Introduction

The understanding of chordates origin and evolution has been a preeminent challenge for biologists over the last two centuries (Swalla and Xavier-Neto 2008). The phylum Chordata consists of three distinct lineages: cephalochordates (lancelets), tunicates or urochordates (sea squirts, salps and appendicularians) and craniates (i.e., cyclostomes and vertebrates). Among these, tunicates embrace a diversity of more than 3,000 filter-feeding and mostly hermaphrodite marine species, characterized by both sessile (ascidians) and planktonic (salps and appendicularians) life styles, associated with a unique developmental program (Cone and Zeller 2005; Lambert 2005; Satoh 2003).

In sharp contrast with vertebrates and lancelets, the vast majority of tunicate adults lack the hallmarks of the typical chordate body plan, namely a dorsal neural tube and a notochord. Since the time it has been pointed out that ascidian larvae do possess these chordate features (Kowalevski 1868), tunicates have been considered as the sister group of Euchordata, i.e., cephalochordates plus craniates (Cameron et al. 2000; Mallatt and Winchell 2007; Swalla et al. 2000; Winchell et al. 2002). Although having widely been accepted during the pre-genomic era, this euchordate view has been substantially impacted by the genome projects of the two model ascidians Ciona intestinalis (Dehal et al. 2002) and Ciona savignyi (Small et al. 2007b), as well as the draft genome of the appendicularian Oikopleura dioica (Seo et al. 2001).

These genomic data allowed including tunicates in phylogenomic studies and accumulating evidence soon indicated that the evolutionary positions of cephalochordates and tunicates should be reversed, erecting a clade called Olfactores that links tunicates with vertebrates (Bourlat et al. 2006; Delsuc et al. 2006; Delsuc et al. 2008; Dunn et al. 2008; Singh et al. 2009). Although it was debated at first, the Olfactores hypothesis of chordate evolution has now gained wide acceptance since it has been corroborated by analyses of the complete genome sequence of the amphioxus (Putnam et al. 2008). Moreover, the latest mitogenomic analyses also provided additional support for such a relationship within chordates (Singh et al. 2009).

Comparative genomics also revealed that tunicates possess the basic developmental genetic toolkit of vertebrates in unexpectedly condensed genomes that have not undergone the subsequent vertebrate-specific duplication events (Cañestro et al. 2003; Dehal and Boore 2005; Dehal et al. 2002). This prompted a growing interest in tunicates as useful experimental systems to decipher the developmental mechanisms underlying chordate origins (Davidson 2007; Holland and Gibson-Brown 2003; Satoh 2003). Despite this renewed interest, our current understanding of tunicate evolution remains fragmentary since phylogenomic and especially comparative genomic studies are taxonomically biased towards model species with available genomes (Bourlat et al. 2006; Delsuc et al. 2006; Delsuc et al. 2008; Donmez et al. 2009; Imai et al. 2006; Kim et al. 2007; Satoh 2003; Sierro et al. 2006; Yandell et al. 2006). Complementary EST data are available in a few additional species like Diplosoma listerianum (Blaxter and Thomas 2004), Molgula tectiformis (Gyoja et al. 2007) and Halocynthia roretzi (Kim et al. 2008). However, the aforementioned species are far from covering the diversity of tunicates, so that genomic data are still disproportionately distributed among the major phylogenetic lineages. Moreover, tunicates are currently subjected to a paucity of nuclear phylogenetic markers, as indicated by the exclusive use of the 18S rRNA associated with open controversies regarding the evolutionary history of the group (Tsagkogeorga et al. 2009; Yokobori et al. 2006; Zeng et al. 2006; Zeng and Swalla 2005).

One reason invoked to explain this situation involves the high level of divergence of tunicate lineages due to their accelerated evolution (Zeng et al. 2006). Several recent genome-based phylogenies inferred from either nuclear or mitochondrial genome data have provided clear-cut evidence for the particularly high rates distinguishing tunicate evolution, as illustrated by the persisting long branches of the group in the reconstructed trees (Bourlat et al. 2008; Delsuc et al. 2008; Gissi et al. 2008; Singh et al. 2009). Similar conclusions have been drawn by studies focusing on individual genes, such as 18S rRNA (Perez-Portela et al. 2009; Yokobori et al. 2006), cox1 (Turon and Lopez-Legentil 2004), Huntingtin (Gissi et al. 2006), P transposase (Kimbacher et al. 2009), or chordate gene families such as CYP 1 (Goldstone et al. 2007), suggesting that tunicate sequence divergence may increase up to 30% between the species of the same genus. Similarly, whole-genome sequence data analyses have also revealed high rates of molecular evolution at a within-species level, as indicated by the extremely high rates of structural (16.6%) and single nucleotide polymorphism (4.5%) characterizing the Ciona savignyi genome (Small et al. 2007a). As far as genomic features are concerned, it has been shown that, besides genome contraction, tunicates have also undergone dramatic genomic rearrangements associated to numerous gene losses (Holland and Gibson-Brown 2003). Breaking prominent chordate paradigms, intron positions are highly variable in Oikopleura doica genes (Edvardsen et al. 2004), the Hox cluster is disintegrated in Ciona intestinalis (Ikuta et al. 2004) and Oikopleura doica (Seo et al. 2004), and the main retinoic acid-signalling genes are lacking in the Oikopleura dioica genome (Cañestro and Postlethwait 2007). All afore-mentioned elements suggest a high degree of tunicate genomic divergence from other chordate lineages and provide compelling evidence that tunicates exhibit high rates of molecular evolution.

