To predict SRP RNA genes from protozoans and fungi we used a method previously described 14. The first step is a heuristic pattern-based search for conserved features of the helix 8 region as described under "Methods". The pattern was relatively degenerate and in many cases the result included a number of false positives. In a second step the candidates from the pattern-based search were screened with a COVE model of SRP RNA. COVE is an implementation of the algorithms described by Eddy & Durbin 15 that make use of probabilistic models to describe the sequence and secondary structure consensus of an RNA family. The COVE analysis is much more stringent than the first pattern-based step and we expect very few false positives among the high-scoring hits from this analysis.

The identified SRP RNA candidates aligned well to a COVE model for eukaryotic RNAs in the conserved S domain region, i.e. the part that corresponded to the helices 5, 6 and 8. The Alu domain displayed a higher degree of variation and in many instances did not align well to the COVE model. As a consequence, the prediction of the 5' and 3' ends of the molecule were in most cases unreliable.

The secondary structure of all candidates were also predicted with MFOLD 16 with default parameters or specifying constraints consistent with known conserved elements of SRP RNA. MFOLD was able to fold the S domain of the SRP RNA in a manner which was consistent with the secondary structure predicted by COVE. However, when used without constraints, MFOLD typically predicted a secondary structure for the Alu domain that was inconsistent with our identifications. As a consequence, for the prediction of Alu domains as well as their folding, we relied on consensus features, such as the presence of a conserved sequence motif UGUNR (where N is any base and R is purine, typically an A) motif and the general secondary structure outline (Fig. 1). In summary, in our prediction and folding of SRP RNA we combined pattern matching, COVE, and MFOLD, and we checked that known consensus motifs of SRP RNAs were present in the predicted RNAs. Finally, we used BLAST to show that the predicted SRP RNA genes did not overlap with predicted protein-coding regions or any other annotated features. Therefore, we believe that the final candidates presented here represent sequences that are evolutionary related to SRP RNA. Still, it should be noted that we cannot distinguish between a bona fide SRP RNA gene and pseudogenes that are known to occur in plants 817 and in mammals. Examples are the two SRP RNA gene candidates that we identified in the C. reinhardtii genome. The covariance models did not allow us to predict the 5' and 3' ends and the folding of the Alu domain of these two sequences. Therefore, it remains to be seen which of these candidates, if any, represents a functional RNA.

We were not able to identify an SRP RNA in Giardia lamblia and Entamoeba histolytica. However, as SRP proteins were identified in these organisms we expect that, as the genomes are completed, SRP RNA genes will be discovered.

An SRP RNA gene was identified in Entosiphon sulcatum (Fig. 3). Its Alu domain was found to contain a very short helix 4 and in this respect resembled the structure of the Alu domain of the trypanosomatids. This relationship was consistent with the known close evolutionary relationship between euglenids and trypanosomatids 18). Interestingly, the E. sulcatum SRP RNA gene was shown to be part of a cluster which also contained genes for 5S rRNA, U1, U2 and U5 snRNAs 1920). This gene organization is reminiscent of that of T. brucei and Leishmania where SRP RNA genes were found to be located adjacent to other RNA genes 2122.

In the group of the Alveolates we found an Eimeria tenella SRP RNA (Fig. 3) with a predicted Alu domain structure similar to that of the metazoans. Interestingly, the Alu domain of Theileria annulata was reminiscent of the SRP RNA Alu domain previously identified in the Ciliophora Tetrahymena 23, in the respect that the helix 4 appeared to be absent.

It has previously been reported that the genome of the malaria parasite Plasmodium falciparum encodes several SRP proteins 24. Here, we were able to identify the corresponding SRP RNA (Fig. 3). The secondary structure of the Alu domain of this RNA was predicted by combining COVE and MFOLD procedures. In addition, the RNA of two other Plasmodium species, P. yoelii and P. knowlesi, were predicted to form the same structure despite significant differences in their primary sequences (Fig. 3). The Alu domain of Plasmodium SRP RNA was different in that it possessed an internal loop in helix 4. Therefore, even within the Alveolates, a considerable variation in the predicted folding of the Alu domain was observed.

