The sequencing of an EST collection of the juvenile-staged chætognath S. cephaloptera offered the opportunity to explore the transcriptome of this evolutionarily significant organism. The survey of sequence length and quality supported the accuracy of these data (Figure S1 in Additional data file 1). During these steps, we noticed that 16% of sequences match mitochondrial rRNA sequences (12S and 16S rRNAs, Figure 1) probably because the long polyadenine stretches of these rRNA molecules were isolated by the oligos-dT employed for mRNA isolation (see Materials and methods). We attempted to build clusters that gathered all transcripts from a unique gene so as to deal with a non-redundant partial transcriptome. However, the low complexity regions of some ESTs, which did not include an accurate open reading frame, hindered this process. Thus, ESTs were sorted into predicted coding and non-coding sequences using conceptual translation, and the coding transcripts were retained for comparative analyses. The overall content of the EST collection was evaluated using these steps (Figure 1). We noticed that up to 54% of the ESTs could be non-coding polyadenylated RNA, a striking figure that is, however, similar to that obtained for the human genome 27. The removal of non-coding sequences greatly improved clustering efficiency, yielding 1,447 clusters, of which 459 include more than one sequence (Figure S1 in Additional data file 1). A total of 694 of these clusters have significant matches within a protein database (TrEMBL, score >50) and 250 have clear homologs in this database with an average of 72% identity (score >150). Among the transcripts that match nuclear coding genes, the RP genes are largely represented compared to other genes similar to SwissProt entries (Figure 1).

The average gene content of the library was checked regarding functional annotation as implemented in Gene Ontology 28. The S. cephaloptera library exhibited a broad diversity of functional classes with a majority of transcripts involved in metabolism or cellular activities and a non-negligible amount of transcripts involved in development (Figure S2 in Additional data file 1), which is consistent with the juvenile stage of the animals used. Hence, this EST collection contains representative, high quality sequences, providing suitable material for comparative analyses.

The set of non-redundant chætognath transcripts was compared with several databases using the Blast program. These databases first included sets of transcripts of representative species belonging to the most important clades of bilaterians: Drosophila melanogaster as an ecdysozoan, Lumbricus terrestris as a lophotrochozoan and Homo sapiens as a deuterostome. These comparisons were depicted through the plotting of respective similarity scores for all transcripts that have a significant match to at least one of these species (score >150, Figure 2). This comparison demonstrated that a pool of 141 transcripts is strongly conserved between these distantly related species (Figure 2a). Conversely, 169 transcripts did not have significant matches in one or two of the species despite their strong similarity between chætognath and the remaining species. This lack of homologs is generally imputed to extensive gene loss 21. Therefore, further comparisons were performed to identify genes whose homology assignment and gene loss in a peculiar lineage were unambiguous. Interestingly, the number of transcripts that did not match to one or more databases decreased from 169 to 74 when the complete set of sequences available for each bilaterian clade was employed as the database, instead of only one representative species (Figure 2b). The lack of homologous matches in some species could then be explained by an increase in evolutionary rates, which could have weakened the sequence similarity signal 29. Additionally, the similarity level of matches increased when composite databases were employed (Figure 2), which supports the interest in this approach for phylogenomic reconstruction 18.

Two classes of genes provide reliable information for phylogeny inference (Figure 2b). Those that are highly shared between distantly related taxa constitute a set of conserved genes that are valuable markers for constructing phylogenomic datasets. In parallel, the genes that lack a homologous copy in one of the considered clades represent meaningful signature genes whose loss is attributable to a discrete event 23.

The candidates for signature genes are the genes inferred to be lost in one of the investigated clades (Figure 2b). Those candidates were carefully examined and their presence checked in the largest sets of available ESTs and full genome sequences. These data include the newly sequenced full genomes of lophotrochozoans and is assumed to include an exhaustive gene set in these species. Numerous candidate genes were invalidated because their homology relationships are disputable or because a homolog was retrieved from the full genome sequences surveyed. Among these candidates, the guanidinoacetate N-methyltransferase (GAMT) enzyme was recovered in the chætognath S. cephaloptera, in all studied deuterostomes, cnidarians and sister groups of metazoans (Figure S3 in Additional data file 1) but was not retrieved in any of the protostomes surveyed. Notably, this GAMT enzyme was also recovered in the acoel Convoluta pulchra, which was recently excluded from the protostomes 30. This enzyme catalyzes the key step of creatine synthesis, an activity that was previously checked biochemically in a variety of organisms but was not found in selected protostomes 31. GAMT was later noticed as missing in D. melanogaster, Anopheles gambiae and Caeorhabditis elegans genomes 32. The presence of this ancient gene provides strong evidence for an early divergence of chætognaths from other protostomes. Indeed, the most parsimonious scenario states that this gene was lost in the protostome lineage after its split with chætognaths 18.