Given these puzzling issues, we conducted an in silico identification of new nuclear coding phylogenetic markers for tunicates from orthologous housekeeping genes. These candidate markers were then validated in a non-model species, Microcosmus squamiger (Ascidiacea: Stolidobranchia, Pyuridae) through an in vitro transcriptomic approach involving high-throughput 454-sequencing (Margulies et al. 2005). In order to shed light on the patterns of evolutionary rate variation among tunicates and other chordates, we carried out a comparative analysis based on a subset of 35 highly conserved orthologs. More precisely, our study aimed at testing the hypothesis of tunicate accelerated evolution at independent nuclear loci and assessing their degree of divergence from the other chordate lineages. To this goal, we addressed the following questions: (1) Are the evolutionary rates of amino acid replacement consistently higher in tunicates than in vertebrates and, if so, to what extent? (2) Do chordates exhibit homogenous within-lineage rates and, if not, which genes and/or which species contribute the most to the observed heterogeneity? (3) Finally, what are the underlying cause(s) of molecular evolutionary rate heterogeneities within Olfactores?

Materials and Methods

Biological Sampling and 454 Sequencing of Microcosmus squamiger

Microcosmus squamiger individuals were collected from an invasive population at the locality of Cubelles (Rius et al. 2009), in the Spanish Northeast littoral (41°11′37.2″ N, 1°39′17.46″ E). Muscle and gonad tissues were dissected and rapidly frozen in liquid nitrogen. Given the small size of this species, tissue material from 8 individuals was mixed and used for standard cDNA library construction in pCMV Sport 6.1 vectors (Invitrogen; Carlsbad, CA, USA). The obtained library was estimated to contain 1.9 × 107 cDNA clones, with inserts of about length 2 kb in average which were subsequently sequenced using the 454 Roche GS FLX standard system.

Identification of Orthologs

First, we formatted three reference databases: (1) a database consisting of all Ciona intestinalis transcript sequences (19,858 transcripts); (2) a second including only the orthologous genes between Ciona intestinalis and Ciona savignyi (9,520 genes); and (3) finally, a protein database built from a previous phylogenomic data set of 179 orthologous markers for 51 metazoans (Delsuc et al. 2008). We scanned then the 454 data generated, by conducting similarity searches against these databases, and vice versa, using the programs BLASTN, TBLASTX, BLASTX and TBLASTN, respectively, with a cut-off e-value of 10−6 (Altschul et al. 1990).

Contigs of the sequences matching the third database were assembled using the program CAP3 (Huang and Madan 1999). Orthology assignment was controlled in two ways: (1) by reciprocal BLAST searches, i.e., by checking that the best hit for a given contig indeed corresponded to the targeted gene, and (2) by phylogenetic tree reconstruction controlling that Microcosmus contigs cluster within tunicates in the highest likelihood trees reconstructed by TREEFINDER (Jobb et al. 2004) under the GTR + Γ4 + I model. The use of the above criteria allowed us to determine the orthologous coding sequences of Microcosmus squamiger for 68 metazoan housekeeping genes from the initial 179 candidates. Among these, 35 genes for which orthology was unequivocally assessed through phylogenetic analyses were finally retained. The resulting 35 Microcosmus squamiger sequences have been deposited in the EMBL nucleotide database under Accession Numbers FN984758 to FN984792.

Data Set Assembly

Under conservative constraints of orthology assessment using phylogenetic reconstruction and adequate taxonomic representation of tunicates, the number of usable markers was restricted to 35 (Table 1). Individual gene data sets for these housekeeping genes were built upon available transcript data for 21 additional metazoan species: three non-bilaterians, one poriferan and two cnidarian species, used as outgroups; and 18 bilaterians, including 7 Protostomia and 11 Deuterostomia species. Nucleotide sequences for these taxa were retrieved from multiple online resources, such as the Ensembl Genome Browser (http://www.ensembl.org/), the EST Database (http://www.ncbi.nlm.nih.gov/dbEST/) and Trace Archive (http://www.ncbi.nlm.nih.gov/Traces/) of GenBank at the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/), as well as other eukaryotic genome databases hosted at the website of the Department of Energy Joint Genome Institute (http://www.jgi.doe.gov).

Table 1 List of the 35 metazoan-conserved housekeeping genes for which orthologous sequences have been identified in the transcriptome of Microcosmus squamiger
Full size table

Each of the 35 nuclear DNA data set was translated into protein. Amino acid alignments were generated using MAFFT version 5 (Katoh et al. 2005). Gaps were then reported on the corresponding DNA sequences to obtain nucleotide alignments. Some divergent regions of these alignments were subsequently refined manually using MEGA 4 (Tamura et al. 2007). Ambiguously aligned sites were then identified at the amino acid level and separately removed from all the individual gene data sets, using the program GBLOCKS (Castresana 2000), according to the following parameters: a minimum of 14 sequences for conserved and flanking positions, a maximum of 8 contiguous non-conserved positions, a minimum of 5 positions for the length of a block, and a maximum of 50% of gaps per position. The resulting blocks of each protein alignment were then transposed at the DNA level. The concatenation of the 35 phylogenetic markers yielded two 28-taxon data sets: a nucleotide data set including 25,026 unambiguously aligned sites (8,342 codons), and the corresponding amino acid data set. The percentage of missing data in these matrices was only 10.7%. All data sets are available upon request.

