PCR experiments were done with a closely related Hydra species. We obtained specimens of Hydra sp. from the Schulbiologie-Zentrum Hannover. DNA from one polyp was prepared with the Chelex method (protocol as described in 22), 1 μl of the undiluted supernatant or 1 μl of a 1:10 dilution was used as template. Phylogenetic analysis with partial cox1 data verified that our Hydra sp. specimen is very closely related to Hydra magnipapillata (see Results).

Primers (Additional file 3) were designed to confirm our bioinformatically derived observations via PCR. The fragments shown in Fig. 1 were amplified and sequenced (some in two overlapping parts, see Additional file 3 for details). Sequences have been submitted to GenBank [GenBank: EU683621–EU683624].

To exclude the possibility that we had amplified a nuclear mt pseudogene (NUMT) of the nd5 and partial cox1 fragment, we started different assemblies with the pipeline originating from blast hits of a 200 bp fragment (100 bp both down- and up-stream of the connection of nd5 and cox1 in the assembly), as well as two assemblies starting from the last 100 bp 3' of nd5, and the first 100 bp of the incomplete copy of cox1. In the first two cases, only one assembly was received, each time consistent with our former assemblies. By starting with 100 bp of the partial cox1, we recovered different assemblies, which were compatible with one chromosomal assembly or the other. In no case did we observe any inconsistencies or any presence of nuclear genes, which would have indicated that the nd5-cox1 arrangement is part of a NUMT.

The two chromosomes of the H. magnipapillata mt genome each carry one rRNA gene (mt1:rnl; mt2: rns). Each of the assembled contigs of mt1 and mt 2 is represented by > 7,000 single sequence reads from the trace archive (Additional file 1). Our consensus sequence for mt1 is 8,194 bp long. This matches the length reported by Bridge et al. 14 for this H. magnipapillata mt chromosome. The sequence of mt2 is shorter (7,686 bp).

The H. magnipapillata mt genome includes 13 protein-coding genes of the respiratory chain usually found in other Metazoa. mt1 contains 6 protein-coding genes, rnl and two tRNA genes; mt2 contains 7 protein-coding genes, rns and one tRNA gene (Fig. 1A). All genes are unidirectionally encoded on each of the two molecules and densely arranged along the chromosomes. As in H. oligactis, the longest non-coding intergenic region is 52 bp between cox3 and nd2 17. Otherwise, subsequent genes are separated by 0–5 bp or overlap for up to 10 bp (in nd6-nd3 and nd1-nd4).

Like many other Cnidaria 1617323334, the H. magnipapillata mt genome possesses only the two tRNA genes for methionine (trnM; CAU) and tryptophan (trnW; UCA). trnW is only found on mt1, whereas identical copies of trnM are present on both chromosomes (Fig. 1B).

Six amino acid codons are not used in the 13 protein-coding genes (Table 1), and all genes are terminated by TAA. Apparently synonymous codons that posses an A or T, instead of a G or C, at the third codon position are preferred in H. magnipapillata. To test whether this observation is caused by mechanisms that affect base composition in the whole mt genome, we analyzed codon usage in the 13 respiratory protein-coding genes in 24 mt genomes of Cnidaria. We plotted the AT content at each of the three codon positions against the AT contents of the rRNA genes for every genome, as rRNA coding genes represent a different part of the mt genomes in terms of functional constraints compared to protein-coding genes. Remarkably, H. magnipapillata showed the highest values for AT content at the third codon positions (89.8%) and in the rRNA genes (78.1%; Fig. 2, black filled symbols). Moreover, a high AT content in rRNA genes generally correlates with the usage of A and T at third codon positions in all Cnidaria (significant at p = 0.001), suggesting that codon usage might be the result of a general selection for base composition on the mt genome caused by interaction of mutational, repair, replication and translational mechanisms 35. The AT content at the first and second codon positions also correlates with that of the rRNA genes (significant at p = 0.001), but here AT content rise at a lower rate than the increasing AT content of the rRNAs (regression line slopes: first codon position: 0.46; second: 0.33; third: 1.18). This is likely the result of selection on certain amino acids. Cnidarians posses a lower AT content at the first codon position than at the second (Fig. 2), with H. magnipapillata and H. oligactis being the only exceptions (73.1% vs. 70.9% for H. magnipapillata, filled symbols in Fig. 2).

Compared to the gene arrangement of A. aurita and H. oligactis, only a few changes can be observed in H. magnipapillata. Neglecting the positions of tRNAs, two blocks (cox2, atp8, atp6, cox3, nd2, nd5 and rns; nd6, nd3, nd4L, nd1, nd4, cob) of genes are identical across the three genomes, occurring on mt1 or mt2, respectively, in H. magnipapillata (Fig. 1A). The mt genomes of H. oligactis and of H. magnipapillata are entirely alignable and display a sequence divergence of 12.3% (excluding the terminal chromosome structures; see below).