We attempted to evaluate the phylogenetic properties of the conserved genes that share equal levels of similarity with the main animal clades with respect to the convenience of their orthology assignment, their abundance in EST data and their molecular evolution properties. The main concerns when constructing phylogenomic-class datasets, especially from EST data, are the discarding of paralogous sequences, the removal of contaminants and the limitation of missing data. According to these criteria, we argue here that the set of RP genes is one of the best for setting up phylogenomic analysis in a large sample of taxa.

Among the 694 chætognath genes similar to a database entry, only 267 genes have homologs in the three main clades of bilaterians (score >150, Figure 2b). Copies of each selected marker were retrieved for all phyla studied for which EST data are available (Figure 3). In this way, the missing data were estimated through the occurrence of each gene in EST collections and preliminary phylogenetic analyses were carried out for all these independent alignments. Such controls unexpectedly highlighted putative paralogy problems for many candidate markers. If the orthologous transcript of a surveyed gene is missing in a non-exhaustive EST collection, a paralogous relative of this gene could be retrieved instead, with little chance of detection. Among candidate marker genes, RPs exhibit no ancient duplicates or out-paralogs and constitute a class of markers free from potential paralogy assignment problems 2533. Moreover, the gene-specific trees allowed detection of some contaminants in the EST collections, through the verification of unexpected clusterings in the tree (for example, several EST collections of parasitic organisms being contaminated by transcripts from their hosts).

Next, the amount of missing data was estimated using these raw alignments and compared with the number of ESTs in each available collection (Figure 3). The positive correlation observed between the number of ESTs and the completeness of the dataset is stronger when dealing with a dataset composed of RPs. For instance, the 5,235 EST collection of tardigrades yielded a dataset that is 77% complete for RPs, but only 35% complete for non-ribosomal markers. Thus, their large representation in EST collections strengthens the usefulness of RPs as phylogenetic markers.

In order to assess the branching of chætognaths and to stress the usefulness of RP genes for phylogenomics, a RP dataset was assembled using the composite dataset approach 18. This method depends on the selection of the least diverging copy of each marker gene in each taxon, such as a phylum, and thus allows reduction of the branch lengths of composite taxa (Table S1 in Additional data file 2). To overcome previous problems, both taxon sampling and inference methods were improved. Several new phyla were included in this analysis and, in particular, numerous protostome groups: priapulids, platyhelminthes, nermerteans, ectoprocts, entoprocts and rotifers 343536. Most rotifer sequences were retrieved from Oryza sativa (rice) ESTs, where they exist as contaminants, using their very specific splice-leader sequence as an anchor (see below and 37). Rotifers constitute a key phylum with respect to chætognaths because they were sometimes grouped together in the gnathifera clade on the basis of morphological criteria 38. Alternatively, a splitting of lophotrochozoans into two main lineages, the platyzoans (uniting platyhelminthes and rotifers) and the trochozoans (mainly annelids, molluscs, lophophorates and nermertes) has been proposed 3940. Otherwise, in addition to the traditional site-homogenous WAG model, we have assessed the phylogeny of bilaterians using the site-heterogeneous CAT model, which recently improved the limitation of the long-branch attraction artifact, a common pitfall in phylogenetic reconstruction 4142. The inclusion of the most recently released EST data for this large set of phyla led to a dataset including 11,730 amino acid positions and 25 taxa (Additional data file 4).