Phylogenetic Analyses

Probabilistic phylogenetic analyses of the two concatenated data sets were conducted following Bayesian Inference (BI) and Maximum Likelihood (ML) reconstruction approaches. Bayesian inferences were conducted under the site-heterogeneous CAT mixture model (Lartillot and Philippe 2004) combined to a 4-category Gamma (Γ) distribution, i.e., CAT + Γ for amino acid and CAT + GTR + Γ for nucleotide alignments. All analyses were run using Phylobayes version 3.1 (Lartillot et al. 2009). For each data set, two independent Markov Chains Monte Carlo (MCMC) were launched with parameters and trees sampled every 10 cycles until 1500 points were attained. Priors were set to default values and convergence of the two chains was checked by monitoring the marginal likelihoods through cycles. Bayesian Posterior Probabilities (PP) were obtained from the 50% majority rule consensus of the 1000 trees sampled during the stationary phase, after a burn-in of 500 cycles.

ML analyses of the nucleotide data set were conducted using PHYML 3.01 (Guindon and Gascuel 2003) and PAUP* 4.0b10 (Swofford 2002). The ML parameters of the GTR + Γ model were first estimated by conducting an initial heuristic search in PHYML using a BIONJ starting tree followed by Subtree Pruning and Regraphting (SPR) branch-swapping. The optimal parameter values were then used in PAUP ML heuristic searches implemented by applying Tree Bisection and Reconnection (TBR) branch-swapping on a Neighbor-Joining (NJ) starting tree. For the amino acid dataset, ML analyses were performed using the program PHYML under the CAT-C20 + Γ model with the number of CAT categories set to 20 (Le et al. 2008). The heuristic ML searches were conducted by performing SPR moves on a NJ starting tree.

For both nucleotide and amino-acid ML trees, statistical support was estimated by bootstrap resampling with 100 pseudo-replicates generated by the program SEQBOOT of the PHYLIP package (Felsenstein 2005). In all replicates, ML analyses were performed as described above for the original data sets. Bootstrap percentages (BP) were obtained from the 50% majority rule consensus of the 100 pseudo-replicated trees using the program TREEFINDER.

Substitutional Saturation

We evaluated the extent of substitutional saturation in our data using saturation plots (Philippe et al. 1994) for all types of concatenated data sets, nucleotide and amino acid, as well as for the first plus second, and third codon positions separately. The amount of saturation was estimated by plotting the observed pairwise distances on the alignments against the tree-inferred distances. The latter patristic distances were calculated from the CAT + Γ branch lengths estimated by PhyloBayes under a fixed topology. All analyses were carried out using the APE package (Paradis et al. 2004) within the R statistical environment (The R Development Core Team 2004).

Testing Relative Rate Variations Between Tunicates and Vertebrates Using Local Clocks

Given the depth of the divergences involved in our phylogeny, all following analyses were conducted at the amino-acid level in an effort to minimize the misleading effects of saturation. The tree topology previously obtained from the Bayesian CAT + Γ analyses of the concatenated amino-acid data set was used as a reference, and the variation of evolutionary rates between the tunicate and vertebrate lineages was assessed using the program CODEML of the PAML package version 4.2 (Yang 2007).

We first specified two local clock models (using option Clock = 2 in CODEML): (1) a two-rate model, where all branches in our phylogeny conform with the clock assumption and have a default rate (r 0 = 1) to the exception of the Olfactores clade, encompassing tunicate and vertebrate branches, which was assumed to have a different rate r 1; and (2) a three-rate model in which tunicate and vertebrate branches were assigned distinct rates, r 1 and r 2.

ML analyses were performed under these two local clock models on the concatenated data set as well as on each individual gene alignment. Given that the first model is hierarchically nested into the second, log-likelihoods of each model were compared using likelihood ratio tests (LRTs) (Felsenstein 1981), with degrees of freedom equal to one (i.e., the difference in the number of parameters included in the models). The associated p-values were corrected for multiple testing (e-values), by being multiplied by the total number of genes tested in parallel according to a Bonferroni correction (Boorsma et al. 2005). In these tests, the two-rate model (1) represented the null hypothesis that the rate of evolution is homogeneous between tunicates and vertebrates.

Protein Evolutionary Rate Estimation Under Autocorrelated Models

Because of practical constraints in using the program MULTIDIVTIME (Thorne et al. 1998), all amino-acid alignments for the 35 orthologous markers were slightly modified in order to ensure the presence of one non-bilaterian outgroup taxa in each gene data set. In this respect, Nematostella sequences had to be removed from all gene alignments because this taxon was missing for some genes. The missing sequences of Reniera (genes rpl10l and rplp2) were replaced by the corresponding orthologs of its relative Suberites domuncula, in order to obtain a complete poriferan outgroup.

