RESULTS AND DISCUSSION

Nucleotide sequence and gene organization: The complete mtDNA sequence of the B. carnosus mtDNA is 15,708 bp long and has an A + T content of 51.4%. As in most other metazoan mtDNAs, the genome contains the genes for 13 proteins (three subunits of cytochrome c oxidase, CO1, CO2, and CO3; one subunit of cytochrome c-ubiquinol oxidoreductase, CYTb; seven subunits of NADH dehydrogenase, ND1, ND2, ND3, ND4L, ND4, ND5, and ND6; and two subunits of ATP synthase, ATP6 and ATP8), 22 tRNAs, and 2 rRNAs (12S and 16S).

The genes are arranged in an extremely economical fashion such that most genes have no bases between their coding bodies, and 11 pairs of genes appear to overlap by 1-10 bases. Of these overlaps, 5 occur between genes encoded on different strands. The remaining 6 cases involve genes encoded on the same strand. In 3 instances, the transcript might be processed such that the upstream protein-coding gene (ATP6, ND5, and CYTb, respectively) gets a UAA stop codon by polyadenylation. A fourth case involves the ND4 gene and the downstream tRNA^His gene. Provided that the extra guanylate residue that tRNA^His carries at its 5′ end (Sprinzlet al. 1989) and that is added post-transcriptionally in other metazoans (L’Abbeet al. 1990) is encoded by the tRNA gene in the hemichordate, a UAA stop codon can be created on the ND4 transcript by polyadenylation. A fifth case involves the genes for ATP8 and ATP6, which overlap by 10 bases. A similar overlap between these genes is seen in most other metazoans, and it has been shown that the transcription of ATP6 is initiated within the ATP8-coding sequence in bovine mitochondria (Fearnley and Walker 1986). The sixth and final case involves the 3′ end of the gene for tRNA^Glu and the 5′ end of the gene for tRNA^Thr. Such overlaps involving mitochondrial tRNAs encoded on the same strand occur in molluscs, annelids, echinoderms, and vertebrates (Börneret al. 1997). It has been shown that they are processed such that the downstream tRNA is released in its intact form, whereas the upstream tRNA is truncated and gets filled in by an RNA-editing activity (Yokobori and Pääbo 1995a,b) that may involve polyadenylation (Yokobori and Pääbo 1997).

The organization of the genes and noncoding regions is presented in Figure 1 with the situations in a vertebrate, human (Andersonet al. 1981), and in an echinoderm, the sea urchin Arbacia lixula (De Giorgi et al. 1996). It can be observed that the gene order of the B. carnosus mtDNA is very similar to that in humans and, therefore, to other vertebrates, requiring a total of only six rearrangements to convert one gene order into the other. Only the gene encoding tRNA^Glu has changed from one strand to the other. In contrast, the order of genes in the echinoderm is remarkably different. One striking feature of the latter genome is the cluster of 13 tRNA genes that also occurs in the two other sea urchins (Jacobset al. 1988; Cantatoreet al. 1989) and in the starfish (Asakawaet al. 1995), although in the latter species, an inversion with respect to the sea urchins of a 4.6-kb region encompassing the tRNA gene cluster is seen.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

—Gene organization of (A) the B. carnosus (hemichordate) mitochondrial genome compared to that of (B) human (vertebrate) and (C) the sea urchin Arbacia lixula (echinoderm). Protein genes are designated using the abbreviations given in the text. Transfer RNA genes are indicated by the single-letter abbreviations. The isoacceptors of serine and leucine tRNAs are given with the corresponding anticodons in parentheses. Genes coded in the opposite strand are shown with an arrowhead. Noncoding regions are shown as black boxes. CR, control region; O_L (in the human mtDNA), the L-strand origin of replication; solid circles, putative signals of transcription outside the control region in the hemichordate and the echinoderm mitochondrial genomes. The minimum number of translocations necessary to interconvert the B. carnosus and the human mitochondrial genomes are indicated with double arrows. An inversion is indicated with an additional circular arrow. The differences with the echinoderm genome are too numerous to be depicted. The dashed line indicates that this translocation may have arisen by a duplication of the tRNA^Leu(TAA) gene.