We previously identified SRP RNA genes in C. albicans and N. crassa 14. Here we also found an SRP RNA candidate in Aspergillus nidulans. As shown for Yarrowia SRP RNA in Fig. 4, in all of these fungi, including S. pombe, the SRP RNA secondary structure were shown to be very similar. For the unusually large (519 nts) SRP RNA of S. cerevisiae12, the COVE model predicted an S domain with helices 5, 6 and 8 as for other eukaryotic RNAs. MFOLD also folded this part of the molecule in accordance with the consensus 2D structure of the S domain. As the 5' and 3' terminal sequences were shown to be related to Alu, we concluded that the S. cerevisiae SRP RNA contained at least one insert as compared to other fungi. Secondary structures for the SRP RNAs including these inserts were constructed for S. mikatae, S. kudriavzevii, S. bayanus, S. castellii and S. kluyveri. The sequences were identified by BLAST using the S. cerevisiae sequence as query. For the prediction of the 5' end of the RNA we took advantage of the fact that the highly conserved Alu domain was present at the very 5' end of the RNA. For prediction of the 3' end we considered a T-rich region which was conserved in all six Saccharomyces strains and likely is part of a transcription termination signal.

Since the number of available Saccharomyces SRP RNA sequences was too low for using covariation or mutual information analysis, and a COVE model that would predict the pairing in the insert regions could not be obtained, MFOLD was used first with each full-length sequence to predict the secondary structure. As expected, all Saccharomyces sequences folded into structures which contained helices 5, 6 and 8 as predicted for S. cerevisiae. Furthermore, all Saccharomyces RNAs possessed a hairpin structure with the Alu UGUNR motif at the 5' end similar to what was observed in the Alu domains of the other fungi. A multiple alignment was obtained using procedures described under Methods and is available in the web supplement to this paper http://bio.lundberg.gu.se/srp03/.

As for the SRP RNA insertions specific to Saccharomyces, MFOLD predicted the structure shown in Fig. 4 containing the helices that we here refer to as 5c-g (Fig. 4). The helices 5h-i were also characteristic of the Saccharomyces RNAs. Smaller corresponding inserts reminiscent of these were found in Yarrowia, Neurospora and Aspergillus. The predicted secondary structure of the 5c-g region was very similar in all Saccharomyces species although there was significant variation in sequence. The bases involved in compensatory base changes in this part of the RNA (Fig. 4) offer support to the predicted folding.

The 5c-g insertion appeared to be specific to Saccharomyces and indicated that at some point during evolution this piece of the RNA was inserted near the 5' end as indicated in Fig. 4. The evolution of Saccharomyces also involved the enlargement of helices 5h and 5i. One may speculate that these species have developed some additional mechanisms related to SRP-mediated translational arrest or additional unknown functions.

Our findings suggested that all fungal SRP RNAs contain a hairpin structure with a Alu UGUNR motif at the 5' end. This part of the RNA was found to be conserved and there was phylogenetic support for the hairpin structure from the covariations indicated in Fig. 4. The fungal hairpin motif was distinct from all other known non-fungal Alu domains where the conserved UGUNR motif was found to be part of a more elaborate pseudoknot.

It has been shown previously that a 5' terminal 99 nt fragment of the S. cerevisiae RNA was able to bind in vitro to the SRP14 protein. A tentative model of this portion of the SRP RNA has been presented where only the 5' terminal portion formed the Alu domain 25. However, based on the analysis presented here it is likely that also the 3' portion is part of the Alu domain.

There is strong evidence that the Microsporidia, such as E. cuniculi, are phylogenetically related to Fungi 13. An analysis of the E. cuniculi genome revealed a candidate SRP RNA in a 396 nt intergenic region (chromosome X, positions 138833–139226) which contained helices 6 and 8, and aligned well with our eukaryotic COVE model. However, we were unable to identify a typical metazoan Alu domain. Fig. 3 shows a tentative model of the Alu domain which is similar to that of other fungi.