For each of the 35 resulting protein data sets, branch-specific rates of amino acid substitution were estimated with MULTIDIVTIME under the JTT model. We used the previously defined reference topology after pruning Nematostella, and set the prior age of the root at 700 Mya (Douzery et al. 2004) with no additional constraints on nodes. Two independent MCMC were launched and run for 1 million generations, sampling every 100 generations and discarding the first 100,000 as the burnin. The rates assigned to terminal branches were taken as species-specific evolutionary rates expressed as the number of amino-acid replacements per 100 sites per million year (Myr).

Estimation of Non-Synonymous and Synonymous Substitution Rates in Olfactores

Given the extent of substitutional saturation at the large-scale in our data set, we selected the three pairs of most closely related species to estimate the ω = d N/d S ratio for tunicates, teleosts and mammals. This criterion resulted in the following pairs being selected: Ciona intestinalis and Ciona savignyi for tunicates, Danio rerio—Tetraodon nigroviridis for teleosts, and Monodelphis domestica—Homo sapiens for mammals.

From the 35 previously assembled individual gene data sets, we extracted the sequences of the six afore-mentioned taxa. Nucleotide alignments were then processed using Gblocks in order to exclude all codons that contained gaps (option: no gap positions allowed). Five markers (psmb1, rplP2, rpl35A, rpl39L and rpsA) had to be discarded during this process because of too many missing characters. Thirty genes were thus analyzed, encompassing in total 19,083 nucleotide sites or 6,361 codons for the three selected pairs of tunicate, teleost and mammal species.

The numbers of synonymous substitutions per synonymous sites (d S) and non-synonymous substitutions per non-synonymous sites (d N) were estimated for each of the three species pairs and all thirty housekeeping genes, using the ML method of Goldman and Yang (Goldman and Yang 1994), implemented in CODEML (options seqtype = 1, runmode = −1 and CodonFreq = 2). Similar results were also obtained when analyses were repeated using the YN00 counting method (Yang et al. 2000) implemented in CODEML. Three genes for which a d S greater than 10 was obtained (rpl15, rpl7 and sars) were excluded, as they undoubtedly present unreliable estimates due to strong substitutional saturation through time. Finally, the distributions of the synonymous and non-synonymous substitution rate (d S and d N) estimates for tunicates, teleosts and mammals across the 27 retained housekeeping genes were plotted using R.

Results

Microcosmus squamiger 454 Transcriptome Sequencing and Identification of Orthologs

The 454 sequencing of the Microcosmus squamiger partial transcriptome on the GS FLX standard platform yielded about 50 million high quality bases in total, which correspond to 211,899 sequence reads with a mean length of 250 bp. Similarity searches using BLASTN against either the intersection between Ciona intestinalis and Ciona savignyi orthologous genes (9,520 genes) or the Ciona intestinalis full transcriptome (19,858 transcripts) resulted in 46,245 and 48,455 matching reads corresponding to 216 and 245 genes, respectively. When using TBLASTX searches, the total number of putative homologous read sequences to Ciona orthologs was raised to 126,830 belonging to 834 genes.

At a larger taxonomic scale, positive BLAST results were obtained for about 11% of the total number of Microcosmus reads which matched 68 of the 179 initially targeted nuclear protein-coding genes conserved across metazoans. Among these, 35 genes for which orthology was unequivocally assessed were finally retained. The 35 orthologous gene sequences for Microcosmus squamiger were finally assembled from a total of 24,209 matching reads, with a mean coverage of 691 ± 24 reads per gene reflecting the high expression level of these housekeeping genes. According to the Gene Ontology (Ashburner et al. 2000), the 35 genes (Table 1) were mostly involved in translational elongation (ribosomal proteins) but also in metabolic or catabolic processes (AHCY, SUCLG1, PSMA6, PSMB1 and HSP90AB1), glycolysis (PDHB), RNA splicing (U2AF1), tRNA processing (SARS) and protein folding (CCT5).

Tunicate Phylogenetics Based on 35 Housekeeping Genes

Saturation plots revealed a high degree of substitutional saturation at 3rd codon positions compared to positions 1 and 2 and amino acids (Fig. 1). Consequently, 3rd codon positions were excluded from subsequent phylogenetic analyses of the nucleotide data set. Phylogenetic reconstructions conducted at both the amino acid and nucleotide levels using ML and BI approaches under standard and mixture models of sequence evolution yielded almost identical trees. The resulting phylogenetic picture (Fig. 2) conforms to the widely accepted view of animal evolution (Telford 2006) according to which Bilateria can be divided into Protostomia and Deuterostomia, the former embracing two reciprocally monophyletic lineages; Lophotrochozoa (PPaa = 1, BPaa = 98, PPnt = 0.98 and BPnt = 98) and Ecdysozoa (PPaa = 1, BPaa = 99, PPnt = 1, and BPnt = 100). All analyses were highly congruent as far as the monophyletic origin of Protostomia was concerned (PPaa = 0.99, BPaa = 93, PPnt = 1 and BPnt = 81), yet Deuterostomia relationships appeared more difficult to resolve, particularly in ML trees (PPaa = 0.89, BPaa < 50, PPnt = 0.87 and BPnt < 50). In detail, ML analyses on nucleotide positions 1 and 2 under the standard GTR + Γ model yielded a ML tree where Xenambulacraria (Xenoturbella + Hemichordata + Echinodermata) appeared as the sister-group of all other bilaterians thereby disrupting deuterostome monophyly, although without statistical support (BPnt < 50). In ML analyses of the amino acid data set under the CAT-C20 + Γ model, Xenoturbella emerged as the sister-group of the remaining bilaterians (BPaa < 50) and Branchiostoma clustered with Ambulacraria with moderate support (BPaa = 77). As pointed out, no significant nodal support was obtained for neither conformations, and similar results were previously shown to likely represent phylogenetic reconstruction artifacts (Bourlat et al. 2006; Delsuc et al. 2008), although the phylogenetic position of Xenoturbella is still contentious (Hejnol et al. 2009).