Given the very different gene arrangements between vertebrates and echinoderms, it has been debated which of these might represent the ancestral gene order of these phyla (Cantatoreet al. 1987; Jacobset al. 1989a). Because the earliest group to branch off within vertebrates, the lampreys (Cyclostomata), have a gene order similar to that of other vertebrates (Lee and Kocher 1995), and partial sequences of a cephalochordate (Delarbreet al. 1997) appear to show the same arrangement as vertebrates, it is likely that the chordate gene order was established more than 400 mya. Because the hemichordate has virtually the same gene order as vertebrates, this pushes the date for the establishment of this organization of the mtDNA back to at least the split between vertebrates and hemichordates, more than 600 mya (Ayalaet al. 1998). Furthermore, because hemichordates are probably more related to echinoderms than to vertebrates (Turbevilleet al. 1994; Wada and Satoh 1994; Halanych 1995; see Figure 4), the vertebrate/hemichordate gene arrangement is likely to be the ancestral, and the echinoderm is likely to be derived.

The similarity between the hemichordate and vertebrate gene arrangements allows more detailed questions regarding translocations of specific genes to be addressed. Taking the high number of identities between the two tRNA^Leu genes in the echinoderm Paracentrotus lividus into account, it has been proposed that the tRNA^Leu(TAG) gene arose from a duplication of the tRNA^Leu(TAA) gene (Cantatoreet al. 1987). Echinoderms show a close similarity between both tRNA^Leu genes, ranging from 67 to 81% identity (compared to 38-48% identity between vertebrate tRNA^Leu genes, Sprinzlet al. 1989), pointing to a gene duplication that preceded the split of sea urchins and starfish. In the B. carnosus mitochondrial genome, both tRNA^Leu genes are also very similar (63% identity), indicating that the duplication took place before the hemichordate/echinoderm split. Furthermore, in the hemichordate mitochondrial genome, the tRNA^Leu(TAG) gene is located immediately upstream of the tRNA^Leu(TAA) gene. Because the latter gene is located upstream of the ND1 gene in echinoderms, vertebrates, and the hemichordate, it is probable that the hemichordate mtDNA represents the situation that existed in an ancestral species after the duplication and subsequent anticodon change in one of the tRNA^Leu(TAA) genes took place. We also note that the 5′ end of the hemichordate ND5 gene, which is adjacent to the tRNA^Leu(TAG) gene in vertebrates (Figure 1), is ∼80 bases longer than in vertebrates, having the same length as the homologous gene in echinoderms. This makes it plausible that this extension represents the remnants of a tRNA^Leu(TAG) gene that was appended to the ND5 gene, as suggested by Cantatore et al. (1987). Because the tRNA^Leu(TAG)-tRNA^Leu(TAA)-ND1 cluster also occurs in some molluscs, such as the blue mussel (Hoffmannet al. 1992) and the black chiton (Boore and Brown 1994), and in an arthropod, the horseshoe crab (Booreet al. 1995), it is conceivable that this arrangement is ancestral to protostomes and deuterostomes. However, the extension of the ND5 gene does not occur in protostomes, and the adjacent localization of the two tRNA^Leu genes is absent in other molluscs and arthropods. Therefore, an independent gene duplication and change of the tRNA^Leu(TAA) gene to a tRNA^Leu(TAG) gene, as well as the addition of the vestigial tRNA^Leu(TAG) gene to the ND5 gene, probably occurred in a common ancestor of echinoderms and hemichordates.

The comparison of the mitochondrial gene order among organisms has proved to be a valuable tool to establish phylogenetic relationships when shared derived characters can be used for such a purpose (Booreet al. 1995). However, the comparison of the gene order of different deuterostome phyla (Figure 1) shows that whereas vertebrates and hemichordates maintain many features of an ancestral gene order, echinoderms have experienced a large number of rearrangements, leaving no shared derived characters between phyla to establish a reliable phylogeny. Thus, the different rate of gene arrangements between groups makes it hard to identify shared derived characters and, thus, complicates the use of mtDNA gene order as a phylogenetic tool for metazoan phyla.

Protein-coding genes and genetic code: The B. carnosus mitochondrial genome encodes the same 13 proteins as other metazoans (Figure 1). While ATG serves as a start codon in most genes, GTG is used in CO2 and ATP8, ATA in ND6, and CTG in CO1. The CTG initiation codon has not been seen to date in other metazoan mtDNAs (Wolstenholme 1992), but it has been shown to initiate translation in mammalian nuclear genes (Hannet al. 1988; Peabody 1989). The occurrence of this unusual mitochondrial initiation codon in B. carnosus is likely because the inferred N-terminal part of the hemichordate CO1 protein is well conserved compared to that of other metazoans. TAA or TAG stop codons are used, except in CYTb and possibly a few other genes that overlap with downstream genes, where only one T or TA is found (see above). In these cases, a TAA stop codon is probably created post-transcriptionally through polyadenylation, as is the case in other organisms.