As mentioned before, we found 191–196 bp of ITR at both ends of mt1 and mt2. In the linear mt genomes of H. oligactis and A. aurita, ITR were also present but were longer (H. oligactis: 1,488 bp; A. aurita: 471 bp, 1617 assuming symmetry for unsequenced ends). Unlike ITR in Aurelia 16, ITR in H. magnipapillata have a higher GC content than the rest of the molecule (52.2% GC in ITR vs. 25.2% GC in 5' IOS [see below], 27.6% GC in 3' IOS [see below] and a mean of 22.5% GC for all remaining regions). We found that a smaller part of 3' cox1 (54 bp) is included in all ITR of H. magnipapillata. Probably because the 3' end of cox1 is not very conserved, Pont-Kingdon et al. 18 missed this feature in their mt1 fragment of H. vulgaris. The ITR regions of H. oligactis contain a larger cox1 fragment (one non-functional copy at the 5' end, functional cox1 at 3' end, Fig. 1A). The remaining sequenced 3' region of ITR in H. oligactis is very similar to those found in H. magnipapillata and H. vulgaris (Fig. 1C), but longer. Between H. magnipapillata and H. vulgaris, the major difference is that a stretch of Gs (31 in H. vulgaris) is significantly shorter in H. magnipapillata (11–16 at the homologous region).

In H. magnipapillata mt1 and mt2, we found additional identical sequences at the 5' and 3' ends following (at the 5' ends) and preceding (at the 3' ends) the ITR. We refer to those regions as identically oriented sequences (5' and 3' IOS, Fig. 1B). After the ITR, the 5' IOS of both molecules contain identical copies of non-coding DNA and trnM. At the 3' IOS we found a larger partial copy of the 5' region of cox1 on mt1. As a consequence of this arrangement, mt1 and mt2 share 310 bp (ITR+5' IOS) at the 5' end and 436 bp (3' IOS+ITR) at the 3' end, giving both molecules a specific orientation.

Using PCR experiments with the closely related Hydra sp., we verified the following arrangements initially observed in the H. magnipapillata sequences (compare Fig. 1A): (i) the presence and orientation of the ITR at all four chromosome ends could be shown, as well as the presence of partial cox1 sequences in the ITR; (ii) identical regions are shared at the 5' and 3' end, respectively, between mt1 and mt2 adjacent to the ITR; and (iii) within the latter regions, the 5' motif contains trnM, which therefore appears in two copies in the genome, and a larger sequence of cox1 forms the shared 3' motif of mt1 and mt2.

The tree topology derived from our phylogenetic analysis of cox1 shows the close relationship of Hydra sp. and H. magnipapillata (Fig. 3), thus ensuring that we used an appropriate taxon to test our results. H. vulgaris (Two sequences from GenBank) is paraphyletic, which reflects the difficult taxonomy of the genus 36. The presented phylogeny, in combination with the mt genome organization, supports the view that the ancestral state of mt genome organization in the genus Hydra was a single linear mt chromosome.

Linearity of mt genomes seems to have evolved once after the divergence of Medusozoa from Anthozoa. Fig. 3A summarizes the results of different studies 13151617, mapped on a cnidarian phylogeny 37. A fragmentation of mt genomes has been reported from several Hydra species (Hydrozoa) 1314 and Cubozoa 15. Uncertainties remain for Cubozoa: Bridge et al. 14 studied the same cubozoan species Carybdea marsupialis as Ender and Schierwater 15, but reported a single ~16 kb linear mt genome, while in the more recent work, a ~4 kb fragment was shown to carry the rnl gene. Because Ender and Schierwater 15 were able to repeat the experiments with different DNA isolates of C. marsupialis and obtained concordant results from an additional cubozoan species (Tripedalia cystophora), an experimental error seems unlikely. However, their conclusion of four equally-sized mt chromosomes in Cubozoa is not directly supported by their identification of a 4 kb chromosome carrying rnl. Alternatively, one could assume the presence of a single ~12 kb mt counterpart, as indicated in Fig. 3A. Such an arrangement is possible, e.g., if rnl and cox1, the two genes that are encoded in different orientation to the other mt genes in A. aurita 16, were encoded in one chromosome in Cubozoa, and the remaining genes on a second chromosome.