Fig. 1
figure 1

Levels of saturation inferred for first plus second codon positions, third codon positions and amino acids. All the three plots represent the relationship between the tree-inferred distances and observed pairwise distances at nucleotide and amino acid level, respectively. Dotted lines represent the hypothesis of absence of saturation on the data, where estimated and observed distances are equal. Solid lines indicate the regression of the linear model fitted on the data

Full size image
Fig. 2
figure 2

Superimposition of the consensus Bayesian trees inferred from amino acids and from first and second codon positions under the CAT + Γ site—heterogeneous mixture model. Values at nodes represent Bayesian Posterior Probabilities (PP)/Maximum Likelihood bootstrap percentage (BP) obtained for the amino acid (aa) and nucleotide (nt) data sets respectively (PPaa/BPaa/PPnt/BPnt)

Full size image

Within Chordata, a sister-group relationship of tunicates with vertebrates was unequivocally recovered with all methods and data sets (PPaa = 1, BPaa = 94, PPnt = 1, BPnt = 95). Of note, the Olfactores clade is retrieved from nucleotide-based reconstructions, giving further credit to the robustness of the new chordate phylogeny and suggesting that the core of the 35 housekeeping genes we considered carries a strong phylogenetic signal largely congruent with that of larger phylogenomic data sets (Delsuc et al. 2006; Delsuc et al. 2008; Dunn et al. 2008; Putnam et al. 2008).

Similarly, the 35 phylogenetic markers appeared informative enough to provide a clear-cut phylogenetic picture within each group. The newly sequenced Microcosmus squamiger was found firmly branched with Halocynthia (PPaa = 1, BPaa = 100, PPnt = 1, and BPnt = 100), a relationship consistent with the traditional classification, both species belonging to the same order and family (Stolidobranchia: Puyridae). Likewise, the resulting relationships for the remaining tunicate species appeared highly concordant with recent phylogenies inferred from 18S rRNA data (Tsagkogeorga et al. 2009).

For the majority of species, branch-length estimates obtained from phylogenetic reconstructions based on amino acid data were similar to those inferred from nucleotide sequences, with the exception of some fast-evolving taxa within Ecdysozoa and Tunicata. Oikopleura dioica constitutes such an exception, since its DNA branch length is 1.6 times longer than its amino acid one, with the mean ratio for the other tunicate branches being equal to 1.1. Moreover, Oikopleura dioica was found to exhibit a highly distinct compositional profile in PCA analyses (data not shown), occupying an outlier position relative to the other tunicate and deuterostome taxa, and being close to Caenorhabitis elegans, the second fastest-evolving species in our phylogeny (see Fig. 2).

Individual Gene Rate Differences Between Tunicates and Vertebrates Under Local Clocks

In all phylogenetic trees inferred from the full data set of 35 proteins, tunicates displayed the longest branches within Chordata indicating an overall faster evolutionary rate than vertebrates and cephalochordates. In order to test whether this acceleration affects all the genes to a similar extent, we followed an ML approach to obtain relative rate estimates for the two Olfactores lineages in each orthologous gene. LRTs between a constant-rate model and a variable-rate model allowed us to test for statistical significance in rate differences between the two groups. The results indicated that the relative rate of amino acid replacement in tunicates and vertebrates significantly differs between the two groups for 20 genes (Supplementary Table). The tunicate rate was consistently found to be superior to the vertebrate rate for all the afore-mentioned 20 genes, upholding the assumption of tunicate rapid evolution for the housekeeping genes considered. Moreover, the rate ratio estimates ranged from 1.4 to 7.5 depending on the marker, which implies that the evolutionary shift in tunicate rates has not affected all the genes to a similar extent (Fig. 3).

Fig. 3
figure 3

Contrasting local molecular clock estimates among tunicates and vertebrates for the 35 housekeeping proteins. The graph shows the tunicate/vertebrate rate ratio estimated for each of the 35 genes. Black bars indicate genes for which rate differences between the two groups were not significant after correcting for multiple tests (LTR test, critical e-value = 0.05)

Full size image

Unexpectedly, although the rate ratio for the remaining 15 genes was superior to 1, except for three of them (rplp2, rpl12 and rpl39l), the LRTs did not significantly reject the hypothesis of equal rate between the two groups. Thus, in these cases the evolutionary rate within Olfactores may be considered roughly constant, a result providing evidence that tunicates do carry genes that have escaped the prevalent genomic acceleration. The outline of the estimated rate variation within Olfactores is shown in Fig. 4 which illustrates the global rate contrast between tunicates and vertebrates across the 35 proteins. When the same analytical protocol was applied to the concatenated data set, an overall rate ratio of 1.9 was obtained.