There are four mitochondrial codons whose meaning varies among the main deuterostome groups. Thus, ATA codes for Met in vertebrates and urochordates, and for Ile in echinoderms; AAA codes for Lys in vertebrates and urochordates, and for Asn in echinoderms; and AGA and AGG are stop codons in vertebrates, but they encode Ser in echinoderms and Gly in urochordates (Yokoboriet al. 1993). Table 1 shows the codons and their amino acid assignments in the 13 protein genes of the B. carnosus mtDNA. The amino acid assignments were achieved by multiple alignments of the B. carnosus protein-coding genes with proteins from 21 species representing five metazoan phyla. The codons ATA and AGA correspond to Ile and Ser, respectively, as is the case in echinoderms. Specifically, out of the 69 ATA hemichordate codons, 20 occur in positions where Ile is conserved in >55% of the sequences (eight of these positions have Ile in >90% of the sequences), while none of them occur at positions where Met is conserved. Of the 19 AGA codons, 6 occur at positions where Ser is conserved in >55% of the sequences (3 of them with Ser in >90% of the sequences), whereas they are never used as stop codons. The 2 other codons that differ between the deuterostome phyla, AAA and AGG, are absent from the B. carnosus mtDNA, while all other 62 codon triplets are used (Table 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

—Alignment of the segments with best similarity between the echinoderms and B. carnosus control regions. The B. carnosus sequence has been inverted with respect to the one represented in Figure 1. An asterisk indicates that the position is completely conserved in the species considered. The shaded blocks cover the best conserved motifs within this segment.

The overall sequences and predicted secondary structures of the B. carnosus tRNAs are similar to those of other deuterostomes (Sprinzlet al. 1989). However, one feature is of special interest. In the B. carnosus mtDNA, the tRNA^Lys gene carries the anticodon CTT, as is the case in echinoderms (Asakawaet al. 1995), whereas this tRNA gene carries the anticodon TTT in most other metazoans. This feature, which probably limits the recognition capacity of the tRNA^Lys to the codon AAG, together with the absence of AAA codons, could reflect an intermediate stage during the reassignment of the AAA codon from Lys to Asn in the echinoderm lineage (Castresanaet al. 1998), and it supports the codon capture hypothesis for codon reassignment (Osawa and Jukes 1989).

Noncoding regions and possible transcription and replication signals: Among vertebrates, the main sequences necessary for transcription and initiation of mtDNA replication are situated in the control region, between the tRNA^Pro and tRNA^Phe genes (Shadel and Clayton 1993, 1997). Two promoters exist in this region, one initiating the transcription of the H strand (the main coding strand) and another initiating the transcription of the opposite strand (L strand). DNA replication is linked to transcription because RNA transcripts initiated at the L-strand promoter serve as primers for replication of the H strand. Many of these initiation products become arrested in a way such that D-loop structures can be observed in many vertebrate mtDNA molecules. The L-strand origin of replication is situated outside the control region, normally between the tRNA^Asn and tRNA^Cys genes. In echinoderms, it has been shown in the sea urchin Strongylocentrotus purpuratus (Jacobs et al. 1988, 1989a,b; Elliott and Jacobs 1989) that the mtDNA may contain one major origin of replication where a D loop is formed, albeit smaller than in vertebrates, and that the initiation of DNA replication in this region is also linked to a transcriptional promoter. In addition, the echinoderm mtDNA contains several palindromic decanucleotide motifs scattered throughout the genome (filled circles in Figure 1) that have been proposed to serve as bidirectional transcription initiation signals (Elliott and Jacobs 1989; Jacobset al. 1989a), a situation reminiscent of the multiple promoters in the yeast mitochondrial genome (Christianson and Rabinowitz 1983).