The analysis of this dataset confirmed the branching of chætognaths at the base of the protostomes with significant support values for both the site-homogeneous WAG model and the site-heterogeneous CAT model (bootstrap proportion (BP) of 76 and posterior probability (PP) of 1; Figure 4a,b). The inclusion of chætognaths within protostomes is still firmly supported (BP 95, PP 1; Figure 4). The inclusion of new taxa strengthens support for both the ecdyozoa and lophotrochozoa clades but the exact relationships within these two clades remain elusive 353643. Chætognaths and rotifers do not exhibit any peculiar affinities, prompting us to reject the gnathifera hypothesis 38. Conversely, the branching of rotifers is problematic since this phylum is alternatively included in ecdysozoans and lophotrochozoans depending on the use of, respectively, site heterogeneous or homogeneous models (Figure 4). Thus, the clustering of platyhelminthes and rotifers in a platyzoa clade is supported by the WAG model but rejected by the CAT model, suggesting that this grouping may be somehow related to long-branch attraction (Figure 4). Alternatively, previous studies based on morphology and SSU genes have not argued for the ecdysozoan affinities of rotifers 3839. Surprisingly, CAT model analysis no longer succeeds in recovering the monophyly of the deuterostomes (Figure 4b). Instead, it provides limited support for the successive divergence of chordates and ambulacrarians (echinoderms and hemichordates; PP 0.9; Figure 4b). This striking topology was recovered by an independent study using the same heterogeneous CAT model 43 but was neither confirmed by WAG analyses (BP 89 for the monophyly of deuterostomes; Figure 4a) nor supported on morphological bases 3438. One can consider that the two unexpected branchings of rotifers and deuterostomes may be related to some artifact affecting the CAT model, such as sensitivity toward compositional biases 44. Finally, the placozoan Trichoplax adherens surprisingly clustered within the poriferans, as a sister group of the homoscleromorphs (BP 91, PP 0.94; Figure 4), although this poriferan status has never been suggested before 4546. These challenging hypotheses will be investigated in further studies because they have deep implications for the evolution of metazoans (F Marlétaz et al., in progress).

Through extended taxon sampling and improved substitution models, these analyses strongly confirm our previous statements about basal-protostome branching of chætognaths 18 and exclude the basal-lophotrochozan hypothesis 19. Although some areas of bilaterian trees are sometimes incongruent depending on models and inference methods, the position of chætognaths remains remarkably stable throughout our analyses. Furthermore, this branching is not only supported by the presence of GAMT, an unambiguous molecular signature, but also by the posterior Hox genes of chætognaths that are not related to the classes specific to ecdysozoans (Abd-B) or lophotrochozoans (Post1/2) 16. Finally, this topology was also recovered by independent studies involving alternative gene and taxon sampling 303543. In a broader perspective, the strengthening of their phylogenetic position makes chætognaths a key model for comparative genomics among bilaterians.

The clustering of similar sequences indicated that alternative nucleotide forms are present among the transcripts encoding the same protein. Two distinct forms are observed in most cases, although three forms encode some proteins. These forms are separated by a large amount of molecular divergence and can also be distinguished by their different 5' and 3' untranslated regions (UTRs), suggesting that they correspond to different genes (Figure 5 and Additional data files 5 and 6).

Ka/Ks ratios were calculated for all pairs of diverging forms to consider the impact of the nucleotide divergence on the protein sequences. The values of Ka/Ks range from 0.001-0.154 with a median value of 0.004, which confirmed the strong conservation of amino acid sequence despite the large synonymous substitutions observed in some cases (Ks values range from 0.8-75; Table S2 in Additional data file 2). These distinct forms were mainly retrieved for the most highly expressed genes, among which RP genes are prominent (Table 1). We verified that the observed molecular divergence could not be explained by the clustering of distant paralogous sequences. For the genes that have clear homologs among metazoans, the sequences of alternative forms always cluster together in phylogenetic analyses and are thus strongly separated from homologous genes of other animals. For instance, the RUX genes have undergone an ancient duplication resulting in the RUX-E and RUX-G paralogs in all metazoans. Interestingly, chætognaths display up to three forms of RUX-E and two forms of RUX-G, all these forms being closely related (Figure S4 in Additional data file 1; Additional data file 7).

Such a pattern could be explained by either the duplication of a large set of genes in the genome of chætognaths or, alternatively, it could be explained by the presence of cryptic species within the sampled population. In the first case, the observed differences would be attributed to the divergence between paralogous genes originating through the duplication, where the genome of one individual is expected to contain the two alternative nucleotide forms. In the second case, the observed genetic differences would be caused by the genetic divergence between the orthologous genes of several cryptic species spread among the population, where one individual is thus expected to contain only one of the alternative forms.