Fig. 4
figure 4

Average contrast of rate variation in amino acid replacements within Olfactores. The figure illustrates tunicate (grey) and vertebrate (white) distributions of the evolutionary rate across the 35 markers. Horizontal bars give the median of rate distributions; boxes give the quartiles; whiskers extend to 1.5 times the interquartile range; and circles are for outliers

Full size image

Within-Group Rate Variation Under an Autocorrelated Model of Rate Evolution

Given the extensive variation in rate among genes, we next asked whether similar fluctuations can also be observed in species rates, and if yes, how they are linked to the afore-described contrasts. Do all tunicates contribute equally to the estimated rate differences among chordates, or the large range of among-gene variations is rather driven by only few species? To answer this question, we estimated the rate of amino acid replacement in all the branches of our phylogeny over the 35 housekeeping genes using an autocorrelated model of rate evolution (Thorne et al. 1998). This allowed us to simultaneously explore two aspects of rate variation within Chordata: both across genes, and across branches of its three subclades.

The branch-specific rate distributions for tunicates, vertebrates, and cephalochordates have been estimated across the 35 molecular markers (Fig. 5). All the branches of the tunicate clade were characterized by an accelerated evolutionary rate as compared to the other chordate groups, whereas the lineage-specific estimates were found to vary extensively across the sampled species. This yielded a particularly heterogeneous picture for tunicate rates, which contrasted sharply with the more homogeneous and mostly overlapping rate distributions of vertebrates, and the slow evolutionary rate characterizing the amphioxus Branchiostoma floridae (Fig. 5). More precisely, the two phlebobranch ascidians (Ciona intestinalis and C. savignyi) seemed to exhibit the lowest rates among tunicates, with the stolidobranchs (Halocynthia roretzi, Microcosmus squamiger and Molgula tectiformis) following next, conforming to previously reported phylogenetic observations (Tsagkogeorga et al. 2009; Yokobori et al. 2006; Zeng et al. 2006). Interestingly, the evolutionary rate heterogeneity appeared radically lessened when considering species of the same order, as best exemplified by the overlap of the gene rate distributions within Stolidobranchia. Notably, the aplousobranch Diplosoma listerianum and, particularly, the appendicularian Oikopleura dioica were detected as highly divergent species, with rates deviating greatly from the average values of all other sampled tunicates in particular, and chordates in general. These two representatives were also characterized by the largest variance in estimates across genes, illustrated by the wide quartiles of their rate distributions (Fig. 5).

Fig. 5
figure 5

Branchwise distributions of tunicate, vertebrate and cephalochordate rates of amino acid replacements as estimated under an autocorrelated rate model across 35 proteins. The top panel shows the Chordate sub-tree extracted from the reference tree with numbering of the corresponding internal branches. In the bottom panel, tunicate rate distributions are shown in dark grey, vertebrates in white and amphioxus in light grey. Horizontal bars give the median of rate distributions; boxes give the quartiles; whiskers extend to 1.5 times the interquartile range; and circles are for outliers

Full size image

Considering the results of the previous local clock analyses, we finally examined the among-branch variations separately for the 20 genes that showed significantly higher evolutionary rates in tunicates, and the remaining 15 genes for which LRTs did not reject the hypothesis of a homogeneous rate among Olfactores. As expected, rate estimates considering the latter 15 genes yielded a more even rate variation across Chordata (Fig. 6). Interestingly, this more homogeneous picture among chordate lineages appeared to hold good mainly in the decrease of Diplosoma listerianum and Oikopleura dioica rates. The two Ciona species, and the three stolidobranch ascidians (Halocynthia roretzi, Microcosmus squamiger and Molgula tectiformis) showed rate profiles almost identical when considering either the full data set or the two sub-sets of 20 and 15 genes, respectively (Fig. 6).

Fig. 6
figure 6

Branchwise distributions of tunicate, vertebrate and cephalochordate rates of amino acid replacements as estimated under an autocorrelated rate model across all 35 proteins, the 20 faster genes, and the 15 slower genes, respectively. All-gene distributions are shown in light grey, the 20 faster-evolving-gene in dark grey and the slower-evolving-gene in white. Horizontal bars give the median of distributions; boxes give the quartiles; whiskers extend to 1.5 times the interquartile range; and circles are for outliers. Boxes highlight the deviating rate distributions of the aplousobranch Diplosoma listerianum and the appendicularian Oikopleura dioica

Full size image

dN/dS Ratio Variation Among Lineages

The synonymous and non-synonymous rate ratios (d N/d S), as estimated for the three most closely related species pairs Ciona intestinalis—Ciona savignyi, Danio rerio—Tetraodon nigroviridis and Monodelphis domestica—Homo sapiens, respectively, and across a subset of 27 genes are shown in Fig. 7. All the d N/d S estimates were much less than 1 reflecting the fact that the set of examined genes is under strong purifying selection in all the represented lineages. This is consistent with the fact that every gene of our analysis is conserved at a notably large taxonomic scale with most of them involved in fundamental housekeeping functions of the cell.