However, given the available data it seems reasonable to assume that fragmented linear genomes occur in both Cubozoa and Hydrozoa (in some members of the genus Hydra). This suggests from an evolutionary perspective that the mt genome in the common ancestor of Medusozoa was linear and then independently split into different chromosomes in Hydra (Fig. 3B), and in at least some Cubozoa (compare Fig. 3A).

A possible mechanism for the origin of linear chromosomes from a circular molecule is the integration of one or more resolution elements 4. The circular DNA molecule would be split into one or more linear molecules with identical ends. In Medusozoa, the processes of linearization and the split of one linear into two linear chromosomes were obviously different processes as shown in the phylogenetic trees (Fig. 3). The linearization, possibly occurring in the last common ancestor of medusozoans, seems to have preceded the splitting of the chromosomes by a long time. If the ancestral linear mt chromosome of Medusozoa originated by introduction of a resolution element, one probably would not expect to observe its original motifs, which would occur as identical repeats at the two ends of the linear molecule 4. Indeed, the ends of linear medusozoan mt chromosomes have inverted terminal motifs (the ITR), instead of direct repeats. The splitting of ancestral linear mt chromosomes as in H. magnipapillata (and possibly Cubozoa) happened much later in evolutionary history, contradicting the view that the two or more linear mt chromosomes in Medusozoa directly originated from one circular DNA molecule.

Fragmented mt genomes are present in various eukaryotic taxa, e.g., in dinoflagellates 3839, Ichthyosporea 12 and Fungi 40. In Metazoa, fragmented mt genomes are known from the genera Globodera (Nematoda 414243), Dicyema (Mesozoa 44) and the rotifer Brachionus plicatilis 45, but unlike in H. magnipapillata, in these taxa the genomes are encoded on several small circular molecules. The mt chromosomal organization observed in H. magnipapillata supports the hypothesis of an ancestral, linear chromosome in Hydra (Fig. 3B), as represented by the mt genome of H. oligactis 17, which has been split in two between nd5 and rns.

Warrior 13 already suggested the presence of identical terminal sequences on both chromosomes of H. vulgaris. We now show that these ends are arranged as ITR on mt1 and mt2, as in other medusozoans 1617. In H. oligactis, which in the phylogenetic tree branches off before Hydra species carrying two mt DNA molecules (Fig. 3B), the single linear mt chromosome has ITR containing a large copy of the 5' end of cox1. Only the ITR at the 3' end has been completely sequenced 17. Based on our findings in H. magnipapillata, we predict that the unsequenced 5' end is almost identical to the 3' motif (Fig. 1), and we expect that about 150 bp remain unsequenced on the 5' end (in contrast to the 65 bp that have been proposed 17). In Hydra, partial copies of cox1 play a crucial role as part in ITR regions at the chromosome ends (Fig. 1, 17). The ITR in H. magnipapillata contains only a short sequence of the 3' end of cox1 (54 bp, compared to the 1,284 bp in H. oligactis), suggesting that large parts of the cox1 copies were lost. A simultaneous duplication of 5' ITR (containing the already shortened partial cox1 copy) and the 5' IOS motif seems likely to have occurred in the process of chromosome splitting. In this case, the longer cox1 copy (containing additional 240 bp of cox1) is a duplication of the functional cox1 of the original 5' end of a single mt chromosome (Fig. 1A).

ITR of linear mt molecules are present in other taxa besides medusozoans, e.g., in yeasts (e.g., 9) and in the green algae Chlamydomonas reinhardtii 10. Furthermore, in the green algae Polytomella parva, identical ITR are present at all ends of the two linear mt chromosomes 11, similar to what we observe in H. magnipapillata. We report 5' and 3' IOS as an additional shared feature of the two mt chromosomes. Interestingly, such an arrangement of ITR and 5' and 3' IOS is also seen in another, highly fragmented eukaryotic mt genome. In the ichthyosporean Amoebidium parasiticum, mt genes are distributed over several hundred different chromosomes, each of which also possesses ITR and 5' and 3' IOS 12.