This cryptic speciation hypothesis may be supported by the strong polymorphism also observed for all genes of the mitochondrial genome, which constitutes an independent lineage from the nuclear genome. For example, cytochrome b (Cytb) transcripts but also cytochrome oxydase I and III are split into distinct forms separated by large molecular distances (Figure 5c; Figure S4 in Additional data file 1; Additional data files 8-10), thus testifying to the presence of distinct mitochondrial lineages within the sampled population.

To decide between these hypotheses, we designed a PCR screen to survey the alternative forms of selected markers in independent individuals. The genes for RPs L36 and L40 were targeted because they are nuclear genes displaying two alternative forms with the highest number of transcripts in the library (Table S2 in Additional data file 2). The mitochondrial Cytb gene served as an independent reference for the interpretation of results from nuclear genes. The three distinct forms of this strongly diverging mitochondrial gene were surveyed in all the individuals tested (Figure 5c). Chætognaths are hermaphroditic and, after fertilization, they store exogenous sperm in their sperm receptacles (Figure 5a), which makes it possible to amplify the DNA from another individual. Hence, in order to detect such contamination, we performed independent amplifications on heads, which are considered free from sperm contamination, as well as on the rest of the body, which contained sperm receptacles (Figure 5a). The experimental design made it possible to detect alternative forms through the amplification of specific DNA fragments of distinct sizes (Figure 5b; Table S3 in Additional data file 2). The PCR products were characterized by sequencing and nucleotide polymorphism was subsequently carefully examined. In addition to their nucleotide divergence in coding sequences, the distinct forms of nuclear genes for RPs L36 and L40 have alternative intron positions and lengths as well as differences in their 5' and 3' UTR regions (Figure 5b).

Performed on nine individuals, the amplifications revealed the presence of the two forms of the nuclear genes for RPs L36 and L40 in each individual (Table 2). Conversely, only one form of the mitochondrial Cytb gene was amplified in each individual with the exception of the body of individual 1, which includes two forms, thus suggesting contamination by exogenous sperm (Table 2). The amplification of the divergent nucleotide forms within one individual indicates that the alternative nucleotide forms correspond to paralogous nuclear copies originating through past gene duplication events (Table 1). Conversely, the alternative forms of the mitochondrial gene correspond to variation within the population. Because some genes, such as that encoding Translationally controlled tumor protein (TCTP), do not present paralogous copies despite their high expression levels (112 TCTP transcripts in the EST collection; Table S2 in Additional data file 2), we addressed the extent of these duplications in evaluating the quantity of duplicated genes. If the clusters of transcripts encoding the same protein include all the transcripts from alternative paralogous genes and if those paralogous genes have similar levels of expression, the probability that transcripts from these paralogous genes are represented in a given cluster is related to the size of this cluster (see Materials and methods). Hence, all the clusters that include more than six transcripts have at least a 95% chance of including transcripts from the two copies if they exist. Such clusters of transcripts were all checked for paralogous copies through sequence alignments and trees. Paralogs were detected within 35 of the 66 clusters investigated, which suggests that up to 69% of chætognath genes are the products of duplications. These paralogs could have arisen through either a whole genome duplication (WGD) event followed by an extensive gene loss, or several segmental duplication events.

The hypothesis of a WGD event is reinforced by the high occurrence of RPs among duplicated genes (Table 1). The trend to retain RP genes was previously observed after WGD for Paramecium tetraurelia, yeast and plants 474849 but is not a common occurrence in small-scale duplications. Conversely, it is difficult to understand why the paralogous genes have been retained after their duplication and maintained under purifying selection as emphasized by Ka/Ks values. This conclusion is in contradiction with the current view of gene destiny after genome duplication, which alternatively predicts that one of the gene duplicates is lost or undergoes the accumulation of substitutions 50. Using a genome-level dataset, similar findings were made about the strongly duplicated genome of Paramecium where the retention of duplicated genes was accounted in part by dosage compensation constraints 47.

