9. Statistics of the searches (numbers of candidates corresponding to different steps in the search process) are shown in Table 1. In an additional search for D. discoideum SECIS elements, the following pattern was used as a query: TGTAATGATT_(10-12 nucleotides)_AAA_(24-35 nucleotides)_TGAT. This search then continued as described for other organisms.

The primary sequence analysis step included searches for SECIS-like structures that satisfy NTGA__AA__GA or NTGA__CC__GA (N is any nucleotide) motifs in nucleotide sequences. Additional requirements were that the distance between the quartet (NTGA) and the unpaired AA in the apical loop is 10-13 nucleotides, and the distance between the unpaired AA and the GA that base-paired with the quartet is 15-39 nucleotides.

The secondary structure analysis step examined for consistency with the eukaryotic SECIS element consensus. Several additional filters were implemented to filter out candidates with unsuitable secondary structures, including SECIS elements with more than two unpaired nucleotides in a row and Y-shaped SECIS elements.

The free energy for each candidate structure was estimated; the free energies for the whole structure (threshold value of -12.6 kcal/mol) and the upper stem-loop (threshold value of -3.7 kcal/mol) were calculated. Only thermodynamically stable structures were considered further.

Based on the location of candidate SECIS elements, candidate ORFs were predicted in upstream regions. SECIS candidates located within coding regions of known proteins were filtered out. An additional requirement was the presence of at least one homologous protein in the NCBI non-redundant database. If SECIS elements and ORFs corresponding to known protein families were on different DNA strands, the candidates were filtered out.

The final step included manual sequence analyses of predicted selenoprotein ORFs located upstream of candidate SECIS elements.

For three organisms, O. tauri, O. lucimarinus and C. merolae, additional procedures for selenoprotein detection included the search for Sec/Cys pairs in homologous sequences. ORFs with in-frame TGA codons were extracted that satisfied the following criteria: Sec-flanking regions for these proteins were conserved; and homologs could be detected that contained Cys in place of Sec. TBLASTX was used to examine all potential ORFs with in-frame UGA codons against NCBI non-redundant protein database. All hits were then tested for the occurrence of SECIS elements. Orthologous proteins were defined as bidirectional best hits. PSI-BLAST was used for identification of distant homologs. Homologs were further confirmed by phylogenetic trees construction.

Numerous attempts to derive a tree of life using various methods that were based on genes encoding ribosomal RNAs and several proteins have been published. However, their principle existence has been questioned recently because of either an insufficient amount of discriminating characters or other biases such as horizontal gene transfer and chimerism. To avoid such problems, we adopted a eukaryotic branch of a phylogenetic tree recently developed by Ciccarelli et al. 33. This highly resolved tree of life utilized 31 concatenated, universally occurring genes with indisputable orthology in 191 species with completed genomes across all three domains of life. The missing organisms were filled in using a 'Tree of Life' web project 44 and selected publications 5-48]. Although the horizontal gene transfer is highly prevalent in prokaryotes, it is less so in eukaryotes, particularly in multicellular organisms. We also analyzed selenoprotein evolution in the eukaryotic domain. To reconstruct the phylogenies of selenoproteins, we adopted a character-based tree estimation method, a maximum parsimony approach that implies that the preferred phylogenetic tree is the tree that requires the least number of evolutionary changes.

D. discoideum cells were grown as previously described 31, the medium was supplemented with 100 μCi of ⁷⁵Se [selenite] (University of Missouri Research Reactor), and the cells were further maintained under continuous shaking for two days. A similar procedure was used for labeling O. tauri cells, except that they were grown in K-medium. The radioactive bands were visualized on the gel with a PhosphorImager. Samples of ⁷⁵Se-labeled mammalian HEK 293 and CV-1 cells were included, which were prepared as described previously 49.