To explore possible models of transcription and replication of the B. carnosus mtDNA, we looked for similarities by dot plots between its main noncoding region of 277 bp and other vertebrate and echinoderm control regions. The only segment exhibiting similarity to other taxa was a G tract followed by several AT dinucleotides that occur at a similar region in the three sea urchin sequences (Figure 2). No similarities were discerned in comparisons with the starfish or with vertebrates. This region of similarity has been previously observed by comparison of the control regions among the three sea urchins (De Giorgi et al. 1996). Interestingly, this is the region where the beginning of the formation of a D loop has been mapped in S. purpuratus (Jacobset al. 1989b). This might indicate that a similar structure could also be formed in other echinoderms and in the hemichordate.

No sequences with 9 or 10 matches to the decanucleotide sequence proposed to be involved in transcription in echinoderm mtDNA can be found in the B. carnosus genome. However, upon comparing the two small non-coding regions (of 32 and 41 bp, respectively), a decanucleotide motif T(T/C)ACCTTTTT was found at the beginning of both regions when the main coding strand was considered (Figure 3). This motif is found nowhere else in the B. carnosus mtDNA. Strikingly, both of these motifs are facing genes that are encoded on the minor coding strand. It is tempting to speculate that these noncoding regions contain initiation signals for two short transcripts that could cover all the genes located on the minor coding strand (Figure 1). If that were true, the hemichordate mtDNA would represent an intermediate stage between the vertebrate (with only two promoters, both located in the control region) and the echinoderm (with multiple promoters scattered around the genome) modes of transcription. Experimental data from these and other animals would be necessary to fully understand the function and evolution of the promoter sequences.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

—The two small noncoding regions of the B. carnosus mtDNA. The shaded blocks indicate the shared motif found in these regions. Arrows indicate the starting point and direction of transcription of the flanking genes. (A) The region between the tRNA^Ser(TGA) and tRNA^Asp genes. (B) The region between the ND6 and tRNA^Phe genes.

Phylogenetic trees: Hemichordates possess adult characters, such as gill slits or a dorsal hollow nerve cord, that link them to the chordates, while many embryological features place them closer to echinoderms (Hyman 1959; Schaefer 1987). As a consequence, their phylogenetic position within deuterostomes has been widely debated (Willmer 1990; Nielsen 1995; Gee 1996). Complete nuclear 18S RNA sequences from the enteropneusts B. carnosus (Wada and Satoh 1994) and Saccoglossus kowalevskii (Turbevilleet al. 1994), as well as partial sequences of S. cambrensis (Hollandet al. 1991) and the pterobranch Rhabdopleura normani (Halanych 1995), have been recently used to try to assess the phylogenetic affinities of hemichordates. With the exception of the analysis of the S. cambrensis partial sequence, these studies have found the hemichordates to be the sister taxon of echinoderms rather than vertebrates, but without sufficient support to arrive at a firm conclusion.

Alignments of the inferred amino acid sequences of 12 B. carnosus mtDNA proteins with the homologous proteins in echinoderms, chordates, and a mollusc used as an outgroup were constructed. Regions of problematic alignment were removed, allowing for less than eight (or four) contiguous, nonconserved positions in the selected segments (see materials and methods), and the resulting alignments were concatenated. Furthermore, because the frequency of the amino acids Lys and Met has changed in echinoderms and hemichordates (Castresanaet al. 1998), all positions containing Lys or Met were eliminated from each alignment. Thus, a total of four alignments were produced (Table 2).

Neighbor-joining trees derived from all these alignments show the sister group status of hemichordates and echinoderms to be supported by >95% of bootstrap replications (Figure 4, A-D). The neighbor-joining trees were used as starting points to search, by means of a local rearrangement algorithm, for better tree topologies according to the maximum-likelihood criterion using the mtREV model of amino acid substitution based on mitochondrial proteins (Adachi and Hasegawa 1996a). The best topologies found for every alignment are shown in Figure 4 (E-H). They all show the hemichordate B. carnosus to group with echinoderms. To further investigate the phylogenetic position of hemichordates, the preferred maximum-likelihood topology found for every alignment was compared with two alternative topologies for the interrelationship of the three deuterostome groups. Table 2 shows the differences in log likelihood relative to the preferred topology and the relevant bootstrap proportions supporting each topology. Only two of the alignments allow one of the alternative topologies to be rejected and support the preferred topology by >90% of bootstrap value. Therefore, although the sequence analyses favor the sister group status of hemichordates and echinoderms, alternative phylogenetic positions of the hemichordate cannot be statistically rejected using these alignments.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