The most plausible dating is that this duplication occurred before the diversification of the major chætognath lineages. Two copies of SSU and LSU were retrieved in members of the phylum dispersed all over the tree of chætognaths 101112. Moreover, the survey of 226 ESTs available for Flaccisagitta enflata also revealed the presence of alternative nucleotide forms for some genes (data not shown), which would confirm that the duplication is not limited to SSU/LSU genes at this taxonomic scale. Further genome data would be required to date the duplication, for instance, in considering the Ks distribution of the set of paralogs 51, and also to definitively state the nature of the duplication through the analysis of synteny in duplicated blocks of the genome. Nevertheless, this preliminary transcriptomic survey stresses the usefulness of the chætognaths to study phylum-level genome duplication events and the destiny of paralogous genes.

Beyond the molecular divergence between the coding sequences of duplicated paralogous genes, a subsequent survey of the genomic sequences of selected genes revealed that the level of polymorphism is strong within each paralogous gene (Table S4 in Additional data file 2). Multiple nucleotide substitutions as well as insertion/deletion events (indels) occurred within the introns of the four selected nuclear genes (paralogous copies of the genes for both RPs L36 and L40; Additional data files 11-14). Similarly, a large number of substitutions have accumulated in the various mitochondrial genes, thus revealing distinct mitochondrial lineages within the sampled population (Figure 5c; Figure S4 in Additional data file 1). However, these strong levels of divergence remain consistent with a population genetic structure because of the regular AT composition and the limited degree of saturation revealed by Ts/Tv ratios, singleton positions being essentially transition substitutions (Table S4 in Additional data file 2; Figure S6 in Additional data file 1).

We attempted to determine the origin of this population genetic heterogeneity, which could, for instance, be due to a cryptic speciation or to a past hybridization. For this, the sequences of each individual were compared using phylogenetic trees and indels as discrete informative characteristics (Figure 6). For each marker gene, individual sequences split into several major clades supported by strong bootstrap and discrete indel events, which allows unambiguous identification of heterozygous individuals (Figure 6). For example, individual 4 is heterozygous for all markers and individuals 6, 9 and 3 are heterozygous for at least one marker. Moreover, the occurrence of several cases of putative recombinations between alleles highlights the heterozygous status of some individuals (individuals 3 and 4, Figure 6b,d). Notably, our PCR-based experimental design provided positive evidence only for heterozygosis because two amplifications (head and body) were carried out per individual, yielding 0.5 probability to detect heterozygosity. Heterozygous individuals could thus be even more abundant than observed. These heterozygous cases convincingly demonstrate that a shuffling occurs between the most divergent alleles of each gene, which constitutes strong evidence for interbreeding within the sampled population. This finding definitely excludes the possibility of cryptic speciation within this S. cephaloptera population. Alternatively, the panmixy hypothesis was confirmed by the unimodal distribution of pairwise divergences in mismatch analysis, which is consistent with constant population size and excludes a past hybridization event (Figure S6 in Additional data file 1). Finally, the distinct mitochondrial lineages are spread within the population but they are not correlated with any haplotype differentiation at the nuclear level, which is a strong argument against the cryptic speciation hypothesis. This type of mitochondrial diversity was previously discovered for the planktonic species Sagitta setosa but was also interpreted with difficulty 52.

Strikingly, these comparisons also highlighted molecular divergence between the head and the body of some individuals for each of the five markers investigated (Figure 6 and Additional data file 4). Such substitutions cannot be explained by a heterozygous status of those individuals because sequences from head and body were firmly clustered in the tree (Figure 6). For example, individual 4 exhibits well-separated alleles present in both head and body but intra-individual substitution took place between head and body for both of these alleles (Figure 6c). This pattern of substitutions may be explained by the occurrence of somatic mutations during the life of individuals. This interpretation is corroborated by the large extent of intra-individual substitutions in all marker genes and all individuals. Somatic mutations are considered as rare conditions, mainly known from related disorders in humans 53. Less clear are the evolutionary implications and putative benefits of this phenomenon 54. They are sometimes suspected to play a prominent role in apoptosis and possibly in the regulation of cell division 54. Moreover, somatic mutations have been demonstrated to be more widespread in Drosophila than in mammals 55, and are sometimes correlated with extensive chromosome rearrangement in the Drosophila lineage 56. However, little is known about the extent and importance of this process in the non-model organisms. In the case of the chætognath, somatic mutation could be due to the high mutation rates that seem to affect both germline and soma and could explain the divergence at the population and individual levels. The possible relationship of these accelerated mutation rates with structural reshaping of the genome after duplication deserves further evaluation.