We used a range of tools to identify and inventory SRP proteins in protozoa and fungi. Genome sequences were analyzed for genes encoding SRP proteins using BLAST 26 or FASTA 27 using previously known eukaryotic SRP proteins as query sequences. In addition, we performed PSI-BLAST 26 searches where a SRP protein sequence, typically the human ortholog, was used to search a database with the proteins in a public protein sequence database combined with the proteins obtained by translating all possible open reading frames of the genome being analyzed. Genomes were also analyzed using Genscan 28 or GlimmerM 29 and predicted peptide sequences were used in a BLAST or PSI-BLAST procedure as above. The results of our findings are shown in Table 1. One should keep in mind that several genomes were incomplete although large portions were covered by raw sequence data or contigs. The fact that we failed to identify a certain component of SRP was therefore not entirely conclusive in these cases (indicated by empty cells in Table 1).

As previously noted for other species, SRP54 was shown to be ubiquitous and highly conserved. Also SRP19, SRP68 and SRP72 were found in most of the genomes analyzed here (Table 1) although we were unable to identify a SRP72 homolog among the Alveolates. In E. cuniculi we failed to identify SRP68/72, as described further below. We obtained evidence that the SRP9/14 proteins were subject to a more rapid evolution, as discussed in the following.

SRP9 and 14 homologs were identified in Plasmodium falciparum and Chlamydomonas reinhardtii. The SRP14 homolog in P. falciparum has been identified previously but annotated as a hypothetical protein (http://www.plasmodb.org, accession PFL0160w). SRP14 was found also in E. tenella and E. histolytica. However, we were unable to identify SRP9/14 in Leishmania major, Trypanosoma cruzi, Theileria annulata, Giardia lamblia and E. cuniculi. Although an intriguing observation, we could not formally rule out the possibility that SRP9/14 homologs would be discovered during the completion of the genome assemblies and that the SRP9/14 protein sequences in these organisms have strongly diverged from known members of this protein family.

Evidence has been provided that Leptomonas and T. brucei possess a tRNA-like RNA that associates with the SRP 3031. It has been speculated that this RNA compensates for the loss of portions of the Alu domain 31. The possibility that the tRNA-like RNA took the role of proteins SRP9/14 was considered as well. Upon completion of additional trypanosomatid genome sequences it will be interesting to determine if SRP in all these organisms carry a tRNA-like component and if they all lack SRP9/14.

The microsporidian E. cuniculi has a highly compact genome that seem to have been under a pressure to eliminate non-essential material. As SRP9/14, 68 and 72 seem to be missing, the evolution of this organism could have involved the loss of these genes. The lack of these proteins would suggest that they are less critical for SRP function. It is interesting to note that in this respect E. cuniculi resembles archaea which also appear to lack SRP9/14, 68 and 72.

Homologs of SRP14, SRP19, SRP54, SRP68, and SRP72 were identified in all the Saccharomyces species (Table 1). It has been reported previously that S. cerevisiae possess SRP21 which was thought to be a new family of SRP proteins 11 unique to Saccharomyces. We have here reexamined the relationship of SRP21 to other proteins, including those of the SRP. Homologs to S. cerevisiae SRP21 in S. mikatae, S. kudriavzevii, S. bayanus, S. castellii, S. kluyveri and S. paradoxus sequences were initially identified using BLAST. The E-values ranged from 1e-77 (S. paradoxus) to 1e-29 (S. kluyveri). A homolog to SRP21 was found also in Candida albicans (E-value 8e-06). Open reading frames in the region of the BLAST hit were identified to obtain a likely full-length protein sequence for the C. albicans SRP21 (Fig. 5). We also identified the S. pombe protein YE07_SCHPO (T37873) as a distantly related homolog, with a E-value of 6.3. Interestingly, this protein displayed a distant homology to metazoan protein SRP9 and was previously listed in the SRPDB as a potential SRP9 homolog. Using the S. pombe protein as query we identified a possible homolog in Neurospora crassa (E-value 5e-6) using TBLASTN to search N. crassa genomic sequences (Fig. 5).