Fig. 7
figure 7

Estimation of the synonymous/non-synonymous rate ratios (d N/d S) according to 27 housekeeping genes in three pairs of closely related taxa. The graph shows the synonymous and non-synonymous substitution rates as estimated using the YN00 counting method (Yang et al. 2000) between the pairs: C. intestinalis–C. savignyi, D. rerio–T. nigroviridis, and M. domestica–H. sapiens for each of the 27 genes with no missing data and a d S ≤ 10

Full size image

Overall, the ω values ranged from 0.001 to 0.14 depending on the lineage and the gene considered, yet none of the lineages presented a systematically higher across-gene d N/d S as compared to the others, ruling out the hypothesis that the observed evolutionary rate variation is due to a relaxation of selective constraints in one particular lineage. Instead, the resulting picture of ω across the three Olfactores groups was rather uniform, as illustrated by the highly similar global distributions of d N/d S observed for tunicates, teleosts and mammals (Fig. 8).

Fig. 8
figure 8

Overall estimation of the synonymous/non-synonymous rate ratios in Olfactores. ω or d N/d S distributions for the pairs: C. intestinalis–C. savignyi, D. rerio–T. nigroviridis, and M. domestica–H. sapiens across the subset of 27 housekeeping genes

Full size image

Discussion

Determining New Phylogenetic Markers for Tunicates Using 454 Sequencing

Over the last 4 years, massively parallel sequencing platforms have become available, enabling the generation of large amounts of genomic data at extremely short times and accessible costs (Shendure and Ji 2008). Using such technologies for tunicates seemed a promising perspective from a phylogenetic and comparative genomics standpoint, mainly because this lineage is far from satisfying the prerequisites for the successful development of phylogenetic markers through traditional PCR approaches. Evolving at elevated rates, tunicate protein-coding genes exhibit high levels of sequence divergence leading to large amounts of saturation, even when considering closely related species such as Ciona intestinalis and C. savignyi. Thus, the identification of conserved regions in a gene, suitable for primer design, appears to be a particularly difficult task. As a result, besides whole-genome projects, the currently available genomic information for tunicates comes from EST and cDNA sequencing projects that do not technically require a priori information and low sequence divergence (Blaxter and Thomas 2004; Gyoja et al. 2007; Kim et al. 2008).

In this study, we applied, for the first time, the high-throughput 454 technology (Margulies et al. 2005) to sequence the partial transcriptome of an ascidian tunicate, Microcosmus squamiger. This invasive species recently became widespread in the western Mediterranean sea while being likely of Australian origin, and has also spread and become invasive in other parts of the world (Rius et al. 2008; Rius et al. 2009). We sought to identify orthologous sequences for a set of 179 previously determined phylogenetic markers used in large-scale comparative analyses (Delsuc et al. 2008). At the time this study was conceived, the 454 sequencing system generated an average of 100 megabases per run, at short 250 base-pair reads. Still this approach permitted us to determine in silico 68 homologous genes, among which 35 were unambiguously identified as orthologous sequences to our query genes.

Although this might seem a small number compared to the initial 50 million base pairs of resulting data, several factors argue for the great potential of the 454 sequencing approach in tunicate phylogenetics and comparative genomics. First, the sequencing of the Microcosmus transcriptome was conducted on a cDNA library that was not normalized. This yielded an important level of coverage or/and redundancy for highly expressed genes in our data, and the 35 housekeeping genes corresponded to a non-negligible amount of 24,209 sequence reads (i.e. about 10% of the initial 211,899 reads). Second, albeit high-throughput sequencing alleviates bench problems, the fast-evolving nature of tunicate species has still a substantial impact on the post-data analyses by hindering the reliable assignment of orthology between highly divergent sequences. Analyses of the Ciona intestinalis genome have demonstrated that in a transcriptome composed of 16,000 genes, only a small fraction (5%) matched to vertebrate genomes, whereas about 60% of the genes are shared with protostome animals (Dehal et al. 2002). In this respect, the limited sequence length of the 454 GS FLX standard platform at the time has reinforced the problem, since short sequences are difficult to assign with confidence through similarity searches.

In conclusion, considering tunicate scarcity in genomic data and available phylogenetic markers, the results from our pilot application of the 454 sequencing approach to the group appears promising. Given the ongoing advances of these high-throughput sequencing technologies in terms of layout, it can be foreseen that the 454 sequencing approach will represent a compelling tool for the construction of large comparative data sets allowing for more comprehensive studies on tunicate phylogenetics and molecular evolution in the near future.

Tunicates Evolve Twice as Fast as Vertebrates: But Not All Genes Nor All Lineages are Equal Determinants of This Contrast

To quantify the tunicate molecular divergence in a comparative chordate framework, we assessed evolutionary rate variation within Chordata on the basis of 35 housekeeping genes. Overall, our results suggested that tunicates have experienced a nearly twofold faster evolution compared to their vertebrate counterparts within Olfactores (Fig. 4), whereas the cephalochordate rate pattern appeared fairly similar to that of vertebrates (Fig. 5).