Notably, this level of somatic mutation generates a strong background noise that hinders the accurate interpretation of point mutations related to the diversity of haplotypes. Moreover, traditional hypotheses of population genetics are challenged by our findings: the genetic distances observed between individuals of a single population reach species-level without any evidence for cryptic speciation or past hybridization. In parallel, multiple mitochondrial lineages diverge and are spread and maintained within a single population 52. If such features are revealed as more widespread than expected, the data collected from numerous population genetic or phylogeographic studies require more cautious interpretation. This observation pleads for an increase of the sampling depth for population genetics, especially through genomic approaches 57.

The survey of the chætognath EST collection detected a common 36 nucleotide motif shared at their 5' end between transcripts from unrelated genes. Selected cases showed that these short motifs are absent from upstream genomic regions of the genes (Figure 7c; Additional data files 15-17). These findings strongly suggest that mRNA of S. cephaloptera undergoes trans-splicing maturation. This mRNA processing occurs through the spliceosomal transfer of a small RNA molecule at the 5' end of the mRNA from an independent spliced-leader (SL) gene and has been described in numerous species 26. To exclude the possibility of an artifact, the 226 available EST sequences of F. enflata, another chætognath species, were screened using the identified SL sequences from S. cephaloptera. This survey led to the recovery of similar SL sequences at the 5' end of several cDNA sequences. As F. enflata and S. cephaloptera belong to the two main orders of chætognaths 10, the trans-splicing mechanism is likely to be present in the whole chætognath phylum.

An extensive study of trans-splicing in the nematode phylum has previously revealed the presence of different forms of SL sequences 58. Similarly, several different SL sequences were retrieved from S. cephalotera (Table 3). This number is consistent with the strong level of polymorphism previously observed for coding sequences but two distinct SL forms alone (SL1 and SL2) represent 87% of the SL sequences (Table 3). These forms do not exhibit any specificity for the different paralogs since SLs are added randomly to transcripts from these paralogs (Figure 7; Additional data files 5 and 6). In an attempt to understand the evolutionary history of trans-splicing within the chætognath phylum, the SL forms of F. enflata were compared with those of S. cephaloptera. This set of chætognath SL sequences splits into four main forms: two of them, SL1 and SL3, are present in both the species investigated whereas the two other forms, SL2 and SL4, are specific for S. cephaloptera and F. enflata, respectively (Table 3). Otherwise, neither of these chætognath SL forms display similarity with the SL of another phyla. This finding suggests that just as in nematodes, the evolution of alternative forms could have occurred at a relatively reduced taxonomic scale 58.

Within the EST collection, 2,914 sequences exhibit SL addition, which represents 30% of nuclear transcripts (Figure 8a). Among the SL population, 72% are coding transcripts, of which 46% have a homolog in SwissProt. The clustering of similar coding trans-spliced transcripts indicated that 41% of putative genes undergo SL addition. Furthermore, the relationship between trans-splicing and expression level was tested through the comparison of the number of ESTs per cluster of trans-spliced or non-trans-spliced transcripts. If we posit that this number can be considered as an estimate of the expression level, trans-spliced genes are significantly more expressed than others (Wilcoxon rank test, p < 2.2e^-16). For instance, among the 50 more expressed genes (that is, biggest EST clusters), only two are not trans-spliced. These values suggest that trans-splicing is involved in the regulation of a set of strongly expressed genes responsible for key cellular functions, for example, the RP set.

Hence, the set of trans-spliced genes of S. cephaloptera was compared with those of other animals that carry out this RNA maturation process (Table 4). Particularly, trans-splicing was characterized in the nematode C. elegans model, for which a large EST set is available 5859. These data were especially useful for extensive comparisons with S. cephaloptera (Figure 8b). The amount of ESTs with a SL is slightly smaller in C. elegans (22%) than in S. cephaloptera (29%) and trans-spliced transcripts are missing from the very reduced non-coding ESTs, only 5% of the total EST collection in C. elegans. This comparison suggests that trans-splicing is at least as widespread in S. cephaloptera as in C. elegans. Furthermore, the set of genes affected by trans-splicing in S. cephaloptera was comparable with those of C. elegans. Among the 119 different genes that are both trans-spliced and annotated using SwissProt with high confidence, 79 have trans-spliced homologous genes in C. elegans (Additional data file 3). These values do not include the 78 RP set, which are conserved and trans-spliced in both species. While the molecular actors involved in trans-splicing do not exhibit evolutionary conservation 26, the genes undergoing this kind of processing could otherwise be similar in distantly related species.