We considered that SRP21 might bind to the regions inserted specifically into Saccharomyces SRP RNA, i.e helices 5c-g or 5h-i. However, this possibility appeared unlikely because SRP21 homologs were identified also in those fungi that did not possess these helix 5 expansions.

To further examine the SRP21 homologs and their relationship with SRP9 we carried out profile-based searches. First, all novel putative yeast SRP21 homologs identified here were merged with the sequences of public protein sequence databases (Genpept and SWISSPROT/TREMBL). The resulting databases were used in profile-based searches including PSI-BLAST, PROFILESEARCH, and hmmsearch. For instance, when the S. cerevisiae SRP21 sequence was used as query in a PSI-BLAST search, after two iterations the S. pombe protein YE07_SCHPO (T37873) was identified above the default threshold (E-value 0.01) together with the Candida and Saccharomyces SRP21 proteins (not shown). Second, PROFILESEARCH was used with a profile based on a multiple alignment of the Saccharomyces sequences (Fig. 6, sequences shown in lighter gray) obtained by CLUSTALW to search SWISSPROT/TREMBL (approximately 1 million protein sequences as of March 2003). In the result of this search the Candida, Neurospora, and S. pombe sequences followed immediately after the Saccharomyces SRP21 sequences (Fig. 6). Interestingly, the SRP9 proteins of maize, Arabidopsis, and C. elegans ranked closely to the top-scoring yeast sequences, although they had scores lower than the other SRP21-related proteins. Similar results were obtained with hmmsearch (not shown).

A multiple alignment of SRP21 and SRP9 as well as SRP14 protein sequences is shown in Fig. 5. The alignment of SRP9 to SRP14 is the structural alignment of Birse et al. 32 and the alignment of SRP9 to SRP21 was the result of a CLUSTALW analysis which was consistent with the results obtained from the profile searches described above. Interestingly, many of the positions that were conserved in the SRP9/14 structural alignment 32 were occupied by the same category of amino acids in the SRP21 proteins (Fig. 5). These data indicated that SRP21 is structurally similar to the SRP9/14 proteins and provided further evidence of the homology between SRP21 and SRP9.

In mammalian SRP, proteins SRP9 and SRP14 were shown to form a heterodimer and share a αβββα topology 32. To determine if the secondary structure predicted for fungal SRP21 was similar to the SRP9 and SRP14 structure, we made predictions with PSIPRED 3334. Saccharomyces SRP21 sequences were used as input and the results are shown in Fig. 7. For human SRP9 and SRP14 the boxed regions indicate the positions of the secondary structure elements as known from the structure of the proteins. The corresponding regions for the fungal proteins are also shown in boxes and are based on the alignment in Fig. 5. The results showed that the predicted secondary structure of SRP21 was remarkably similar to the predicted or known structures of SRP9 and SRP14. With the exception of the first α-helix of Saccharomyces SRP21 all the predicted secondary structure elements were consistent with those of the SRP9/14 structure. Similar results were obtained with the secondary structure prediction method SAMT02 35 (not shown).

In metazoan cells, SRP9/14 forms a complex with the Alu domain of SRP RNA. The structure of this complex has been determined at high resolution 532. On the other hand, the Alu domain of yeast is less well characterized. An analysis of yeast SRP showed that it was missing an obvious SRP9 homolog 11. SRP21 was not believed to be the SRP9 equivalent in yeast because its sequence similarity to SRP9 was not recognized and SRP21 did not appear to be stably associated with SRP14. Furthermore, evidence was provided that SRP14 is present in two copies in the yeast SRP 3 and SRP14 was shown to form a homodimer which bound to the Alu domain 25. On the basis of these observations it was assumed that the SRP14 homodimer was functionally equivalent to the SRP9/14 heterodimer. In contrast, we suggest that SRP21 is not only structurally but also functionally related to SRP9. We suggest that the protein is an integral component not only of Saccharomyces, but of every yeast SRP. In the light of these findings it will be important to reexamine experimentally the role of SRP21 in yeast.