From a gene standpoint, shifts in tunicate rates have been detected in the majority of the examined genes, yet the degree of the predicted acceleration varied extensively across genes. Our estimates indicated a 1.4- to an uppermost 8-fold accelerated rate for tunicates versus vertebrates depending on the gene, clearly suggesting that not all the genes are equal determinants of rate discrepancies within Chordata (Fig. 3). For a substantial proportion of genes (15 out of 35 genes), the estimated evolutionary rate in tunicates was not significantly different from that of vertebrates (Table 1). This result implies that despite the generally striking genomic divergence (Dehal et al. 2002; Seo et al. 2001; Small et al. 2007b), at least some coding parts of tunicate genomes have escaped the prevalent acceleration bursts.

Considering lineage-specific rate profiles, Phlebobranchia and Stolidobranchia exhibited lower among-gene variation. The inclusion of Microcosmus allowed expanding the taxonomic coverage of Stolidobranchia, providing clues for a rather uniform rate at the ordinal taxonomic level. Finally, Aplousobranchia and, particularly, Appendicularia were characterized instead by the highest and the more heterogeneous rates across the examined markers.

In overview, tunicates presented a propensity for extensive rate variation across genes and lineages but certainly our results cannot be generalized to a genome scale, since 35 housekeeping genes constitute only a small portion of the chordate gene repertoire. Owing to the necessity of accurate orthology assessment among the compared sequences, studies in comparative genomics are frequently based on highly conserved molecular markers, being most of the time restricted to only few functional classes of genes. The main challenge for forthcoming studies would be to counterbalance the trade-off between fast-evolving sequences and orthology assessment, in order to explore tunicate, and thereby chordate, rate variation at a larger scale. This would allow testing the validity of two prefatory hypotheses issued from this work: (1) The high heterogeneity of across-gene rates of protein evolution and (2) the preponderant impact of Appendicularia and Aplousobranchia on the whole image of chordate evolutionary rate variation.

From another point of view, it would also be interesting to attempt lower taxonomic scale analyses, with the goal to decrease the evolutionary distance separating sampled species. The number of shared orthologous genes between two species is expected to be negatively correlated to their evolutionary distance, more closely relatives have thus a much broader potential in terms of number of molecular markers to be examined.

Insights into the Underlying Causes of Tunicate Accelerated Evolution

Evolutionary rate shifts in protein-coding genes among as well as within lineages could result from different factors, such as changes on selective constraints and positive selection, or changes in mutation rates, i.e. a general acceleration of nucleotide substitutions (Nabholz et al. 2008). In a preliminary effort to understand the causes that drove the observed tunicate rate acceleration, we compared synonymous (d S) and non-synonymous (d N) rates of Ciona intestinalis and Ciona savignyi versus those of teleosts and mammals.

The d N/d S ratio was roughly similar for the three lineages, albeit some pronounced differences were observed for some genes (Fig. 7). Tunicates were not systematically associated with higher d N/d S, and no signature of positive selection was traced in our data. On the contrary, all the examined genes evolved under strong purifying or negative selection, which in conjunction with the rather uniform picture of ω (Fig. 8) implies that tunicate rate shift with regards to vertebrates, and also chordates, most probably resulted from an increase of their mutation rate rather than changes in selection regime.

Yet, a generalization of these findings would be premature, obviously because of the limited gene sampling considered, but also because our estimations, of d S particularly, might suffer from the uncertainty due to the putatively confounding effects of saturation (Kryazhimskiy and Plotkin 2008) and from potential branch length effects (Wolf et al. 2009). Underestimated d S might indeed explain why we observe essentially similar d N/d S ratio in tunicates and vertebrates, although the former have presumably larger population sizes than the latter (Small et al. 2007a)—theory would predict more efficient purifying selection, and hence reduced d N/d S ratio, in large populations (Charlesworth 2009). Still, as a first insight into the causes of tunicate acceleration, our results should stimulate a more thorough research for the underlying determinant of mutation rate variation within tunicates. Variability in mutation rate has been so far attributed to a number of intrinsic factors, such as generation time and metabolic rate, mating system and, more recently, longevity (Nabholz et al. 2008). Undoubtedly, the highly diversified life-history traits of the group, encompassing, for example, both sexually and asexually reproducing species (Lambert 2005), provide a challenging ground for identifying links between changes in mutation rates and environmental, functional, and biological factors such as population size.

Conclusions

In this study, the comparative analysis of 35 highly conserved nuclear genes at both amino acid and nucleotide levels provides a first assessment of the within-chordate rate variation and bears a sphere of elements worthy of being tested in larger-scale analyses. More precisely, our results revealed (1) a two-fold faster evolution of tunicates as compared to other chordates; (2) a marked asymmetry and heterogeneity in tunicate evolutionary rates, both across genes and among major lineages; (3) the prevailing effects of Appendicularia and Aplousobranchia in the extent of rate contrasts within Chordata; and finally (4) the predominant role of changes in mutation rate as the most probable underlying cause of tunicate accelerated genomic evolution.