Trans-splicing has been discovered in a set of organisms spread all over the phylogenetic tree of eukaryotes, but the molecular actors involved in this RNA maturation process do not exhibit any evolutionary conservation: the SL sequences cannot be aligned and the ribonucleoproteic machineries involved in this process have no homologs in other species 2637606162. However, the discovery of SL trans-splicing in new animal phyla strongly argues for an ancient origin of this process, especially if these phyla have significant phylogenetic positions, such as chætognaths or acoels. Noticeably, we also found evidence for SL addition in recently released ESTs from the acoels (unpublished observations), worm-like animals that were recently excluded from the protostomes 30. Moreover, an extensive comparison of available genomic data revealed that the set of genes undergoing trans-splicing is evolutionarily conserved between those species, which is another indication of a putative common evolutionary origin of trans-splicing (Additional data file 3 and Table 4).

In the nematode chætognath C. elegans and the tunicate Ciona intestinalis, polycistronic mRNA molecules that contain two or more genes are transcribed from operon structures and subsequently resolved by trans-splicing 5859. We mined for similar eukaryotic operons in our pool of sequenced bacterial artificial chromosomes (BACs) in looking for clusters of genes with the same transcription orientation and very short intergenic regions. The trans-splicing maturation of these putative operons was confirmed by mapping the tran-spliced ESTs onto these genomic sequences. One operon including RP S14 and PCNA genes was successfully identified in the BAC 35A21 that otherwise contains a large number of genes undergoing trans-splicing (Figure 7a; Table S5 in Additional data file 2; Additional data files 15-17). EST mapping and ab initio gene prediction suggested that only three bases separate the polyadenylation site of the upstream RP S14 gene from the acceptor splice site at the PCNA downstream gene. This distance is even shorter than that in C. intestinalis operons 61 and clearly excludes the possibility for a re-initiation of transcription between the genes (Figure 7c). This BAC sequence contains several other predicted genes that are closely clustered, share the same orientation and could thus belong to operonic structures (Figure 7; Table S5 in Additional data file 2). The transcripts of gene RP S14 include the two major SL 1 or 2 forms (Figure 7c). Similarly, transcripts from both paralogous copies of duplicated genes bear any of the alternative SL forms. Hence, contrary to C. elegans for which a SL form is specific for the upstream and the downstream genes in operons 58, no specificity of the SL for gene position in the operon was recovered in S. cephaloptera (Figure 7).

Hence, we have provided arguments for polycistronic transcription in another animal phylum, the chætognaths. Animal operons have previously been described in the nematodes and in the tunicate C. intestinalis and were often interpreted as an adaptation to the genome size reduction observed in both these organisms 26. However, the genome size of S. cephaloptera is rather large at 1.05 Gb 63 and the transcriptome of chætognaths is conserved among bilaterians, in terms of gene similarity and content. Furthermore, extensive study of genes co-transcribed in C. elegans led to the conclusion that the operons constitute functional units in grouping genes involved in similar pathways 59. Hence, the presence of operonic transcription in the chætognath, a basal protostome, strongly suggests that this mechanism should be more likely considered as an original mode of gene regulation, whose evolutionary importance may have been underestimated. Thanks to the pivotal comparative genomic abilities of chætognaths, the comparison of operon structure and content between nematodes and chætognaths should allow the extension of the functional results obtained in C. elegans to a larger set of organisms, including vertebrates. Such an approach would require further genomic data in the chætognath but will likely lead to the characterization of new gene regulatory networks, potentially leading to a better understanding of some genetic disorders 59.

The juvenile-staged cDNA library of S. cephaloptera, Bush 1851 was previously described in 18. Briefly, polyA+ RNAs were isolated from hatchling to one-day-old juveniles collected in the coastal area near Marseille (Le Brusc, La Ciotat). After reverse transcription and selection of longer transcripts by size fractionation, cDNAs were cloned into Lambda Triplex 2 vector (Clontech, Palo Alto, CA, USA). There were 12,324 clones sequenced from a 5' primer at Génoscope (CNS, France). A total of 390 sequences were discarded due to vector contamination or cloning artifacts, yielding a total 11,934 sequences.