—Phylogenetic trees relating several deuterostomes as estimated from alignments of inferred mtDNA proteins. (A-D) Trees estimated by the neighbor-joining algorithm using Dayhoff distances. Numbers represent local bootstrap values from 100 replicates. (E-H) Trees estimated by maximum likelihood with the mtREV model of amino acid substitution. Horizontal branch lengths are proportional to evolutionary distances. The scale bar represents a distance of 0.1 substitutions per site. When the JTT and Dayhoff models of amino acid substitution were used, the preferred topologies were almost identical to those presented here, with the major exception being that the cephalochordate B. floridae was placed with the outgroup in one of the trees using the JTT model. Trees A and E, B and F, C and G, and D and H were calculated from four different alignments, respectively, as explained in the text and in Table 2. Sequences used for constructing the trees (with GenBank accession numbers of the mtDNA sequences in parentheses) are from the following species: the mollusc Katharina tunicata (U09810); the cephalochordate B. floridae (AF035164-AF035176); the cyclostomes P. marinus (U-11880) and M. glutinosa (Y15180-Y15191); the ray-finned fishes Cyprinus carpio (X61010) and Polypterus ornatipinnis (U62532); the lobe-finned fishes Protopterus dolloi (L42813) and Latimeria chalumnae (U82228); the amphibian Xenopus laevis (M10217); the bird Gallus gallus (X52392); the mammals Didelphis virginiana (Z29573), Mus musculus (J01420), and Bos taurus (J01394); the sea urchins P. lividus (J04815), S. purpuratus (X12631), and A. lixula (X80396); the starfish Asterina pectinifera (D16387); the crinoid Florometra serratissima (AF049132); and the hemichordate B. carnosus (AF051097).

• Download figure
• Open in new tab
• Download powerpoint

With respect to other taxa, the neighbor-joining trees (Figure 4, A-D) deviate from the expected morphological trees (e.g., Young 1981) in three main aspects: tetrapods are not monophyletic, cyclostomes (the lamprey, Petromyzon marinus, and the hagfish, Myxine glutinosa) are always paraphyletic, and the cephalochordate amphioxus (Branchiostoma floridae) is not at the base of vertebrates in one of the trees. Furthermore, the relation of tetrapods to fishes is very weakly supported. In most of these aspects, the maximum-likelihood trees are closer to the expected morphological trees. For example, they all have the cephalochordates at the base of vertebrates and cyclostomes as a monophyletic group. However, the tetrapods are monophyletic in only one maximum-likelihood tree, and the relation of tetrapods to different fish groups is unstable.

Previous studies based on mitochondrial protein-coding genes failed to recover the expected position of cephalochordates at the base of the vertebrates (Naylor and Brown 1997, 1998). However, the results presented here may indicate that the use of inferred amino acid sequences and a mitochondrial model of amino acid substitution (Adachi and Hasegawa 1996a) may allow a better estimation of the phylogenetic relationships of these divergent taxa. In another analysis of inferred mitochondrial proteins (Rasmussenet al. 1998), cyclostomes were shown to be paraphyletic, in contrast with the prevalent morphological view. The cyclostome paraphyly is seen here in the neighbor-joining trees (Figure 4, A-D), but not in the maximum-likelihood trees (Figure 4, E-H). Further work is obviously necessary to resolve these issues. This is particularly clear in view of the fact that the generally accepted tetrapod monophyly fails to be recovered in most of the trees, something that has also been observed when vertebrate mitochondrial proteins (Nei 1996) and 18S RNA sequences of several animals were considered (Philippeet al. 1994). The reason for this is unclear, but it may involve changes in the patterns and rates of substitution in different lineages (Philippeet al. 1994; Nei 1996).

In addition to the phylogenetic analyses of the inferred protein sequences, other features of the B. carnosus mtDNA strongly support its phylogenetic association with echinoderms. For example, the hemichordate shares aspects of its genetic code, sequence motifs in the control region, the close similarity of the two tRNA^Leu genes, and the presence of a N-terminal extension of ND5 with echinoderms. In contrast, the overall order of the mtDNA genes in hemichordates is quite similar to that of vertebrates, while it shows substantial differences to that of echinoderms. Thus, while a large number of gene rearrangements has occurred in echinoderms, both hemichordates and vertebrates have maintained a mitochondrial gene order similar to the one that existed in their common ancestor >600 mya.