Some of the genomic sequences used in this work were unfinished sequences and represent as yet unpublished material. In these cases permission to present the results in this paper was obtained from the respective research groups. SRP RNA sequences from S. mikatae, S. kudriavzevii, S. bayanus, S. castellii and S. kluyveri were retrieved using BLAST searches against the corresponding genomes using the S. cerevisiae sequence as query using the BLAST server at the Genome Sequencing Center at Washington University http://www.genome.wustl.edu/blast/yeast_client.cgi. Saccharomyces protein sequences were identified using the same BLAST server and the Synteny viewer at the Saccharomyces Genome Database http://www.yeastgenome.org. Sequences of Candida albicans were from the Stanford Genome Technology Center website at http://www-sequence.stanford.edu/group/candida (Assembly 6). Neurospora sequences were from the Neurospora Sequencing Project, Whitehead Institute/MIT Center for Genome Research http://www-genome.wi.mit.edu. The dataset used in these studies was http://www.broad.mit.edu/ftp/pub/annotation/neurospora/assembly3/neurospora_3.fasta.gz. Aspergillus nidulans sequences were downloaded from the Whitehead Institute, Center for Genome Research at http://www-genome.wi.mit.edu/cgi-bin/annotation/aspergillus/download_license.cgi. Plasmodium falciparum sequences were from PlasmoDB at http://www.plasmodb.org.

Trypanosoma cruzi and Entamoeba histolytica were downloaded with permission from TIGR.

Eimeria tenella http://www.sanger.ac.uk/Projects/E_tenella/,

Theileria annulata http://www.sanger.ac.uk/Projects/T_annulata/,

Leishmania major http://www.sanger.ac.uk/Projects/L_major/ were from the Sanger Centre.

Other sources were Chlamydomonas reinhardtii http://genome.jgi-psf.org/chlre1/chlre1.home.html,

Giardia lamblia http://jbpc.mbl.edu/Giardia-HTML/index2.html

and Encephalitozoon cuniculi

ftp://ftp.ncbi.nih.gov/genomes/Encephalitozoon_cuniculi/.

SRP RNA genes were predicted as described previously 14 by applying a combination of pattern searches using rnabob http://www.genetics.wustl.edu/eddy/software/#rnabob and COVE 15. The rnabob searches made use of descriptors based on consensus features of the helix 8 region of the RNA. The search pattern was typically (XX) YYAGR (NNN) GRRA (N'N'N') AGCAR (X'X') or minor variations of it, where X pairs with X' and N with N'. RNA secondary structure predictions were carried out by COVE and MFOLD 16 and were complemented by analysis of compensatory base changes 36. Saccharomyces RNAs were first aligned with CLUSTALW 37 using a gap extension penalty of 0, and this alignment was manually edited to be consistent with observations from MFOLD predictions to assure that bases involved in the formation of helices were properly aligned.

The program 'sixpack' of the EMBOSS package 38 was used to obtain all possible translation products of genome sequences. PROFILESEARCH was part of the GCG package (Wisconsin package version 10.2, Genetics Computer Group (GCG), Madison, Wisc.). Hmmer package, developed by S. Eddy, was obtained from http://hmmer.wustl.edu/. Genscan was downloaded from http://genes.mit.edu/GENSCANinfo.html and GlimmerM from http://www.tigr.org/software/glimmerm/. Protein secondary structure was predicted using PSIPRED 3334 and SAM-T02 35.

All predicted RNA and protein sequences, secondary structures and multiple alignments are shown in a web supplement at http://bio.lundberg.gu.se/srp03/.