The BAC library was constructed by BioS&T Inc. (Montreal, Québec, Canada) in the pIndigoBAC-5 vector (Epicentre, Madison, WI, USA) from adult S. cephalotera genomic DNA. Average insert size is 135 kb. The library was arrayed and selected clones were sequenced using a shotgun approach at Génoscope.

The sequences were annotated through Blast searches 68 against SwissProt and TrEMBL databases. The transcripts were searched for reliable coding sequences using ESTScan and subsequently sorted into coding and non-coding sequences 69. Transcripts from the same putative genes were grouped into clusters using CLOBB 70. The functional Gene Ontology classification was retrieved from SwissProt matches using Fatigo 71. Transcripts that bear splice-leader at their 5' extremity were searched with Blastn using the SL sequences obtained from a preliminary screen. Data parsing processes were conducted using Perl scripting language and the Bioperl resource 72.

The set of databases compared with the chætognath transcriptome includes the transcriptomes of D. melanogaster and H. sapiens downloaded from Ensembl 73. The partial transcriptome of Lumbricus rubellus as well as clade-level EST databases were obtained from the NCBI EST database using appropriate keywords 74. The recently released genomes of several lophotrochozoans (Capitella capitata, Helobdella robusta, Lottia gigantea, Aplysia californica, Schmidtea mediterranea and Schistosoma mansoni) were accessed directly on the JGI web site 75 or retrieved from the NCBI trace archive 76. The respective scores were visualized using Simitri 77.

All the transcript sequences encoding similar protein coding sequences were clustered on the basis of a similar SwissProt hit or a reciprocal blast hit after conceptual translation by ESTScan 69. Gene clusters were analyzed using Clustalw alignment, phylogenetic analyses and visual checking for the inclusion of potential diverging sequences (Table S1 in Additional data file 2). The minimal EST number (N) per cluster to detect the occurrence of two duplicated paralogous sequences at a 5% probability is 6 as estimated by 1 - 2(1/2)^N. Ka/Ks were computed using PAML 78.

The RP dataset was built using a composite database approach so as to select the least divergent sequences for each taxon 18. The species set involved in each taxon as well as the data on missing sequences are available in Table S1 in Additional data file 2. This dataset was analyzed using the CAT model as implemented in Phylobayes 79. Maximum-likelihood and Bayesian inferences were performed using PhyML 80, Treefinder 81 and MrBayes 82, respectively, and both assumed a WAG+Γ4+I model. For bayesian analyses (MrBayes and Phylobayes), at least two chains were launched, their convergence was verified and the corresponding burn-in period discarded.

The individuals of S. cephaloptera employed for polymorphism analysis were collected across the Posidonia seagrass, in the calanque of Sormiou, in the coastal area near Marseille, France. Copies of nuclear RP S36 and RP L40 and mitochondrial Cytb genes were amplified by PCR using allele-specific primers (Table S3 in Additional data file 2). Genomic DNA extractions were carried out on heads and bodies of S. cephaloptera individuals from the calanque of Sormiou using the Wizard SV Kit (Promega, Madison, WI, USA) and PCR reactions were carried out using GoTaq polymerase (Promega). PCR conditions were defined according to the melting temperature determined for each pair of primers. Fragments were subsequently gel-purified, cloned in pGEM-T easy (Promega) and sequenced. PCR artifacts were excluded by two approaches: first, amplifications were performed on head and body of the same individual, which allows checking of the congruence between the sequences; then, the accuracy of DNA polymerase was verified by amplifying and sequencing cloned DNA. These validations were confirmed by the recovery of the amplified sequences in the EST library and by the high levels of substitutions between sequences, which are too large to be interpreted as PCR artifacts.

Pairwise sequence comparisons and molecular evolution analyses were performed with Mega2 software assuming the kimura 2 parameters model 83. The demographic history of the populations was evaluated by mismatch analyses using Tajima's D and Fu and Li's D neutrality test implemented in DNAsp software 84.

The set of sequences analyzed in this paper has been deposited in GenBank with the following accession numbers: CU555081-CU563075 and CU693608-CU693592 for the 11,934 ESTs, CU672232 for the BAC 35YA21, and EU529890-EU529968 for the sequences of RP L36 and L40 paralogs in nine individuals (Table 2).