Pont-Kingdon et al. 18 speculated that there may be a role for transcription initiation at the 240 bp 5' of trnM, which they found in their H. vulgaris (as H. attenuata) partial mt1 sequence. Considering that transcription initiation within the ITR would result in energetically expensive nonsense transcripts (since all genes are encoded on only one strand), transcription is more likely to start in the adjacent, non-coding regions of the 5' IOS. In H. magnipapillata and H. vulgaris this region within the 5' IOS is 40 bp long and lies between the cox1 copy and trnM (Fig. 1B). In H. oligactis, the non-coding region between the ITR and trnM is only 6 bp. However, a striking sequence similarity can be observed near trnM between H. oligactis and H. vulgaris (with the same sequence in this region as H. magnipapillata) 17. There is a 14-bp motif (TTATTTRRTCTTCT) that is shared between the species and differs by the last 3 bp from the 3' ITR+3 bp counterpart in H. oligactis. This motif might be involved in transcription initiation. If so, the difference in the very last 3 bp between the 5' end and its counterpart on the reverse strand in the ITR of H. oligactis prevents a functional transcription signal on the non-coding strand in this species. A crucial function for transcription initiation would explain selective pressure for maintaining the 5' IOS of both molecules after the ITR in H. magnipapillata. All mt chromosomes from Amoebidium parasiticum that contain coding genes are transcribed from 5' IOS to 3' IOS, as in H. magnipapillata 12. This observation led Burger et al. 12 to the conclusion that the IOS in Amoebidium are responsible for transcription initiation (5' IOS) and termination (3' IOS). While in H. magnipapillata we expect the same function for 5' IOS, the role of the additional partial cox1 copy within the 3' IOS of mt1 and mt2, if any, remains unknown; considering that the end of cox1 is part of the ITR, transcription can only be terminated in ITR and not in the 3' IOS. The sequence homologies of ITR and IOS within or between mt1 and mt2 are probably not the result of a relatively recent origin from ancestral sequences, as a first duplication of partial cox1 is already observed in H. oligactis and therefore predates the separation process. The substitutions between ITR (partial cox1) of the two species are found in all ITR copies in each mt genome. Considering this and the fact that similar arrangements are found in other eukaryotes 8910111216, it seems more likely that concerted evolution maintains the almost identical sequences, probably caused or influenced by the yet unknown mt genome replication mechanism. Terminal sequences of linear DNA molecules play a crucial role in mt replication 4. The main problem in the process of linear chromosome replication is the maintenance of the 5' ends. In nuclear (nc) chromosomes this is normally achieved by telomerase, an enzyme that adds short sequence motifs in tandem repeats at the end of each molecule to compensate for loss at the 5' end that occurs in each replication cycle 46. Consequently, in Hydra as in most Metazoa, the motif (TTAGGG)_n is found at the end of nc chromosomes 47. The termini of Hydra mt chromosomes are much more complex, and their maintenance during replication is not yet understood but is most likely telomerase independent 13. Warrior 13 suggested an mt replication mechanism for the two H. vulgaris mt chromosomes similar to that of T7 bacteriophages. His conclusion was based upon observations from hybridization experiments, which showed the presence of identical terminal sequences that he assumed to have the same orientation at the 5' and 3' ends. According to this model, intermediate concatamers are formed, and via ligation, site-specific nicking and elongation, the 5' ends are finally filled. Based on our data we can reject this model, because we showed that the terminal sequences in H. magnipapillata are inverted (ITR) and therefore do not allow the necessary concatamerization in the proposed way. Similarly, ephemeral circularization as in the phage lambda is not possible when terminal sequences are not direct repeats, but inverted.

Different replication mechanisms for linear chromosomes have been reported or proposed (for review, see 48). Solutions for maintaining the terminal structure in sequences with ITR include (a) covalently bound proteins, which also could serve as primers for a 'racket frame' replication (e.g., in linear mt chromosomes of plants and fungi or adenoviruses 4849), (b) 5' and 3' ends of chromosomes that are connected by a hairpin-loop (e.g., in some yeasts 948) and (c) single-stranded 3' overhangs (e.g., in Chlamydomonas reinhardtii 10). Warrior 13 showed that the first two possibilities are not realized in H. vulgaris. The mt-replication mechanism in Chlamydomonas requires an internal repeat of the single-stranded 3' overhangs 10. We cannot entirely rule out the existence of short single-stranded 3' overhangs, as it is possible they might have been missed due to our methods. The outermost sequence at least cannot be part of a repeat motif, as our PCR amplification of Hydra sp. did not yield fragments of different sizes. Furthermore, neither in mt1 of H. vulgaris nor in the mt genome of H. oligactis were additional sequences or repeats found 1317. Therefore, although a similarity to the mt replication of C. reinhardtii cannot be excluded, we find that mechanisms for linear mt chromosome replication are too diverse and that too many details are still unknown 448 for us to draw further conclusions about the replication process in H. magnipapillata from the presence of the ITR alone. Considering that ITR are a shared feature among all three available sequences of medusozoan mt genomes (H. oligactis 1617, A. aurita 1617 and H. magnipapillata), it is very likely that the mechanisms for mt replication are similar in all medusozoans and are still the same after the fragmentation of the mt genome. Keeping in mind that similar arrangements of ITR and IOS are found in Amoebidium parasiticum, the mt replication mechanisms of H. magnipapillata and Medusozoa are probably not unique among eukaryotes.